id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0002/quant-ph0002090.html
|
ar5iv
|
text
|
# Acknowledgement
## Acknowledgement
The authors thank Prof Brian Wybourne for independently repeating our calculation and correcting an error. Prof Ron King (private communication) has confirmed that the degree 12 coefficient in $`F(q)`$ is 964 in agreement with . We further thank Dr Markus Grassl for correspondence about our results.
|
warning/0002/gr-qc0002029.html
|
ar5iv
|
text
|
# 1 Values of ϵ_𝑛 as computed in the WKB approximation on the basis of (19) and within perturbation theory (PT) for various values of the potential 𝑉 (here, 𝑈_{𝑚𝑎𝑥}=1/4𝑉 is the barrier height; Δ𝑎=𝑎₂-𝑎₁ 1s the barrier width; and Σ is the number of levels at given 𝑉), along with the decay-probability values Γ_𝑛 as given by (18) under the same conditions
Evolution of the Quantum Friedmann Universe Featuring Radiation <sup>a</sup><sup>a</sup>a Published in Physics of Atomic Nuclei, Vol. 62, No. 4, 1999, pp. 708-714. Translated from Yadernaya Fizika, Vol. 62, No.4, 1999, pp. 758-764.
V. V. Kuzmichev <sup>b</sup><sup>b</sup>b e-mail: vvkuzmichev@yahoo.com; specrada@bitp.kiev.ua
Bogolyubov Institute for Theoretical Physics, National Academy of Sciences of Ukraine, Metrolohichna St. 14b, Kiev, 03143 Ukraine
Abstract: The classical and quantum models of the Friedmann universe originally filled with a scalar field and radiation have been studied. The radiation has been used to specify a reference frame that makes it possible to remove ambiguities in choosing the time coordinate. Solutions to the Einstein and Schrödinger equations have been studied under the assumption that the rate of scalar-field variation is much less than the rate of universe expansion (contraction). It has been shown that, under certain conditions, the quantum universe can be in quasistationary states. The probability that the universe goes over to states with large quantum numbers owing to the interaction of the scalar and gravitational fields is nonzero. It has been shown that, in the lowest state, the scale factor is on order of the Planck length. The matter- and radiation-energy densities in the Planck era have been computed. The possible scenarios of Universe evolution are discussed.
1. INTRODUCTION
That quantum gravity theory cannot rely on experimental data adds importance to exactly soluble cosmological models. However, the application of basic ideas underlying quantum theory to a system of gravitational and matter fields runs into difficulties of a fundamental character, which do not depend on the choice of a specific model. By way of example, we will consider a homogeneous, isotropic, and closed universe characterized by the Friedmann-Robertson-Walker metric; that is,
$`ds^2=a^2(\eta )[N^2(\eta )d\eta ^2d\mathrm{\Omega }^2],`$ (1)
Here, $`N(\eta )`$ is a function that specifies the time-reference scale; $`a(\eta )`$ is a scale factor; $`d\mathrm{\Omega }^2`$ is an interval element on a unit 3-sphere; and $`\eta `$ is the parameter that is related to the synchronous proper time $`t`$ by the differential equation $`dt=Nad\eta `$. Considering that scalar fields play a fundamental role both in quantum field theory and in the cosmology of the early Universe , we assume that, originally, the Universe was filled with matter in the form of a uniform scalar field $`\varphi `$. If the field $`\varphi `$ varies slowly in the early Universe, its potential $`V(\varphi )`$ specifies the vacuum-energy density (cosmological term) and ensures Hubble expansion. Restricting our analysis to the case of minimal coupling between geometry and the scalar field, we represent the action functional in the conventional form
$`S={\displaystyle 𝑑\eta \left[\pi _aa^{}+\pi _\varphi \varphi ^{}H\right]},`$ (2)
where a prime denotes differentiation with respect to $`d/d\eta `$; $`\pi _a`$ and $`\pi _\varphi `$ are the momenta canonically conjugate with the variables $`a`$ and $`\varphi `$, respectively; and $`H`$ is the Hamiltonian given by
$$H=\frac{1}{2}N\left[\pi _a^2+\frac{2}{a^2}\pi _\varphi ^2a^2+a^4V(\varphi )\right]N.$$
(3)
Here, the variables $`a`$ and $`\varphi `$ are taken, respectively, in units of the length $`l=\sqrt{2G/3\pi }`$ and in units of $`\stackrel{~}{\varphi }=\sqrt{3/8\pi G}`$. The function $`N`$ plays the role of a Lagrange multiplier, and the variation $`\delta S/\delta N`$ leads to the constraint equation $`=0`$. The structure of the constraint is such that true dynamical degrees of freedom cannot be singed out explicitly. This creates problems in the interpretation of quantum geometrodynamics \[4 6\]. It is commonly thought that the main reason behind such difficulties is that there is no natural way to define a spacetime event in general covariant theories . In the model being considered, the above difficulties are reflected in that the choice of the time variable is ambiguous.
For the choice of the time coordinate to be unambiguous, the model must be supplemented with a coordinate condition. When the coordinate condition is added to the field equations, their solution can be found for a fixed time variable. However, this method of removing ambiguities in specifying the time variable does not solve the problem of a quantum description, because undesirable consequences of this ambiguity in eventual equations cannot be avoided in this way.
In this study, we propose specifying a reference frame with the aid of an additional matter source. This method does not come into conflict with the adopted ideas of the early Universe . At the same time, it enables us to study the evolution of the Universe not only in the semiclassical approximation but also at a purely quantum level.
2. CLASSICAL DESCRIPTION
2.1. Fundamentals of the Model
The ambiguity associated with choosing the time coordinate in (1) will be removed with the aid of a coordinate condition imposed prior to varying the action functional, but its coordinate invariance will be restored . We will choose the coordinate condition in the form $`T^{}=N`$, where $`T`$ is the privileged time coordinate, and include it in the action functional with the aid of a Lagrange multiplier $`P`$; that is,
$$S=𝑑\eta \left[\pi _aa^{}+\pi _\varphi \varphi ^{}+PT^{}\right],$$
(4)
where
$$=N[P+]$$
(5)
is the new Hamiltonian. The constraint equation reduces to the form
$$P+=0.$$
(6)
Integrating the canonical equation $`P^{}=[P,]=0`$, we immediately obtain $`P=E`$, where $`E`$ is a constant. The full set of equations for the model in question becomes
$$\dot{a}^2\frac{a^2}{2}\dot{\varphi }^2+U=E,$$
(7)
$$\ddot{\varphi }+2\frac{\dot{a}}{a}\dot{\varphi }+a^2\frac{dV}{d\varphi }=0,$$
(8)
where overdots denote differentiation with respect to$`T`$ and $`Ua^2^4V(\varphi )`$. Equation (7) represents the Einstein equation for the $`({}_{0}{}^{0})`$ component, while equation (8) is the equation of motion for the field $`\varphi `$. A modification to the Einstein equations that is associated with including the coordinate condition in the action functional is that, on the right-hand side, there additionally arises an energy-momentum tensor $`\stackrel{~}{T}_0^0=E/a^4,\stackrel{~}{T}_1^1=\stackrel{~}{T}_2^2=\stackrel{~}{T}_3^3=E/3a^4,\stackrel{~}{T}_\beta ^\alpha =0\text{for}\alpha \beta `$ that can be interpreted as the energy-momentum tensor of radiation . The choice of radiation as the matter reference frame is natural for the case in which relativistic matter (electromagnetic radiation, neutrino radiation, etc.) is dominant at the early stage of Universe evolution. If our Universe were described by the model specified by equation (4), it would be possible to relate the above radiation at the present era to cosmic microwave background radiation.
2.2. Solving the Einstein Equations
A feature peculiar to the model in question is that it involves a barrier in the variable $`a`$. This barrier, described by the function $`U`$, is formed by the interaction of the scalar and gravitational fields. It exists for any form of the scalar-field potential $`V(\varphi )`$ and becomes impenetrable on the side of small $`a`$ in the limit $`V0`$. In just the same way as in inflation models (see ), we assume that the rate at which the scalar field $`\varphi `$ changes is much smaller than the rate of universe evolution, $`|\dot{a}/a||\dot{\varphi }|`$. In this case, equation (7) can be integrated in a general form. Presented immediately below are explicit solutions in the regions $`aa_1`$ and $`aa_2`$, where they are assumed to satisfy the boundary conditions $`a(0)=0`$ and $`a(t_{in})=a_2`$, respectively; here, $`a_1`$ and $`a_2`$ are the turning points ($`a_1<a_2`$) specified by the condition $`U=ϵ`$, and $`t_{in}`$ is the initial instant of time in the second region. We have
$$a(t)=\left[\frac{1}{2V}\left(1\text{cosh}\mathrm{\hspace{0.17em}2}\sqrt{V}t\right)+\sqrt{\frac{ϵ}{V}}\text{sinh}\mathrm{\hspace{0.17em}2}\sqrt{V}t\right]^{1/2}$$
(9)
for $`aa_1`$ and
$$a(t)=\left\{\frac{1}{2V}\left[1+\sqrt{14Vϵ}\text{cosh}\mathrm{\hspace{0.17em}2}\sqrt{V}(tt_{in})\right]\right\}^{1/2}$$
(10)
for $`aa_2`$. The quantities $`ϵ`$ and $`U`$ depend parametrically on $`\varphi `$. In the zero-order approximation, the former is given by $`ϵ=E`$. The above solution to equation (7) can be refined by taking into account a slow variation of the field $`\varphi `$ with the aid of the equation
$$\frac{a^2}{2}\dot{\varphi }^2+ϵ=E,$$
(11)
where $`ϵ(\varphi )`$ stands for a potential term, which is bounded by the inequality $`Eϵ1/4V`$. The case of $`ϵ>1/4V`$, which corresponds to an infinite motion, will not be considered in this study. From equations (8) and (11), it follows that, in general, a change in the potential $`V(\varphi )`$ entails a change in the quantity $`ϵ(\varphi )`$.
The solution in (9) describes the universe expanding from the point of the initial cosmological singularity to the maximum possible value of $`a_1`$ achieved at the instant $`t_m=\frac{1}{4\sqrt{V}}\mathrm{ln}\left(\frac{1+2\sqrt{Vϵ}}{12\sqrt{Vϵ}}\right)`$; after that, the expansion gives way to contraction, and the universe collapses by the instant $`t_c=2t_m`$. For $`2\sqrt{V}t1`$, the solution in (9) takes the form
$$a(t)\left[2\sqrt{ϵ}t\right]^{1/2}.$$
(12)
It is independent of $`V`$ and describes the evolution of the universe that is dominated by radiation and which expands in the de Sitter mode from the point $`a=a_2`$. In the extreme case of $`ϵ=0`$, where there is no radiation, the region $`aa_1`$ contracts to the point $`a=0`$, and the expansion can proceed only from the point $`a=a_2`$. Since the region $`a<a_2`$ cannot be treated in terms of classical theory, it is assumed that the classical spacetime with $`a=a_2`$ is formed as the result of a tunnel transition from ”nothing” taken to mean some quantum state of the protouniverse (see, for example, ). If, originally, the universe was filled not only with matter but also with radiation, it can undergo evolution in the region $`aa_1`$ as well. In the general theory of relativity, the solutions in (9) and (10) describe two independent scenarios of the evolution. The inclusion of the mechanism of quantum tunneling through the barrier $`U`$ requires a joint analysis of these scenarios. It is then legitimate to consider the probabilities of finding the universe in each of the classically accessible regions.
The evolution of the universe depends on the initial distribution of the classical field $`\varphi `$ and its subsequent behavior as a function of time. The chaotic-inflation scenario , which is realized in the region $`a>a_2`$, is described by equations (7) and (8) as applied to the case specified by the inequalities $`(\frac{d\mathrm{ln}V}{d\varphi })^21,V|\frac{1}{a^2}\frac{ϵ}{a^4}|,\text{and}\frac{1}{a^2}|\ddot{\varphi }||\frac{dV}{d\varphi }|`$. In the model where the scalar-field potential $`V`$ is taken to be proportional to $`\varphi ^n`$, the chaotic-inflation process proceeds between scalar-field values greatly exceeding a level of $`\frac{n}{3\sqrt{2}}`$ (initial stage) and those achieving this level (final stage). In this approach, radiation has virtually no effect on the degree of inflation, and the scalar field represents the field of an inflaton. The de Sitter regime of inflation persists as long as the potential $`V(\varphi (t))`$ varies rather slowly with time. From equations (8) and (11), it follows that the inequality $`\dot{V}+\dot{ϵ}/a^4<0`$ holds in the expanding universe $`(\dot{a}>0)`$. If the potential $`V`$ increases with time, the quantity $`ϵ`$ is bound to decrease. But if $`V`$ decreases, $`ϵ`$ can increase, and the rate of this increase is higher for greater $`a`$. We will now estimate $`ϵ`$ by using the relation $`ϵ\stackrel{~}{T}_0^0a^4`$. In our Universe, with $`a10^{28}`$ cm, the main contribution to the radiation-energy density comes from cosmic microwave background radiation with energy density $`\rho _\gamma ^010^{10}`$ GeV/$`\text{cm}^3`$. Setting $`\stackrel{~}{T}_0^0=\rho _\gamma ^0`$, we find that, at the present era, the result is $`ϵ=ϵ_\gamma 10^{117}`$. In the early Universe, the scale parameter is $`a10^{33}`$ cm, while the energy density $`\stackrel{~}{T}_0^0`$ is on the order of the Planck value. On this basis, it can be found that $`ϵ1`$ corresponds to that era. It follows that $`ϵ`$ increased in the evolution process. This increase can be explained by a considerable redistribution of energy between the scalar field and radiation at the initial stage of Universe existence. Quantum theory is able to account for this phenomenon in a natural way (see below). In the region $`a>a_2`$, the possible variation of $`ϵ`$ with time does not affect the quasiexponential expansion of the universe, because the inflation stage terminates in a rather short time interval of $`t10^{37}`$ s , and the evolution process is then determined by other factors (particle production, heating, etc.). In the region $`a<a_1`$, the role of the increase in $`ϵ`$ with decreasing $`V`$ may prove substantial. In principle, the dependence of $`ϵ`$ on $`\varphi (t)`$ makes it possible to provide the missing power in the $`t`$ dependence of $`a`$ and to solve the problem of the Universe size. In order to demonstrate this explicitly, we assume that, up to the present time $`t_0`$, the Universe has expanded according to the law specified by (12) . We then have $`a(t_0)/a(t_p)=\left(\sqrt{ϵ_0/ϵ_p}t_0/t_p\right)^{1/2}`$, where $`ϵ_0=ϵ(\varphi (t_0))`$, $`ϵ_p=ϵ(\varphi (t_p))`$, and $`t_p`$ is the Planck time. The ratio of $`ϵ_0`$ and $`ϵ_p`$ can be estimated as $`ϵ_0/ϵ_pV_p/V_0`$, where $`V_p=V(\varphi (t_p))`$ and $`V_0=V(\varphi (t_0))`$. Assuming that, in the Planck era, $`V_p`$ is on the order of the Plank energy density and that $`V_0`$ is on the order of the mean matter-energy density at the present era, $`\rho _0=10^5\text{GeV}/\text{cm}^3`$, we find that the value of $`a(t_0)10^{28}`$ cm corresponds to $`a(t_p)10^{33}`$ cm.
3. QUANTIZATION
3.1. Schrödinger Equation
In quantum theory, the constraint equation (6) comes to be a constraint on the wave function that describes the universe filled with a scalar field and radiation. Replacing the canonically conjugate variables involved in equation (7) by the operators $`\widehat{a}=a\times ,\widehat{\pi _a}=i_a,\widehat{\varphi }=\varphi \times ,\widehat{\pi _\varphi }=i_\varphi ,\text{and}\widehat{P}=i_T`$, we find that the state vector $`a,\varphi |\mathrm{\Psi }(T)`$ satisfies the equation $`a,\varphi |\mathrm{\Psi }(T)`$
$$2i_T|\mathrm{\Psi }(T)=\left[_a^2\frac{2}{a^2}_\varphi ^2U\right]|\mathrm{\Psi }(T).$$
(13)
where the order parameter is assumed to be zero \[5, 10-13\]. Equation (13) represents an analog of the Schrödinger equation with a Hamiltonian independent of the time variable $`T`$. The momentum $`\widehat{P}`$ associated with radiation appears linearly in equation (13). We can introduce a positive definite scalar product $`\mathrm{\Psi }|\mathrm{\Psi }<\mathrm{}`$ and specify the norm of a state . This makes it possible to define a Hilbert space of physical states and to construct quantum mechanics for the universe model being considered.
A partial solution to equation (13) has the form
$$|\mathrm{\Psi }(T)=|\psi \mathrm{exp}\left\{\frac{i}{2}E\left(TT_0\right)\right\},$$
(14)
where the state $`\psi `$ satisfies the time-independent equation
$$\left(_a^2+\frac{2}{a^2}_\varphi ^2+UE\right)|\psi =0.$$
(15)
The quantity $`E`$ is arbitrary in the general theory of relativity, but, in quantum theory, it is quantized in accordance with solutions to equation (15).
3.2. Quasistationary States
In considering the quantum case, we assume that, at the initial stage, the motions occurring in the system under study can be separated into two types: a slow variation of the scalar field, in which case the operator $`(2/a^2)_\varphi ^2`$ can be treated as a perturbation, and fast changes in geometry. This assumption is a quantum analog of the adiabaticity hypothesis, which leads, in the zero-order approximation, to the solutions given by (9) and (10). In quantum theory, the problem being considered reduces to solving the equation
$$\left[_a^2U+ϵ_n(\varphi )\right]|\phi _n=0.$$
(16)
The wave functions $`\phi _n`$ and the eigenvalues $`ϵ_n`$, which depend on $`\varphi `$ parametrically, describe the evolution of the universe for very slow variations of the potential $`V`$ associated with the field $`\varphi `$ (more specifically, under the condition $`|d\mathrm{ln}V/d\varphi |1`$). In order to take into account the variations of the field $`\varphi `$, we can represent $`\psi `$ as an expansion in terms of the states $`\phi _n`$ and integrate then equation (15). The quantum number $`n`$ of the system unperturbed by the operator $`(2/a^2)_\varphi ^2`$ will be a good quantum number for the universe in the perturbed state $`\psi `$ as well.
We will further consider solutions to equation (16), allowing for the possible boundary conditions. For $`ϵ_n1/4V`$ the classically accessible regions $`aa_1`$ and $`aa_2`$ are bounded by the turning points $`a_1`$ and $`a_2`$, which are now dependent on the state of the quantum system. In the region $`a>a_2`$, a general solution has the form of a superposition of converging and diverging waves. In the Wentzel-Kramers-Brillouin (WKB) approximation, we can write
$`\phi _n={\displaystyle \frac{1}{(ϵ_nU)^{1/4}}}\left\{C_1\text{e}^{i\underset{a_2}{\overset{a}{}}\sqrt{ϵ_nU}𝑑a\frac{i\pi }{4}}+C_2\text{e}^{i\underset{a_2}{\overset{a}{}}\sqrt{ϵ_nU}𝑑a+\frac{i\pi }{4}}\right\},`$ (17)
where $`C_1`$ is the amplitude of an ”incident” wave describing the universe whose scale factor decreases, while $`C_2`$ is the amplitude of the wave ”traveling” toward greater values of $`a`$ and describing the expanding universe. For the extreme case of $`ϵ_n=0`$, which corresponds to the radiation-free universe with an undetermined time variable, the wave function in the form (17) was studied by many authors (see, for example, \[3, 5, 10-13\]). It coincides with the Vilenkin wave function at $`C_1=0`$ and generalizes the Hartle-Hawking wave function to the case of $`ϵ_n0`$.
If we consider a universe formed at an instant separated by a comparatively large time interval \[$`\mathrm{}<(TT_0)0`$\] from the commencement of observation, the boundary condition $`C_1=0`$ will imply that the diverging wave corresponding to a quasistationary state is singled out from the superposition in (17). No situation that is physically realizable can exactly correspond to the requirement $`C_1=0`$ for all instants of time because, in that case, the process that leads to the formation of a quasistationary state and which involves converging waves would be eliminated from the analysis. According to the general concepts of quantum theory , a quasistationary state can be implemented approximately by requiring that the region where the asymptotic form (17) with $`C_1=0`$ is realized be bounded by the condition $`aa_{max}`$, in which cases $`\phi _n`$ is set to zero for $`a>a_{max}`$, $`a_{max}\sqrt{ϵ_n}T`$ being some boundary value of the scale factor. That the instants of time that satisfy the conditions $`T<T_0`$ and $`T>a_{max}/\sqrt{ϵ_n}`$ are excluded from the analysis is physically justified because this makes it possible to avoid speculations about the properties of the scalar field and radiation in the Universe at times that are practically inaccessible to observation and for which there are no reliable hints from high-energy physics.
The possibility of quantum tunneling through the region $`a_1aa_2`$ of the potential barrier results in that stationary states cannot be realized in the region $`aa_1`$. If, however, the potential $`V(\varphi )`$ is sufficiently small, quasistationary states whose lifetime is much greater than the Planck time can exist in the region $`aa_1`$. The probability $`\mathrm{\Gamma }_n`$ of the decay of the universe occurring in a given quasistationary state $`\phi _n`$ can be found by requiring that the wave function in (17) satisfy the radiation condition $`C_1=0`$, which selects the discrete complex values $`\stackrel{~}{ϵ}_n=ϵ_n+i\mathrm{\Gamma }_n`$.
We further impose the boundary condition $`\phi _n|{}_{a=0}{}^{}=0`$ on the wave function $`\phi _n`$. For the case of $`\mathrm{\Gamma }_nϵ_n`$, we then obtain
$`\mathrm{\Gamma }_n=2\left[{\displaystyle \underset{0}{\overset{a_1}{}}}{\displaystyle \frac{da}{\sqrt{ϵ_nU}}}\right]^1\mathrm{exp}\left\{\mathrm{\hspace{0.17em}2}{\displaystyle \underset{a_1}{\overset{a_2}{}}}\sqrt{Uϵ_n}𝑑a\right\},`$ (18)
where $`ϵ_n`$ is determined from the equation
$$\underset{0}{\overset{a_1}{}}\sqrt{ϵ_nU}𝑑a=\pi \left(n+\frac{3}{4}\right).$$
(19)
In the extreme case of $`V=0`$, equation (19) can easily be integrated, which yields $`ϵ_n|_{V=0}ϵ_n^{(0)}=4n+3`$; that is, we can see that, for all values of $`n`$, $`ϵ_n^{(0)}`$ coincides with the energy of an isotropic oscillator with zero orbital angular momentum . According to (19), the first level (that at $`ϵ_03`$) emerges at $`V0.08`$.
At small $`V`$, equation (19) leads to $`ϵ_n`$ values coincident with those that are obtained directly from equation (16) by perturbation theory in $`a^4V(\varphi )`$. The table displays $`ϵ_n`$ values calculated by perturbation theory and in the WKB approximation \[that is, with the aid of equation (19)\]. Also presented in this table are the decay-probability values $`\mathrm{\Gamma }_n`$ computed for various potentials $`V`$. Since $`\mathrm{\Gamma }_n\text{Re}\stackrel{~}{ϵ}_n`$, the decay-probability values $`\mathrm{\Gamma }_n`$ found on the basis of (18) are expected to be close to true values for small $`n`$ as well.
We note that the smaller the value of $`a^4V`$ at given $`ϵ_n`$, the higher and the broader is the potential barrier $`U`$ and, hence, the smaller is the decay probability $`\mathrm{\Gamma }_n`$. If some state $`\phi _n`$ is characterized by a small value of $`\mathrm{\Gamma }_n`$, the possibility that this state decays can be disregarded over the decay time $`\tau =1/\mathrm{\Gamma }_n`$, so that this state can be considered to be stationary in this limit. This corresponds to defining a quasistationary state as that which takes the place of a stationary state when the probability of its decay becomes nonzero .
In describing the universe on the basis of equation (15), the process of universe production from ”nothing” in the radiation-free model ($`E=0`$) is replaced by quantum tunneling from a quasistationary state with a definite (complex) value of $`E`$. It can easily be seen that the solution given by (17) describes the de Sitter regime of expansion according to (10). In order to demonstrate this explicitly, we note that, in the WKB approximation, we have $`i_a\phi _n\sqrt{ϵ_nU}\phi _n`$; that is, the classical momentum is given by $`\pi _a=\dot{a}=\sqrt{ϵ_nU}`$, whence we obtain equation (7) in approximation $`|\dot{a}/a||\dot{\varphi }|`$.
In quantum models not featuring radiation, the DeWitt-Wheeler equation determines the time-independent wave function of the universe. This leads to well-known difficulties in interpreting this function and in comparing results obtained on its basis with the observed evolution of our Universe . If, however, the case of $`E=0`$ is considered as the limit to which the model with a privileged reference frame reduces when $`E0`$, we can also speak about the time evolution of the Universe free from radiation .
3.3. Dynamics in the Prebarrier Region
By studying inflationary scenarios of the evolution of a universe filled with a scalar field, it was revealed that a ”realistic” potential $`V`$ must decrease with time . With decreasing $`V`$, the number of quantum states in which the universe can occur increases, while the decay probabilities decrease sharply (see table). The first instants of the existence of the universe are especially favorable for its tunneling through the potential barrier $`U`$.
A quasistationary state $`\phi _n`$ takes the place of the stationary state whose wave function in the region $`a<a_1`$; is close to the wave function $`|n`$ of the state unperturbed by the interaction $`a^4V`$. In the approximation of a slowly varying field $`\varphi `$, transitions in the system being studied can be considered as those that occur between the states $`|n`$ and which are induced by the interaction $`a^4V`$. Since this interaction modifies the physical properties of the system, a finite number of its levels and their nonzero widths must be taken into account in calculating the probabilities $`W_{nm}`$ of the $`m(T_0)n(T)`$ transitions. As a result, we arrive at
$$W_{nm}\left|n|𝒰_I(T,T_0)|m\right|^2\mathrm{exp}\left\{\mathrm{\Gamma }_n\mathrm{\Delta }T\right\},$$
(20)
where $`\mathrm{\Delta }T=TT_0`$ and $`𝒰_I`$ is the evolution operator in the interaction representation . Adiabaticity in the field $`\varphi `$ enables us to consider specific transitions in the time interval $`\mathrm{\Delta }T`$ that correspond to a given value of $`V(\varphi )`$.
The figure displays the total probability of universe decay, $`W_{dec}=1\left(W_{00}+W_{10}\right)`$, and the quantity $`W_{10}`$ as calculated at $`V=0.03`$, in which case there are only two levels in the system. It can be seen that, over the time interval $`\mathrm{\Delta }T50`$, the transitions in the system predominate and only for $`\mathrm{\Delta }T100`$ the probability that the universe tunnels through the barrier becomes commensurate with the probability that it undergoes the $`01`$ transition in the prebarrier region.
Since the rate at which the level width $`\mathrm{\Gamma }_n`$ tends to zero is greater than the rate at which the potential decreases, the reduction of $`V`$ with time results in that transitions become much more probable than tunnel decays, in which case the former fully determine the evolution of the quantum universe in the prebarrier region. If the universe has not tunneled through the barrier before the potential $`V`$ of the field $`\varphi `$ decreases to a value less than $`0.01`$, a sufficiently large number of levels such that the probabilities of decays from them can be neglected are formed in it. Assuming that the amplitudes of transitions over the time interval $`\mathrm{\Delta }T`$ are small, $`\left|Vn|a^4|m\right|\mathrm{\Delta }T1`$, we then find that
$`{\displaystyle \frac{W_{n+1,n}}{W_{n1,n}}}>1,{\displaystyle \frac{W_{n+1,n}}{W_{n+2,n}}}>1,{\displaystyle \frac{W_{n+2,n}}{W_{n2,n}}}>1,{\displaystyle \frac{W_{n1,n}}{W_{n+2,n}}}>1,`$ (21)
that is, the $`nn+1`$ transition is more probable than the $`nn1`$ and $`nn+2`$ transitions. This means that the quantum universe can undergo transitions to ever higher levels with a nonzero probability. It is well known that the oscillator amplitude is quantized according to the condition $`\overline{a}\sqrt{n}`$; therefore, it can be concluded that the characteristic size $`\overline{a}`$ of the universe that did not undergo a tunnel transition increases as it is excited to higher levels.
3.4. Parameters of the Early Universe
In the adiabatic approximation, the expectation value $`\overline{a}`$ for the universe occurring in the lowest state $`\phi _0`$ is given by
$$\overline{a}\phi _0|a|\phi _0=\frac{2}{\sqrt{\pi }}\left[1+\frac{21}{16}V+O\left(V^2\right)\right].$$
(22)
whence it follows that $`\overline{a}0.9\times 10^{33}\text{cm}`$ for $`0<V<0.08`$. The value $`\overline{a}`$ determines the mean amplitude of oscillations of the classical universe filled with matter and radiation, thereby specifying its actual linear dimension. The maximal proper distance in a closed universe can be estimated at $`d\pi \overline{a}3\times 10^{33}\text{cm}`$; that is, the universe in the lowest state has a proper dimension on the order of the Planck length . The presence of the minimal length removes the problem of the initial cosmological singularity.
Equations (7) and (8), which are obtained within the general theory of relativity, also dictate the relationship between the quantities $`a,ϵ`$ and $`V`$. By using the value of $`ϵ_02.6`$, which we found for $`V0.08`$, we can estimate the classical turning points at $`a_11.4\times 10^{33}\text{cm}`$ and $`a_22.2\times 10^{33}\text{cm}`$. The value of $`a_1`$ determines the maximal dimension of the universe occurring in the lowest state to the left of the barrier along the $`a`$ axis, while $`a_2`$ characterizes its initial dimension upon tunneling from this state. In this era, the matter- and radiation- energy densities are $`T_0^0V1.3\times 10^{77}\text{GeV}/\text{fm}^3,`$ and $`\stackrel{~}{T}_0^0\frac{ϵ_0}{\overline{a}^4}1.7\times 10^{78}\text{GeV}/\text{fm}^3`$, respectively; that is, we can see that, according to our model, the energy density in the early universe is determined primarily by the radiation-energy density. This result is fully consistent with what is commonly thought about the properties of the universe for $`a0`$ . The total matter-energy density is $`\rho 0.64m_p^4`$, where $`m_p`$ is the Planck mass. Thus, we can see that quite reasonable results are obtained when the parameters $`ϵ`$ and $`V`$ as derived on the basis of quantum theory are used in the equations of the general theory of relativity.
It is interesting to estimate the quantity $`n`$ at the $`\overline{a}`$ value coincident with the dimension of the presently observed part of the Universe. From the relation $`\overline{a}\sqrt{n}`$ at $`\overline{a}10^{28}\text{cm}`$, we obtain $`n10^{122}`$, whence we can see that, if the quantum model being considered is extrapolated to the observed Universe, it occurs in a highly excited state. Quantum corrections to the classical equations of the general theory of relativity are extremely small in this case (they are of order $`1/n`$). The resulting value of $`n10^{122}`$ is consistent with the estimates presented in and is confirmed by rigorous quantum-mechanical calculations within the radiation-free model that was considered in and which is justified for the present, large, values of $`\overline{a}`$ at the matter-dominated stage.
4. CONCLUSION
The presence of radiation in the universe makes it possible to associate a privileged reference frame with it and to remove thereby an ambiguity in choosing the time coordinate. This opens new possibilities both in classical and in quantum cosmology. Upon performing quantization, there naturally arises the Schrödinger equation (13) with an effective interaction $`U`$ in the form of a potential barrier. The evolution of the universe involves a quantum stage that is realized in the prebarrier region and which precedes the process of tunneling through the barrier. The dynamics of this stage is governed by the interaction of the gravitational and scalar fields. That the system in question possesses a spectrum of quantum (quasistationary) states and that transitions can occur between these states enable us to take a fresh look at the problem of the dimension of the Universe. In the approach developed here, the universe is characterized by a minimal length, so that the singularity problem does not arise in it. The probability for the universe to undergo a tunnel transition is maximal in the lowest quantum state, where the energy density and the scale factor are on the same orders of magnitude as the corresponding Planck values. If a quantum universe tunnels from higher states, the dimensions of the region from which tunneling occurs can considerably exceed the Planck length. The constants $`E`$ and $`V`$ appearing in the Einstein equations are determined by the preceding, quantum stage. The use of the parameters in the general theory of relativity that were obtained on the basis of quantum theory leads to conclusions that are consistent with the currently adopted concepts of the early Universe and its subsequent evolution.
REFERENCES
1. Isham, C.J., gr-qc/9510063.
2. Dolgov, A.D., Zeldovich, Ya.B., and Sazhin, M.V., Kosmologiya rannei vselennoi (Cosmology of the Early Universe), Moscow: Mosk. Gos. Univ., 1988.
3. Linde, A.D., Elementary Particle Physics and Inflationary Cosmology, Chur: Harwood, 1990.
4. Arnowitt, R., Deser, S., and Misner, C.W., Gravitation: An Introduction to Current Research, Witten, L., Ed., New York, 1963.
5. DeWitt, B.S., Phys. Rev., 1967, vol. 160, p. 1113.
6. Ponomarev, V.N., Barvinskii, A.O., and Obukhov, Yu.N., Geometrodinamicheskie metody i kalibrovochnyi podkhod k teorii gravitatsionnykh vzaimodeistvii (Methods of Geometrodynamics and Gauge Approach in the Theory of Gravitational Interactions), Moscow: Energoatomizdat, 1985.
7. Kuchař, ., Proc. 4th Canadian Conf. on General Relativity and Astrophysics, Kunstatter, G., Vincent, D., and Williams, J., Eds., Singapore: World Sci, 1992.
8. Kuchař, K.V. and Torre, C.G., Phys. Rev. D: Part. Fields, 1991, vol. 43, p. 419.
9. Landau, L.D. and Lifshitz, E.M., The Classical Theory of Fields, Oxford: Pergamon, 1975.
10. Vilenkin, A., Phys. Rev. D: Part. Fields, 1994, vol. 50, p.2581.
11. Hartle, J.B. and Hawking, S.W., Phys. Rev. D: Part. Fields, 1983, vol. 28, p. 2960.
12. Vilenkin, A., Phys. Rev. D: Part. Fields, 1986, vol. 33, p. 3560.
13. Vilenkin, A., Phys. Rev. D: Part. Fields, 1988, vol. 37, p. 888.
14. Baz’, A.I., Zel’dovich, Ya.B., and Perelomov, A.M., Scattering, Reactions, and Decays in Nonrelativistic Quantum Mechanics, Jerusalem: Israel Program of Sci. Transl., 1966.
15. Blatt, J.M. and Weisskopf, V.R, Theoretical Nuclear Physics, New York: Springer-Verlag, 1979.
16. Davydov, A.S., Quantum Mechanics, Oxford: Pergamon, 1976.
17. Flügge, S., Practical Quantum Mechanics, Berlin: Springer-Verlag, 1971.
18. Wheeler, J.A., Battelle Rencontres, DeWitt, C. and Wheeler, J.A., Eds., New York: Benjamin, 1968, p. 242.
19. Kuzmichev, V.V, Yad. Fiz., 1997, vol. 60, p. 1707 \[Phys. At. Nucl. (Engl. Transl.), vol. 60, p. 1558\].
20. Dirac, P.A.M., The Principles of Quantum Mechanics, Oxford: Clarendon, 1958.
21. Zeldovich, Ya.B. and Novikov, I.D., Relativistic Astrophysics, vol. 2: The Structure and Evolution of the Universe, Chicago: Univ. of Chicago Press, 1983.
|
warning/0002/cond-mat0002432.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
In a recent paper , ben-Avraham et al. treated the 1D dynamics of the diffusion limited reaction $`A+AA+S`$. The analysis led to an infinite hierarchy of differential equations (HDE) for the time evolution of the set of quantities $`\left\{E_k(t)\right\}`$, where $`E_k(t)`$ denotes the probability of finding an interval of $`k`$ contiguous empty lattice sites (c.f. eq. (2.2) and (2.5) in ). Ben-Avraham et al. were primarily interested in the continuum limit for the probability densities $`E_k(t)/(\mathrm{\Delta }x)^k`$ as the lattice spacing $`\mathrm{\Delta }x`$ is allowed to vanish. Here we rederive the HDE taking as a starting point a detailed cellular automaton description using Boolean occupation numbers . Instead of going to the continuum limit, we solve the HDE for a discrete lattice and obtain explicit expressions for the concentration $`c(t)=\left(1E_1(t)\right)/\mathrm{\Delta }x`$ and the two-point nearest neighbour correlation. We consider both the case of an initially fully occupied lattice and the case in which the initial probabilities $`E_k(0)`$ are given by the random “geometrical” distribution $`(1\rho )^k`$, where $`\rho <1`$ is the initial probability of a single site being occupied. Some exact solutions to related problems and a special case can be found in .
In section II, we derive the HDE starting from the cellular automaton dynamical rule. The next section is a brief reminder of the off-lattice solution for the HDE obtained in . In section IV, we solve the solution of the HDE for a discrete lattice and discuss the transient dynamics of the concentration. As expected, the long time behavior is the same as that found in . However, the early time behavior turns out to be significantly different.
## II Derivation of the HDE
Consider a 1D lattice with $`N`$ sites. Each lattice site $`i`$ is characterized by a boolean occupation number $`n_i=1`$ or $`0`$ depending on whether it is occupied by a single particle (A) or empty (S). At each time step $`\delta t`$, we choose randomly a site $`i`$ in the lattice via the boolean stochastic parameter $`\xi _N^{(i)}`$. This parameter is equal to one for the chosen site and $`0`$ for all other sites. Simultaneously, a second boolean variable is used to select the left ($`\xi _L=1`$) or the right neighbour site ($`\xi _L=0`$) with equal probability $`1/2`$. If the site $`i`$ is occupied, the particle will hop to the chosen neighbour site with a rate $`k_D`$ given by the mean value of the stochastic boolean parameter $`\xi _D`$. In this case, if the neighbour site is empty, the particle will occupy it vacating the site $`i`$. We express this by including the loss terms
$`\xi _N^{(i)}(t)\xi _L(t)\xi _D(t)n_i(t)(1n_{i1}(t))`$ (1)
$`\xi _N^{(i)}(t)(1\xi _L(t))\xi _D(t)n_i(t)(1n_{i+1}(t))`$ (2)
in the dynamical rule for $`n_i(t)`$. If the neighbour site is filled, the particle will “react” with it and instantaneously disappear. Again, the occupation number at site $`i`$ will be decreased from $`1`$ to $`0`$. Thus, we have the reactive loss terms:
$`\xi _N^{(i)}(t)\xi _L(t)\xi _D(t)n_i(t)n_{i1}(t)`$ (3)
$`\xi _N^{(i)}(t)(1\xi _L(t))\xi _D(t)n_i(t)n_{i+1}(t)`$ (4)
Finally, as an empty site $`i`$ can only be occupied by hopping from a particle at a neighbour site, one has the two gain terms:
$`+\xi _N^{(i+1)}(t)\xi _L(t)\xi _D(t)n_{i+1}(t)(1n_i(t))`$ (5)
$`+\xi _N^{(i1)}(t)(1\xi _L(t))\xi _D(t)n_{i1}(t)(1n_i(t))`$ (6)
Clearly, the dynamics described above corresponds to the particular implementation of the reaction $`A+AA+S`$ given in . Adding up all the contributions (1)-(6), we obtain the following dynamical rule:
$`n_i(t+\delta t)`$ $`=`$ $`n_i(t)\xi _N^{(i)}(t)\xi _D(t)n_i(t)`$ (9)
$`+\xi _N^{(i+1)}(t)\xi _L(t)\xi _D(t)n_{i+1}(t)(1n_i(t))`$
$`+\xi _N^{(i1)}(t)(1\xi _L(t))\xi _D(t)n_{i1}(t)(1n_i(t)),i=1,\mathrm{},N.`$
In the first and last equation (9), we set $`n_0(t)=n_{N+1}(t)=0`$. We can rewrite (9) by using the complementary occupation numbers $`s_i(t)=1n_i(t)`$:
$`s_i(t+\delta t)`$ $`=`$ $`s_i(t)+\xi _N^{(i)}(t)\xi _D(t)[\xi _N^{(i1)}(t)+\xi _N^{(i)}(t)]\xi _D(t)s_i(t)`$ (12)
$`+\xi _N^{(i1)}(t)(1\xi _L(t))\xi _D(t)s_{i1}(t)s_i(t)`$
$`+\xi _N^{(i+1)}(t)\xi _L(t)\xi _D(t)s_i(t)s_{i+1}(t),i=1,\mathrm{},N.`$
This is a more convenient form, since, as it turns out, the evolution law for a string of $`k`$ consecutive empty sites $`_{j=i}^{i+k1}s_j`$ involves only products of contiguous occupation numbers
$`{\displaystyle \underset{j=i}{\overset{i+k1}{}}}s_j(t+\mathrm{\Delta }t)`$ $`=`$ $`{\displaystyle \underset{j=i}{\overset{i+k1}{}}}s_j(t)+\xi _N^{(i)}(t)\xi _L(t)\xi _D(t){\displaystyle \underset{j=i+1}{\overset{i+k1}{}}}s_j(t)`$ (17)
$`+\xi _N^{(i+k1)}(t)(1\xi _L(t))\xi _D(t){\displaystyle \underset{j=i}{\overset{i+k2}{}}}s_j(t)`$
$`[\xi _N^{(i1)}(t)(1\xi _L(t))+\xi _N^{(i)}(t)\xi _L(t)`$
$`+\xi _N^{(i+k1)}(t)(1\xi _L(t))+\xi _N^{(i+k)}(t)\xi _L(t)]\xi _D(t){\displaystyle }_{j=i}^{i+k1}s_j(t)`$
$`+\xi _N^{(i1)}(t)(1\xi _L(t))\xi _D(t){\displaystyle \underset{j=i1}{\overset{i+k1}{}}}s_j(t)+\xi _N^{(i+k)}(t)\xi _L(t)\xi _D(t){\displaystyle \underset{j=i}{\overset{i+k}{}}}s_j(t)`$
Let us take the time step $`\delta t`$ equal to $`1/N`$, implying that each site has been visited once on average after one time unit. If we now average (17) over an ensemble of realizations for a given initial configuration, we obtain in the thermodynamic limit $`N\mathrm{}`$:
$$\frac{dE_k^i}{dt}=\frac{k_D}{2}\left(E_{k1}^{i+1}+E_{k1}^i4E_k^i+E_{k+1}^{i1}+E_{k+1}^i\right),$$
(18)
where $`E_k^i(t)=\overline{_{j=i}^{i+k1}s_j(t)}`$. For a single site ($`k=1`$) the corresponding equation
$$\frac{dE_1^i}{dt}=k_D\left(12E_1^i+\frac{1}{2}E_2^{i1}+\frac{1}{2}E_2^i\right)$$
(19)
is obtained by averaging the dynamical rule (12). If we perform the average over realizations and translationally invariant initial conditions, we get the following hierarchy of differential difference equations for the evolution of the averaged products $`E_k(t)=_{j=i}^{i+k1}s_j(t)`$:
$$\frac{dE_k}{dt}=k_D\left(E_{k+1}2E_k+E_{k1}\right),k=1,2,\mathrm{}$$
(20)
with the boundary condition $`E_0(t)=1`$. As described in , the rhs of (20) represents the net flux due to particle diffusion into and out of an empty site interval, whereas the effect of reaction enters through the boundary condition.
## III Off-lattice solution
Following ref. , we set the hopping rate $`k_D/2`$ to either of both sides equal to $`D/(\mathrm{\Delta }x)^2`$, where $`\mathrm{\Delta }x`$ is the lattice spacing. On long length and time scales this yields normal diffusion with a diffusion coefficient $`D`$. With this definition, eqs. (20) are identical to those derived in . If we now let $`\mathrm{\Delta }x0`$, eqs. (20) become
$$\frac{E(x,t)}{t}=2D\frac{^2E(x,t)}{x^2}.$$
(21)
with the boundary conditions $`E(0,t)=1`$ and $`E(\mathrm{},t)=0`$. In this limit, the concentration (number of particles per unit length) is expressed as
$$c(t)=\frac{E(x,t)}{x}|_{x=0}$$
(22)
Thus, one can determine the time dependence of the concentration by solving (21) with the boundary conditions given above (see for details). For the special case of an initially random particle distribution with a concentration $`c_0`$, one has
$$\frac{c(t)}{c_0}=1\left(\frac{8c_0^2Dt}{\pi }\right)^{\frac{1}{2}}+o(c_0^2Dt)\text{as }t0$$
(23)
and
$$c(t)\frac{1}{\left(2\pi Dt\right)^{\frac{1}{2}}}\text{as }t\mathrm{}$$
(24)
## IV Solution for a discrete lattice
We now proceed to solve the HDE (20) for the case in which one has an initially full lattice, i.e. $`E_k(0)=0`$ for all $`k1`$. To begin with, we absorb the rate constant $`k_D`$ into the time scale by introducing the dimensionless time variable $`\tau =k_Dt`$. The hierarchy then reads:
$$\frac{dE_k}{d\tau }=E_{k+1}2E_k+E_{k1},k=1,2,\mathrm{}$$
(25)
Next we apply the Laplace transform to both sides of (25) and obtain the homogeneous difference equation
$$\overline{E}_{k+1}(2+s)\overline{E}_k+\overline{E}_{k1}=0,k1,$$
(26)
where $`\overline{E}_k(s)=L_{\tau s}\left\{E_k(\tau )\right\}={\displaystyle _0^{\mathrm{}}}\mathrm{exp}(s\tau )E_k(\tau )𝑑\tau `$. The boundary condition is given by $`\overline{E}_0(s)=1/s`$. This second-order difference equation is solved with the ansatz $`E_k(s)=\lambda ^k(s)`$. This leads to the quadratic equation
$$\lambda ^2(2+s)\lambda +1=0$$
(27)
which has the two solutions
$$\lambda _\pm =\frac{s+2\pm \sqrt{s^2+4s}}{2}$$
(28)
The general solution of (26) is obtained as a linear superposition of $`\lambda _+^k`$ and $`\lambda _{}^k`$:
$$\overline{E}_k(s)=A(s)\lambda _{}^k(s)+B(s)\lambda _+^k(s)$$
(29)
For $`k\mathrm{}`$, the physically acceptable solution of (26) must satisfy the implicit boundary condition $`E_{\mathrm{}}(s)=0`$. To avoid the divergence of the second term in (29), we must therefore set $`B(s)=0`$. Using the other boundary condition at $`k=0`$, we find $`A(s)=1/s`$. Thus, we have:
$$\overline{E}_k(s)=\frac{1}{s}\left(\frac{s+2\sqrt{s^2+4s}}{2}\right)^k$$
(30)
Clearly, the most interesting quantity is $`\overline{E}_1(s)`$, whose inverse Laplace transform $`E_1(\tau )`$ is the probability that a randomly chosen site be empty. By virtue of a theorem , the inverse transform $`L_{s\tau }^1\{\overline{E}_k(s)\}`$ is given by the integral
$$_0^\tau v_k(\tau ^{})𝑑\tau ^{},k=1,2,\mathrm{}.$$
(31)
where
$$v_k(\tau )=L_{s\tau }^1\left\{\left(\frac{s+2\sqrt{s^2+4s}}{2}\right)^k\right\}=k\frac{\mathrm{exp}(2\tau )I_k(2\tau )}{2\tau }$$
(32)
(see e.g. \[5, page 379\]), where
$$I_n(x)=\underset{r=0}{\overset{\mathrm{}}{}}\frac{(x/2)^{2r+n}}{r!\mathrm{\Gamma }(r+n+2)}$$
(33)
are the modified Bessel functions. In particular, one has
$$E_1(\tau )=_0^\tau \frac{\mathrm{exp}(2\tau ^{})I_1(2\tau ^{})}{\tau ^{}}𝑑\tau ^{}.$$
(34)
### A Asymptotics for early times
For sufficiently short times ($`\tau 1`$), we can use (33) and the series expansion of the exponential function $`\mathrm{exp}(2\tau ^{})`$ to expand the integrand in (34) in powers of $`\tau ^{}`$. Neglecting terms of order $`o(\tau ^3)`$, performing the integration and undoing the time scaling, we obtain:
$$E_1(t)=k_Dtk_D^2t^2+\frac{5}{6}k_D^3t^3+o(t^4)$$
(35)
The particle concentration is
$$c(t)=\frac{P_1(t)}{\mathrm{\Delta }x}=\frac{1E_1(t)}{\mathrm{\Delta }x}=c_0\left[12c_0^2Dt+4c_0^4D^2t^2+o(c_o^6D^3t^3)\right],$$
(36)
where $`c_0=1/\mathrm{\Delta }x`$. This is in clear disagreement with the decay law (23).
For sufficiently short times, we can neglect the effect of large clusters of vacant sites, since the chain is initially full. This is done by setting $`E_k=0`$ for $`k`$ larger than a certain cutoff size $`k_c`$ in the truncation hierarchy (25). If we set $`k_c=1`$, we obtain the differential equation
$$\frac{dE_1}{d\tau }=12E_1(\tau ).$$
(37)
The solution reads
$$E_1(\tau )=\frac{1}{2}\left(1\mathrm{exp}(2\tau )\right)=\tau \tau ^2+\frac{2}{3}\tau ^3+o(\tau ^4)$$
(38)
Setting $`\tau =k_Dt`$, we see that the early times expansion (35) is reproduced correctly up to the quadratic term. The discrepancy between (35) and the off-lattice solution arises due to the finite propagation velocity of a local perturbation in concentration, as opposed to the infinite propagation velocity characteristic of diffusion. Thus, we expect a slower decay of the concentration $`c(t)`$ on the lattice.
### B Long time asymptotics
For large times ($`\tau 1`$), we write $`E_1(\tau )`$ as follows:
$$E_1(\tau )=A_\tau ^{\mathrm{}}\frac{\mathrm{exp}(2\tau ^{})I_1(2\tau ^{})}{\tau ^{}}𝑑\tau ^{}.$$
(39)
where $`A`$ is the definite integral
$$_0^{\mathrm{}}\frac{\mathrm{exp}(2\tau ^{})I_1(2\tau ^{})}{\tau ^{}}𝑑\tau ^{},$$
(40)
which is equal to 1 \[6, page 236\]. The integrand in the second term of (39) can be expanded using the asymptotic form
$$I_1(x)=\frac{\mathrm{exp}(x)}{\sqrt{2\pi x}}\left(1\frac{3}{8x}+o\left(\frac{1}{x^2}\right)\right).$$
(41)
for large $`x`$ ( see e.g. \[7, page 489\] ). Thus, we get
$`E_1(\tau )`$ $`=`$ $`1{\displaystyle \frac{1}{2\sqrt{\pi }}}{\displaystyle _\tau ^{\mathrm{}}}\tau ^{\frac{3}{2}}𝑑\tau ^{}+{\displaystyle \frac{3}{32\sqrt{\pi }}}{\displaystyle _\tau ^{\mathrm{}}}\tau ^{\frac{5}{2}}𝑑\tau ^{}+o(\tau ^{\frac{5}{2}})`$ (42)
$`=`$ $`1{\displaystyle \frac{1}{\sqrt{\pi \tau }}}+{\displaystyle \frac{1}{16\sqrt{\pi \tau ^3}}}+o(\tau ^{\frac{5}{2}}).`$ (43)
This is in agreement with the long time asymptotics of the continuum limit solution (24).
### C Two-point correlation
The explicit expression for the two interval probability
$$E_2(\tau )=2_0^\tau \frac{\mathrm{exp}(2\tau ^{})I_2(2\tau ^{})}{\tau ^{}}𝑑\tau ^{}.$$
(44)
can be used as a starting point to compute the asymptotics of the two-point nearest neighbour correlation $`c^{(2)}(t)=P_2/(\mathrm{\Delta }x)^2`$, where $`P_2=12E_1+E_2`$ is the probability of finding two contiguous particles in the lattice. For early times one gets $`c^{(2)}(t)=c_0^2\left[1c_0^2Dt+o(c_0^4D^2t^2)\right]`$, whereas for long times $`c^{(2)}(t)1/c_0\sqrt{32\pi }(Dt)^{3/2}`$.
### D Solution for an arbitrary initial concentration
In this case, we average over all possible initial configurations of the lattice with a given concentration $`c_0=\rho /\mathrm{\Delta }x`$ ($`\rho <1`$). The initial conditions for the empty interval probabilities now read
$$E_k(0)=(1\rho )^k,k=1,2,\mathrm{}$$
(45)
The boundary conditions are the same as in the case of an initially full lattice. If we now apply the Laplace transform to (25), we obtain
$$\overline{E}_{k+1}(2+s)\overline{E}_k+\overline{E}_{k1}=(1\rho )^k,k1,$$
(46)
which differs from (26) by the inhomogeneity on the rhs. Like in the theory of ordinary differential equations, the general solution of (46) can be written as the sum of the general solution (29) for (26) and a particular solution which we seek in the form
$$\overline{E}_k^{par}(s)=(1\rho )^kC(s).$$
(47)
Inserting this ansatz in (46), we find
$$C(s)=\frac{(1\rho )}{(1\rho )s\rho ^2}.$$
(48)
The boundary condition for $`k\mathrm{}`$ again imposes $`B(s)=0`$. From the boundary condition for $`\overline{E}_0(s)`$ we obtain
$$A(s)=\frac{1}{s}\frac{1\rho }{(1\rho )s\rho ^2}.$$
(49)
Putting (48) and (49) into the equation
$$\overline{E}_k(s)=A(s)\lambda _{}^n(s)+\overline{E}_k^{par}$$
(50)
we find
$$\overline{E}_k(s)=\left(\frac{1}{s}\frac{1\rho }{(1\rho )s\rho ^2}\right)\left[\frac{s+2\sqrt{s^2+4s}}{2}\right]^k+\frac{(1\rho )^{k+1}}{(1\rho )s\rho ^2},k=0,1,\mathrm{}$$
(51)
We can now use the convolution theorem for the Laplace transform to invert (51). This yields
$`E_k(\tau )`$ $`=`$ $`k{\displaystyle _0^\tau }{\displaystyle \frac{\mathrm{exp}(2\tau ^{})I_k(2\tau ^{})}{\tau ^{}}}𝑑\tau ^{}`$ (53)
$`+\left[\left(1\rho \right)^kk{\displaystyle _0^\tau }{\displaystyle \frac{\mathrm{exp}\left(\left[2+\frac{\rho ^2}{1\rho }\right]\tau ^{}\right)I_k(2\tau ^{})}{\tau ^{}}}𝑑\tau ^{}\right]\mathrm{exp}\left({\displaystyle \frac{\rho ^2\tau }{1\rho }}\right).`$
However, we can study the asymptotics directly from (51) by making use of the Tauberian theorems , which allow us to determine the behaviour of $`E_1(\tau )`$ for $`\tau 0`$ and $`\tau \mathrm{}`$ by inverting respectively the series expansion of $`\overline{E}_1(s)`$ around $`s=\mathrm{}`$ and $`s=0`$. For small $`s`$ one has
$$\overline{E}_1=\frac{1}{s}\frac{1}{\sqrt{s}}+\frac{1}{2}+\frac{1\rho }{\rho }+o(s^{\frac{1}{2}})$$
(54)
from which we get
$$E_11\frac{1}{\sqrt{\pi \tau }}\text{for}\tau \mathrm{}.$$
(55)
Again, this is in agreement with $`(\text{24})`$. The long time asymptotics does not depend on the initial concentration $`c_0`$, suggesting that a universal behavior also takes place on a finite lattice. In the opposite limit we have
$$\overline{E}_1=\frac{1\rho }{s}+\frac{\rho ^2}{s^2}\frac{\rho ^2(1+\rho )}{s^3}+o(s^4)$$
(56)
Inverting term by term the rhs of (56), we get
$$E_1=1\rho +\rho ^2\tau \frac{\rho ^2(1+\rho )}{2}\tau ^2+o(\tau ^3)$$
(57)
leading to
$$c(t)=c_0\left[1\frac{2}{\rho }c_0^2Dt+2\frac{\rho ^2(\rho +1)}{\rho ^5}c_0^4D^2t^2+o(c_0^6D^4t^4)\right]$$
(58)
Thus, the discrepancy with (23) appears to be robust. We can again compare the exact early time expansion (57) with the solution of the differential equation (37) with the initial condition $`E_1(0)=1\rho `$, which is given by
$$\frac{1}{2}+\left(\frac{1}{2}\rho \right)\mathrm{exp}(2\tau )=1\rho +(2\rho 1)\tau +(12\rho )\tau ^2+o(\tau ^3).$$
(59)
As expected, the approximation based on the neglect of empty intervals becomes worse as $`\rho `$ decreases.
## V Acknowledgments
This work was supported, in part, by NSF grant DMR 9628224, the Training and Mobility of Researchers program of the European Commission and by the Interuniversity Attractions Poles program of the Belgian Federal Government. E. Abad acknowledges helpful discussions with F. Vikas and F. Baras.
|
warning/0002/math0002083.html
|
ar5iv
|
text
|
# Deformation of singular lagrangian subvarieties
## 1 Introduction
In this paper, we develop some ideas of a deformation theory of singular lagrangian subvarieties. Lagrangian submanifolds are quite fundamental objects, so in a sense it is natural to extend the study of them to a larger class of objects which are allowed to have singularities. This has been done by Arnold, Givental and others (\[Giv88\]). However, not much is known on the behavior of lagrangian singularities under deformations. The aim of this article is to describe the spaces of infinitesimal deformations and obstructions of a lagrangian subvariety and to perform calculations for some concrete examples. It turns out that the lagrangian property of a space has a strong influence on its deformations, e.g., there are examples of spaces $`X`$ with $`dim(T_X^1)=\mathrm{}`$, which have nevertheless a versal deformation space for the lagrangian deformations.
In the sequel, we will consider the following situation: Let $`M`$ be a $`2n`$-dimensional symplectic manifold over $`𝕂=`$ or $`𝕂=`$ (that is, a $`C^{\mathrm{}}`$ or complex analytic manifold of real resp. complex dimension $`2n`$ endowed with a closed, non-degenerated 2-form $`\omega `$, holomorphic in the second case) and $`L`$ a reduced analytic subspace of dimension $`n`$, given by an involutive ideal sheaf $``$, i.e. an ideal sheaf satisfying $`\{,\}`$ where $`\{,\}`$ denotes the Poisson bracket corresponding to $`\omega `$. This condition ensures that $`L`$ is a lagrangian submanifold in a neighborhood of each of its smooth points. A lagrangian deformation of $`L`$ will be a deformation in the usual sense (a flat family $`L_SS`$) with the additional condition that all fibers are lagrangian subvarieties of $`M`$. More precisely, we will call a diagram
a lagrangian deformation of $`L`$ iff $`L_SS`$ is flat and $`\{_S,_S\}_S_S`$. Here $`_S`$ is the ideal sheaf defining $`L_S`$ in $`M\times S`$ and $`\{,\}_S`$ is the Poisson structure defined on $`M\times S`$ by the (degenerate) form $`\omega _S:=p^{}\omega `$, $`p:M\times SM`$ being the canonical projection. This definition can be formalized using the language of deformation functors (see \[Sev99\] and \[Sch68\]). This more formal approach yields the definition of morphisms of deformations, in particular, two deformations $`L_SM\times S`$ and $`L_T^{}N\times T`$ are called equivalent iff there is an fibrewise isomorphism $`F:M\times SN\times T`$ satisfying $`F^{}\omega _T=\omega _S`$. Such an $`F`$ comes from a symplectic diffeomorphism $`f:MN`$ and in case that $`M`$ is simply connected (which we will suppose from now on), $`f`$ is induced by an hamiltonian vector field, see lemma 3.
The tangent space to the functor of lagrangian deformations of $`L`$ (that is, the space of lagrangian deformations of $`L`$ over $`Spec(𝕂[ϵ])`$ up to those induced by hamiltonian vector fields of the ambient manifold) will be denoted by $`LT_L^1`$. However, we will focus our attention to the local case mainly, that is, we will study the *sheaf* $`𝒯_L^1`$ of lagrangian deformations of $`L`$. For lagrangian submanifolds, it follows from \[Voi92\], that the versal deformation space is smooth, i.e., deformation of such objects are unobstructed. This is probably not true in the singular case, although an example has not been found yet. See theorem 1 for further details.
Acknowledgements: We would like to thank A. Givental who suggested to investigate the deformation theory of lagrangian singularities in december 1992.
## 2 The complex $`𝒞^{}`$
We start with a slightly more general situation: Let $`𝒪_M`$ be an involutive ideal sheaf, $`𝒪_L`$ the structure sheaf of the subvariety $`L`$ described by $``$ and denote by $`:=/^2`$ the conormal sheaf. The formula $`\{^i,^j\}^{i+j1}`$, which can be easily verified, shows that there are well-defined operations
$$\begin{array}{ccc}\hfill \times 𝒪_L& & 𝒪_L\hfill \\ \hfill (g,f)& & \{g,f\}\hfill \end{array}\text{and}\begin{array}{ccc}\hfill \times & & \hfill \\ \hfill (g,h)& & \{g,h\}\hfill \end{array}$$
compatible in the sense that $`\{g,fh\}=\{g,f\}h+f\{g,h\}`$. This implies that the first operation can be rewritten as a morphism
$$𝒟er(𝒪_L,𝒪_L)=\mathrm{\Theta }_L$$
One says that $``$ is a *Lie algebroid* (for details on Lie algebroids, see \[Mac87\]).
###### Definition 1.
Let $`𝒞_L^p`$ be the following $`𝒪_L`$-module
$$𝒞_L^p:=om_{𝒪_L}(\stackrel{p}{},𝒪_L)$$
and define a differential:
$$\begin{array}{c}\left(\delta \left(\varphi \right)\right)\left(h_1\mathrm{}h_{p+1}\right):=\hfill \\ _{i=1}^{p+1}\left(1\right)^i\{h_i,\varphi \left(h_1\mathrm{}\widehat{h}_i\mathrm{}h_{p+1}\right)\}\hfill \\ +\underset{1i<jp+1}{}\left(1\right)^{i+j1}\varphi \left(\{h_i,h_j\}h_1\mathrm{}\widehat{h}_i\mathrm{}\widehat{h}_j\mathrm{}h_{p+1}\right)\hfill \end{array}$$
It is a straightforward computation to check that $`\delta \delta =0`$, so we get indeed a complex. Following \[Mac87\], it is called the standard complex for the Lie algebroid $``$. Remark that $`𝒞^0=𝒪_L`$ and $`𝒞^1=om_{𝒪_L}(/^2,𝒪_L)=:𝒩_L`$, the normal sheaf of $``$ in $`𝒪_M`$. For the definition of $`\delta `$, the fact that $``$ is involutive is essential: the second term would not make sense otherwise.
We may define a product on the complex $`(𝒞^{},\delta )`$:
$`𝒞^p\times 𝒞^q`$ $``$ $`𝒞^{p+q}`$
$`(\mathrm{\Phi },\mathrm{\Psi })`$ $``$ $`\mathrm{\Phi }\mathrm{\Psi }`$
with
$$\begin{array}{ccc}(\mathrm{\Phi }\mathrm{\Psi })(f_1\mathrm{}f_{p+q})=\hfill & & \\ & & \\ \underset{\begin{array}{c}I{\scriptscriptstyle J}=\{1,\mathrm{},n\}\\ i_1<\mathrm{}<i_p\\ j_1<\mathrm{}<j_q\end{array}}{}sgn(I,J)\mathrm{\Phi }(f_{i_1}\mathrm{}f_{i_p})\mathrm{\Psi }(f_{j_1}\mathrm{}f_{j_q})\hfill & & \end{array}$$
The sign is defined as
$$sgn(I,J):=sgn\left(\genfrac{}{}{0pt}{}{1,,p+q}{i_1,\mathrm{},i_p,j_1,\mathrm{},j_q}\right)$$
###### Proposition 1.
Let $`\mathrm{\Phi }𝒞^p`$, $`\mathrm{\Psi }𝒞^q`$ et $`\mathrm{\Gamma }𝒞^r`$. Then we have
1. $`\mathrm{\Phi }\mathrm{\Psi }=(1)^{deg(\mathrm{\Phi })deg(\mathrm{\Psi })}\mathrm{\Psi }\mathrm{\Phi }`$
2. $`(\mathrm{\Phi }\mathrm{\Psi })\mathrm{\Gamma }=\mathrm{\Phi }(\mathrm{\Psi }\mathrm{\Gamma })`$
3. $`\delta (\mathrm{\Phi }\mathrm{\Psi })=\delta (\mathrm{\Phi })\mathrm{\Psi }+(1)^{deg(\mathrm{\Phi })}\mathrm{\Phi }\delta (\mathrm{\Psi })`$
###### Proof.
The first two points are trivial, while the third has to be checked by an explicit calculation. ∎
Note that the last proposition says that $`(𝒞_L^{},\delta ,)`$ is a *differential graded algebra*, furthermore, we have $`𝒞_L^0=𝒪_L=\mathrm{\Omega }_L^0`$. As one might hope, there is indeed a tight connection between $`\mathrm{\Omega }_L^{}`$ and $`𝒞_L^{}`$.
###### Proposition 2.
Suppose that $`L`$ is lagrangian. Then there exists a morphism $`J:\mathrm{\Omega }_L^1𝒞_L^1`$ which is an isomorphism outside the singular locus of $`L`$.
###### Proof.
On a symplectic manifold, there is a canonical isomorphism $`\beta `$ between vector fields and one forms, given by $`\beta (V):=i_V\omega `$. On the other hand, for each analytic subspace $`LM`$ we have two exact sequences, dual to each other, namely, the conormal and the normal sequence, thus, there is the following diagram:
Now the fundamental fact is that this diagram can be completed: the morphism $`\mathrm{\Theta }_L`$ above commutes with $`\alpha `$, so we have
(1)
Note that the image of an element $`g`$ under $`\alpha ^{}`$ is just the hamiltonian vector field $`H_g`$. The morphisms $`J:\mathrm{\Omega }_L^1𝒞_L^1=𝒩_L`$ we are looking for can now be defined as the map induced by $`\alpha `$, explicitly
$$J(df)=\left(g\{f,g\}\right)$$
To see that $`J`$ is an isomorphism near a smooth point of $`L`$ it will be sufficient to prove this for the map $`\alpha ^{}`$ (because at smooth points $`x`$ we have $`𝒯_{(L,x)}^1=0`$ and the map $`_x\mathrm{\Omega }_{(L,x)}^1𝒪_{L,x}`$ is injective). So assume the sheaves $``$, $`\mathrm{\Omega }_L^1`$, and $`\mathrm{\Theta }_L`$ to be defined in a neighborhood of a smooth point which means that they all become locally free. $``$ then has to be identified with the conormal bundle. To prove that $`\alpha ^{}`$ is an isomorphism, we will construct an inverse. First note that, by the fact that $`L`$ is coisotropic, the morphism $`\beta :\mathrm{\Theta }_{M}^{}{}_{|L}{}^{}\mathrm{\Omega }_{M}^{1}{}_{|L}{}^{}`$ actually sends an element of $`\mathrm{\Theta }_L`$ to a form vanishing on all vectors tangent to $`L`$. So the restriction of $`\beta `$ to $`\mathrm{\Theta }_L`$ defines a morphism $`\beta ^{}:\mathrm{\Theta }_L`$. The situation is as follows:
A diagram chase shows that $`\beta ^{}`$ is injective. On the other hand, we have $`dim()=dim(\mathrm{\Theta }_L)`$, as $`L`$ is lagrangian. So $`\beta ^{}`$ is an isomorphism and the inverse of $`\alpha ^{}`$. ∎
###### Corollary 1.
The morphism $`J:\mathrm{\Omega }_L^1𝒞_L^1`$ can be extended to a morphism of DGA’s
$$J:(\mathrm{\Omega }_L^{},d,)(𝒞_L^{},\delta ,)$$
which is an isomorphism at smooth points of $`L`$.
###### Proof.
Set
$$J(\omega _1\mathrm{}\omega _p):=J(\omega _1)\mathrm{}J(\omega _p)$$
where $`\omega _i\mathrm{\Omega }_L^1`$. Then it is immediate that $`J`$ is an isomorphism on $`L_{reg}`$. To prove that $`Jd=\delta J`$, it suffices to check this in the lowest degrees, that is, we have to show that the diagram
commutes. This follows directly from $`\mathrm{\Omega }_L^0=𝒞_L^0=𝒪_L`$. ∎
In the last section, we use the following elementary fact.
###### Lemma 1.
The kernel of $`J`$ is the complex $`𝒯ors(\mathrm{\Omega }_L^{})`$ consisting of the torsion subsheaves of $`\mathrm{\Omega }_L^p`$.
###### Proof.
We have $`𝒯ors(\mathrm{\Omega }_L^{})𝒦er(J)`$ as $`𝒞_L^{}`$ is torsion free. On the other hand, the kernel is supported on the singular locus of $`L`$, so it must be a torsion sheaf, hence $`𝒦er(J)𝒯ors(\mathrm{\Omega }_L^{})`$. ∎
### Remark:
Although the definition of the modules $`𝒞_L^p`$ involves the ideal $``$, they are probably intrinsic. This is at least clear in some special cases as the following lemma shows.
###### Lemma 2.
Suppose $`L`$ to be Cohen-Macaulay and regular in codimension one. Then there is an isomorphism
$$(\mathrm{\Omega }_L^p)^{}\stackrel{}{}𝒞_L^p$$
where for an $`𝒪_L`$-module $``$, $`^{}`$ denotes $`om_{𝒪_L}(,𝒪_L)`$.
###### Proof.
We will make use of the following fact: Let $``$ be an $`𝒪_L`$-module of type $`𝒢^{}`$, then $``$ is reflexive, i.e. $`^{}=`$. The morphism $`h:(\mathrm{\Omega }_L^p)^{}𝒞_L^p`$ we are looking is obtained by dualizing twice the morphism $`J:\mathrm{\Omega }_L^p𝒞_L^p`$, this yields $`J^{}:(\mathrm{\Omega }_L^p)^{}(𝒞_L^p)^{}=𝒞_L^p`$ as $`𝒞_L^p`$ is of type $`om(,𝒪_L)`$. Clearly, $`h`$ is an isomorphism on the regular locus. We have an exact sequence
$$0𝒦(\mathrm{\Omega }_L^p)^{}\stackrel{h}{}𝒞_L^p𝒢0$$
where $`𝒦`$ and $`𝒢`$ are the kernel resp. cokernel sheaves of the map $`h`$. This sequence can be split
$$\begin{array}{c}0𝒦(\mathrm{\Omega }_L^p)^{}0\\ 0𝒞_L^p𝒢0\end{array}$$
with $`=m(h)`$. Applying $`om_{𝒪_L}(,𝒪_L)`$ yields
$$\begin{array}{c}0^{}((\mathrm{\Omega }_L^p)^{})^{}𝒦^{}\\ 0𝒢^{}(𝒞_L^p)^{}^{}xt^1(𝒢,𝒪_L)\end{array}$$
Now we use the lemma of Ischebeck (see \[Mat89\]): Given a local ring $`R`$, two $`R`$-modules $`M`$ and $`N`$ with $`k=dim(M)`$ and $`r=depth(N)`$, then for all $`p<rk`$, the modules $`Ext^p(M,N)`$ vanish. It follows that $`𝒦^{}=𝒢^{}=xt^1(𝒢,𝒪_L)=0`$, so we have $`((\mathrm{\Omega }_L^p)^{})^{}=(𝒞_L^p)^{}`$. Then obviously $`((\mathrm{\Omega }_L^1)^{})^{}=(𝒞_L^1)^{}`$ and by the argument above $`(\mathrm{\Omega }_L^1)^{}=𝒞_L^1`$ so the map $`h`$ is an isomorphism. ∎
## 3 Deformations
Recall that the space of infinitesimal embedded deformations of an analytic algebra $`R`$, given as $`R=S/I`$ where $`S`$ is the ring of convergent power series, is equal to the normal module of $`I`$ in $`S`$, i.e. $`Hom_R(I/I^2,R)`$. Dividing out trivial deformations gives the space $`T_R^1`$, defined by the sequence
On the other hand, the deformations of a manifold $`X`$ over $`Spec(𝕂[ϵ]/(ϵ^2))`$ are parameterized by $`H^1(X,\mathrm{\Theta }_X)`$. The cotangent complex is a tool to handle these two special cases in an integrated manner: infinitesimal deformations of an analytic space $`L`$ are in bijection with $`^1(𝕃_X)`$. It seems that the complex $`𝒞_L^{}`$ has to be seen as a first approximation to an equivalent for the cotangent complex in the lagrangian context. More precisely, the following holds:
###### Theorem 1.
The first three cohomology sheaves of $`𝒞_L^{}`$ are
* $`^0(𝒞_L^{})=𝕂_L`$.
* $`^1(𝒞_L^{})=𝒯_L^1`$.
* $`^2(𝒞_L^{})=𝒯_L^2`$. This symbol denotes the lagrangian obstructions, that is, $`𝒯_L^2`$ is the sheaf of obstructions to extend a lagrangian deformation to higher order regardless whether it can be extended as a flat deformation.
The proof of the following preliminary lemma can be found in \[Ban94\].
###### Lemma 3.
If $`H^1(M,𝕂)=0`$, then each diffeomorphism $`f:MM`$ satisfying $`f^{}\omega =\omega `$ is the time $`1`$ map of a flow $`\phi _t`$ of a hamiltonian vector field $`H_g`$ for some function $`g`$ on M.
###### Proof of the theorem.
$`^0(𝒞_L^{})`$ equals $`𝒦er(\delta :𝒪_L𝒞_L^1)`$. Take an element $`f`$ of $`𝒦er(\delta )`$. Then $`\{f,g\}`$ for all $`g`$. If $`f`$ is not a constant, then the ideal $`(,f)`$ is strictly larger than $``$, not the whole ring and still involutive. This is a contradiction to the fact that $`L`$ is lagrangian, which means that $``$ is maximal under all involutive ideals. So the kernel must be the constant sheaf.
To prove that $`^1(𝒞_L^{})=𝒯_L^1`$, two things have to be checked: As $`𝒞_L^1=𝒩_L`$, we must first identify the elements of $`𝒦er(\delta ^1:𝒞_L^1𝒞_L^2)`$ with the flat lagrangian deformations. Then we have to show that the image of $`\delta ^0:𝒪_L𝒞_L^1`$ are the trivial deformations. But this is easy, because for $`f𝒪_L`$, $`\delta (f)`$ acts as $`H_f`$, thus inducing a trivial deformation. Furthermore, by lemma 3, of all deformations coming from vector fields on $`M`$, only those induced by hamiltonian vector fields are trivial in the lagrangian sense. Now we choose an open set $`UL`$ and sections $`(f_1,\mathrm{},f_k)`$ generating $`(U)`$. Take an element $`\mathrm{\Phi }𝒦er(\delta ^1)`$, which means that
$$\varphi \left(\{g,h\}\right)\{g,\varphi (h)\}\{\varphi (g),h\}=0$$
for all $`f,g/^2`$. Then $`\mathrm{\Phi }`$ corresponds to the deformation given by
$$\stackrel{~}{}=(f_1+ϵ\varphi (f_1),\mathrm{},f_k+ϵ\varphi (f_k))$$
The ideal $`\stackrel{~}{}`$ is involutive iff for any two elements $`f+ϵ\varphi (f),g+ϵ\varphi (g)`$, we have $`\{f+ϵ\varphi (f),g+ϵ\varphi (g)\}\stackrel{~}{}`$, which is equivalent to
$$F:=\{f,g\}+ϵ\left(\{f,\varphi (g)\}+\{\varphi (f),g\}\right)\stackrel{~}{}$$
Consider $`G:=\{f,g\}+ϵ\varphi \left(\{f,g\}\right)`$, which is an element of $`\stackrel{~}{}`$, so the condition $`F\stackrel{~}{}`$ is equivalent to $`FG\stackrel{~}{}`$, that is
$$\{f,\varphi \left(g\right)\}+\{\varphi \left(f\right),g\}\varphi \left(\{f,g\}\right)$$
This means exactly that $`\varphi 𝒦er(\delta ^1)`$.
In order to interpret the second cohomology group, we define the bilinear mapping
$$\begin{array}{ccc}\hfill \stackrel{~}{ob}:𝒞_L^1\times 𝒞_L^1& & 𝒞_L^2\hfill \\ \hfill (\mathrm{\Phi },\mathrm{\Psi })& & \left(gh\{\mathrm{\Phi }(g),\mathrm{\Psi }(h)\}\right)\hfill \end{array}$$
In this way we get a quadratic form $`ob(\mathrm{\Phi }):=\stackrel{~}{ob}(\mathrm{\Phi },\mathrm{\Phi })`$. It can be immediately verified that this induces a map $`ob:^1(𝒞_L^{})^2(𝒞_L^{})`$. We will now prove the following: Given a lagrangian deformation $`\mathrm{\Phi }𝒯_L^1`$. Then there is a lift to second order defining an involutive ideal iff $`ob(\mathrm{\Phi })=0𝒯_L^2`$. The last condition is equivalent to the existence of $`\mathrm{\Psi }𝒯_L^1`$ with $`ob(\mathrm{\Phi })=\delta (\mathrm{\Psi })`$, i.e.
$$\{\mathrm{\Phi }(f),\mathrm{\Phi }(g)\}=\mathrm{\Psi }\left(\{f,g\}\right)\{f,\mathrm{\Psi }(g)\}\{\mathrm{\Psi }(f),g\}f,g$$
But this means that the following ideal is involutive.
$$J=(f_1+ϵ\mathrm{\Phi }(f_1)+ϵ^2\mathrm{\Psi }(f_1),\mathrm{},f_k+ϵ\mathrm{\Phi }(f_k)+ϵ^2\mathrm{\Psi }(f_k))$$
### Remark:
The fact that $`𝒯_L^2`$ is not the real obstruction space make precise what was meant by saying that complex $`𝒞_L^{}`$ is a first approximation of the object we are looking for: Hopefully, there is a modified version of this complex whose cohomology gives, in complete analogy with the cotangent complex, the spaces $`T^1`$ and $`T^2`$ for flat lagrangian deformations. On the other hand, it is perhaps not even necessary to impose flatness as the involutivity condition implies that the dimension cannot drop, see also \[Mat\].
###### Corollary 2.
There is an exact sequence
Furthermore, there are two special cases:
* Let $`L`$ be a contractible space. Then $`^1(𝒞_L^{})=H^0(L,𝒯_L^1)`$ and in fact: $`LT_L^1=H^0(L,𝒯_L^1)`$.
* Let $`L`$ be Stein and smooth. Then it follows that $`^1(𝒞_L^{})=H^1(L,𝕂_L)`$ and the space of global deformations is indeed $`LT_L^1=H^1(L,𝕂_L).`$
###### Proof.
The first fact is just the definition of the sheaf $`𝒯_L^1`$. In the second case, note that the space of embedded flat deformations is $`H^0(L,𝒩_L)`$, where $`𝒩_L`$ is the normal bundle of $`L`$ in $`M`$. As $`L`$ is smooth, this happens to be $`H^0(L,\mathrm{\Omega }_L^1)`$, so each infinitesimal flat deformation corresponds to globally defined one-form on $`L`$. It is closed iff the deformation is lagrangian and the subspace of exact one-forms are deformations induced by hamiltonian vector fields (isodrastic deformations, see \[Wei90\]), these are the trivial ones. $`L`$ is assumed to be a Stein manifold, in this case the first *de Rham*-cohomology group is exactly $`H^1(L,𝕂_L)`$. ∎
By analogy with the cotangent complex, the following generalization is probably true although we did not check the details.
###### Proposition 1.
The space of infinitesimal lagrangian deformations of a complex space $`L`$ which is a lagrangian subvariety of a symplectic manifold $`(M,\omega )`$ is given by
$$LT_L^1=^1(𝒞_L^{})$$
## 4 Finiteness of the cohomology
This section is devoted to the proof of the following result.
###### Theorem 2.
Let $`LM`$ be a lagrangian subvariety as above. Assume that the following condition is satisfied: Denote by $`edim(p)`$ the embedding dimension of a point $`pL`$, that is $`edim(p):=dim_𝕂(𝐦_p/𝐦_p^2)`$, where $`𝐦_p`$ is the maximal ideal in the local ring $`𝒪_{(L,p)}`$. Let $`S_k^L`$ be the following set
$$S_k^L:=\{pL|edim(p)=2nk\}L$$
for all $`k\{0,\mathrm{},n\}`$, then suppose that we have
$$dim(S_k^L)k$$
for all $`k`$. Under this condition (which will be called “condition P”), all $`^i(𝒞_L^{})`$ are constructible sheaves of $`𝕂`$-vector spaces with respect to the stratification given by the $`S_k^L`$.
Before going into the details of the proof, we would like to explain the meaning of the condition (2).
###### Lemma 4.
Let $`pS_k^LL`$ with $`k>0`$. Then the germ $`(L,p)`$ can be decomposed into a product
$$(L,p)=(L^{},p^{})\times (𝕂,0)$$
where $`(L^{},p^{})`$ is a germ of a lagrangian variety in the symplectic space $`𝕂^{2n2}`$. Furthermore, we have $`p^{}S_{k1}^L^{}`$.
###### Proof.
Let $`x_1,\mathrm{},x_{2n}`$ be coordinates of $`M`$ centered at $`p`$. Then the fact that $`edim(p)<2n`$ implies that there are coefficients $`\alpha _i𝒪_{L,p}`$ such that the following equation holds in $`𝒪_{L,p}`$
$$\underset{i=1}{\overset{2n}{}}\alpha _ix_i+h=0$$
where $`h`$ is an element of $`𝒪_{L,p}`$ vanishing at second order. So we have an element in the ideal describing $`(L,p)`$ whose derivative do not vanish. Then $`(L,p)`$ is fibred by the hamiltonian flow of this function. Explicitly, we can make an analytic change of coordinates, such that $`\alpha _1=1`$, $`\alpha _i=0`$ for all $`i>1`$ and $`h=0`$. Than the ideal of $`(L,p)`$ is of the form $`(x_1,f_1,\mathrm{},f_m)`$ for some functions $`f_i`$ which are independent of the variable $`x_{n+1}`$ (provided that we have chosen the symplectic form to be $`_{i=1}^ndx_idx_{n+i}`$). ∎
According to the lemma, the set of points of the variety $`L`$ can be divided into two classes, those with maximal embedding dimension (these are the “bad points”) and those (with $`edim(p)<2n`$) at which $`L`$ is decomposable. Condition P implies that the bad points are isolated. As usual, the proof of the theorem consists of two parts: First, we will show that the cohomology sheaves are locally constant on the strata $`S_k^L`$. This is an immediate consequence of the following lemma. Then it suffices to show that all stalks of $`^p(𝒞_L^{})`$ are finite-dimensional.
###### Lemma 5 (Propagation of Deformations).
Let
$$(L,0)(𝕂^{2n},0)$$
be a germ of a lagrangian subvariety which can be decomposed, i.e., there is a germ $`(L^{},0)`$ (which is lagrangian in $`(𝕂^{2n2},0)`$) such that $`(L,0)=(L^{},0)\times (𝕂,0)`$. Denote by $`\pi :LL^{}`$ the projection. Then there is a quasi-isomorphism of sheaf complexes
$$j:\pi ^1𝒞_L^{}^{}𝒞_L^{}$$
###### Proof.
The proof of lemma 4 shows that the ideals $`I`$ and $`I^{}`$ describing the two germs differ by exactly one element whose differential do not vanish at the origin. This implies that the conormal sheaves $``$ of $`L`$ and $`^{}`$ of $`L^{}`$ are related by the formula $`=\pi ^{}^{}𝒪_L`$. It follows that
$$𝒞_L^p=om_{𝒪_L}(\pi ^{}\stackrel{p}{}^{},𝒪_L)om_{𝒪_L}(\pi ^{}\stackrel{p1}{}^{},𝒪_L)$$
Now we have to describe the differential on $`𝒞_L^{}`$. We choose local Darboux coordinates $`(p_1,\mathrm{},p_n,q_1,\mathrm{},q_n)`$ on $`𝕂^{2n}`$ and $`(p_2,\mathrm{},p_n,q_2,\mathrm{},q_n)`$ on $`𝕂^{2n2}`$. Suppose that the two ideals are $`I=(f_1,\mathrm{},f_m,p_1)`$ and $`I^{}=(f_1,\mathrm{},f_m,p_1,q_1)`$ (if we consider $`L^{}`$ as embedded in $`𝕂^{2n}`$). Let $`\mathrm{\Phi }`$ be an element of
$$om_{𝒪_L}(\pi ^{}\stackrel{p}{}^{},𝒪_L)$$
Then it can be written as a power series in $`q_1`$ with coefficients in $`𝒞_L^{}^{}`$. A direct calculation shows that the differential on $`𝒞_L^{}`$ is
$$\begin{array}{ccccc}\delta :& 𝒞_L^p& & 𝒞_L^{p+1}& \\ & \underset{i=0}{\overset{\mathrm{}}{}}(\mathrm{\Phi },\mathrm{\Psi })q_1^i& & \underset{i=0}{\overset{\mathrm{}}{}}(\delta \mathrm{\Phi }_i,\delta \mathrm{\Psi }_i+(1)^{p+1}(i+1)\mathrm{\Phi }_{i+1})q_1^i& \end{array}$$
It is clear that the morphism $`j`$ must be the obvious inclusion
$$om_{𝒪_L^{}}(\stackrel{p}{}^{},𝒪_L^{})om_{𝒪_L}(\pi ^{}\stackrel{p}{}^{},𝒪_L)om_{𝒪_L}(\pi ^{}\stackrel{p1}{}^{},𝒪_L)$$
We will now show that the cokernel of this inclusion is acyclic. Then it follows immediately that $`j`$ induces an isomorphism on the cohomology. So let $`\mathrm{\Gamma }`$ be an element of $`𝒞oker(j)𝒦er(\delta )`$, that is,
$$\mathrm{\Gamma }=\underset{i=1}{\overset{\mathrm{}}{}}(\mathrm{\Phi }_i,\mathrm{\Psi }_i)q_1^i+(0,\mathrm{\Psi }_0)$$
where $`\delta \mathrm{\Phi }_i=0`$ and $`\delta \mathrm{\Psi }_i=(1)^p(i+1)\mathrm{\Phi }_{i+1}`$ for all $`i`$. But then $`\mathrm{\Gamma }`$ vanishes in the cohomology because it can be written as $`\mathrm{\Gamma }=\delta \mathrm{\Lambda }`$ with
$$\mathrm{\Lambda }:=\underset{i=1}{\overset{\mathrm{}}{}}(\frac{(1)^p\mathrm{\Psi }_{i1}}{i},0)q_1^i𝒞_L^{p1}$$
###### Corollary 3.
We have isomorphisms of sheaves
$$\pi ^1^i(𝒞_L^{}^{})^i(𝒞_L^{})$$
###### Proof.
This is obvious since $`\pi ^1`$ is an exact functor. ∎
Let $`pS_k^L`$ be a point at which $`L`$ is decomposable, i.e. $`k>0`$. By induction, we find a neighborhood $`UL`$ of $`p`$ such there is an analytic isomorphism $`h:U\stackrel{}{}Z\times B_ϵ(0)^k`$, where $`Z`$ is lagrangian in $`𝕂^{2(nk)}`$, $`B_ϵ(0):=\{z𝕂||z|<ϵ\}`$ and each $`qUS_l^L`$ corresponds via $`h`$ to a point $`(q^{},b)Z\times B(ϵ)^k`$ with $`q^{}S_{lk}^Z`$. In particular, the image of $`US_k^L`$ under $`h`$ is $`(\{pt\},B(ϵ)^k)`$, so by the last corollary, $`^p(𝒞_L^{})`$ is constant on $`US_k^L`$.
It remains to show that the stalks of the cohomology are finite-dimensional. Again by corollary 3, this is done once we have shown it for points with maximal embedding dimension. We will use a method developed in \[BG80\]. In this paper, the following situation is considered. Let $`f:XS`$ be a morphism of complex spaces (with $`dim(S)=1`$) and $`𝒦^{}`$ a certain sheaf complex on $`X`$. Then, under suitable conditions, the relative hypercohomology $`^if_{}𝒦^{}`$ are coherent sheaves of $`𝒪_S`$-modules. The proof of this theorems relies on a functional analytic argument of Kiehl and Verdier (see \[KV71\] or \[Dou74\]) which states, roughly speaking, that if the mapping induced on the complex of sections of $`𝒦^{}`$ by a small shrinking of the open set (over which the sections are taken) is a quasi-isomorphism, then the hypercohomology groups are finite dimensional vector spaces. We are going to use this result in the form of \[vS87\].
###### Lemma 6.
Let $`(L,p)(𝕂^{2n},0)`$ be a germ of a lagrangian variety satisfying condition P which is indecomposable at $`p`$. Then the stalk $`^i(𝒞_L)_p`$ is a finite-dimensional $`𝕂`$-vector space.
###### Proof.
Choose a representative $`V`$ for the the germ such that $`edim(q)=2n`$ iff $`q=p`$ for all points $`qV`$. We refer the reader to theorem 1 in \[vS87\]. We do not consider a relative situation here, so the map $`f:XS`$ in this theorem is replaced by $`V\{0\}`$ (Obviously, $`V`$ can be chosen such that this map is a standard representative of the germ $`(L,p)`$ in the sense of definition 1 in \[vS87\], i.e., $`V=LB_ϵ(p)`$). The complex of sheaves in the theorem is the complex $`𝒞_L^{}`$, which satisfies the first two properties ($`C_L^p`$ is $`𝒪_L`$-coherent and the differential is $`𝕂`$-linear). Our task is to verify the third axiom, that is, we have to find a vector field of class $`C^{\mathrm{}}`$ such that $`^p(𝒞_L^{})`$ is *transversally constant* (see definition 2 in \[vS87\]), this will be done in corollary 4. Now the proof of the theorem shows that there is a smaller neighborhood $`V_1`$ of $`p`$ such that $`\mathrm{\Gamma }(V,^p(𝒞_L^p))=\mathrm{\Gamma }(V_1,^p(𝒞_L^p))`$. This gives the result by using \[KV71\] in the same way as in \[vS87\] or \[BG80\]. ∎
###### Lemma 7.
Let $`qVS_k^L`$ with $`k>0`$. Then there is a $`C^{\mathrm{}}`$-vectorfield in a neighborhood $`W`$ of $`q`$ in $`M`$, tangent to $`VS_k^L`$ and transversal to $`B_ϵ(p)`$.
###### Proof.
It follows from lemma 4 that there exist $`k`$ linear independent hamiltonian vector fields on $`M`$ which respects the stratum $`S_L^k`$. Now we have to distinguish the cases $`𝕂=`$ and $`𝕂=`$, in the first one, since $`S_L^k`$ is of real dimension $`k`$ and since the intersection of $`L`$ and $`B_ϵ(p)`$ was transversal, it follows immediately that we can find a linear combination of theses $`C^{\mathrm{}}`$-fields which is transversal to $`B_ϵ(p)`$. The same is true in the complex case, here we have $`k`$ independent hamiltonian fields $`\eta _1,\mathrm{},\eta _k`$ which are *holomorphic*. As the holomorphic tangent space at each point is canonically isomorphic (over $``$) to the real one, we get $`2k`$ linear independent $`C^{\mathrm{}}`$-fields by applying this isomorphism to $`\eta _1,\mathrm{},\eta _k,i\eta _1,\mathrm{},i\eta _k`$. These can be used to find a field transversal to $`B_ϵ(p)`$. ∎
###### Corollary 4.
There is a $`C^{\mathrm{}}`$-vectorfield $`\vartheta `$ on a neighborhood $`U`$ of $`B_ϵ(p)`$ in $`M`$ such that $`^p(𝒞_L^{\mathrm{}})`$ is transversally constant with respect to $`U`$ and $`\vartheta `$.
###### Proof.
Set $`U:=(V\backslash \{p\})^{}`$. Then the last lemma yields a covering $`U_i`$ of $`U`$ and vector fields $`\vartheta _i`$ defined in a neighborhood of $`U_i`$ in $`M`$. Chose a partition of unity subordinate to this covering to obtain a field on $`U`$ which is still transversal to $`B_ϵ`$. For each point $`qU`$, which is contained in some stratum $`S_k^L`$, $`\vartheta `$ is necessarily tangent to $`S_k^L`$, so the cohomology sheaves are constant on the local integral curves of $`\vartheta `$. ∎
### Remark:
By the *Riemann-Hilbert-correspondence* (see \[Bj 93\]), the complex $`^{}:=(𝒞_L^{})`$, viewed as an object of $`𝒟_c^b(𝕂_M)`$ (the derived category of constructible sheaves of $`𝕂`$-vector spaces on $`M`$) corresponds via the *de Rham*-functor to a unique complex of coherent $`𝒟_M`$-modules with regular holonomic cohomology supported on $`L`$ (i.e., an object of $`𝒟_{\text{r.h.}}^b(\mu _L(𝒟_M))`$).
###### Lemma 8.
The complex $`^{}`$ satisfies the first perversity condition, that is, the following inequality holds.
$$\text{dim}(^i(𝒞_L^{}))ni$$
###### Proof.
Let $`pS_k^L`$. Then $`(L,p)=(L^{},p^{})\times (𝕂^k,0)`$ and $`^i(𝒞_L^{})_p=^i(𝒞_L^{}^{})_p^{}`$. But $`dim(L^{})nk`$, so $`^i(𝒞_L^{}^{})_p^{}=0`$ for all $`i>nk`$. ∎
In case that the second perversity condition is also satisfied, the $`^i`$’s are the de Rham-cohomology modules of some $`𝒟_M`$-module supported on $`L`$. The following consideration gives more evidence that the complex $`𝒞_L^{}`$ is closely related to $`𝒟`$-module theory: Every complex manifold is lagrangian in its own cotangent bundle. Consider *Spencer’s* complex, which is a resolution of $`𝒪_X`$ as a $`𝒟_X`$-module, explicitly:
$$Sp(𝒪_X)^{}:\mathrm{}𝒟_X_{𝒪_X}\mathrm{\Theta }_X^{p+1}𝒟_X_{𝒪_X}\mathrm{\Theta }_X^p\mathrm{}𝒟_X𝒪_X0$$
The *de Rham*-complex of $`𝒟_X`$-module $``$ is obtained as
$$DR(M):=om_{𝒟_X}(Sp(𝒪_X)^{},)$$
If we define a generalized version of the complex $`𝒞_L^{}`$ as
$$𝒞_L^p():=om_{𝒪_L}(\stackrel{p}{},)$$
for some module $``$ over the Lie algebroid $``$, then $`𝒞_X^p()`$ (for $`X`$ lagrangian in $`T^{}X`$) is exactly the *de Rham*-complex of the $`𝒟_X`$-Module $``$.
## 5 Examples and results
In this section we will describe some of the basic examples of singular lagrangian submanifolds, in particular those for which results on their deformation spaces are available. We start with the easiest case, a plane curve $`C`$ in $`𝕂^2`$, given as the zero set of a mapping $`f:𝕂^2𝕂`$. Such a curve $`C`$ is obviously lagrangian. In this case the complex $`𝒞_C^{}`$ is simplifies to
$$\begin{array}{ccc}\hfill 𝒞_C^0=𝒪_C& \stackrel{\delta }{}& 𝒞_C^1=om_{𝒪_C}(/^2,𝒪_C)=om_{𝒪_C}(𝒪_C,𝒪_C)=𝒪_C\hfill \\ \hfill h& & \{h,f\}\hfill \end{array}$$
It follows immediately that $`^2(𝒞_C^{})=0`$, while $`𝒯_C^1=^1(𝒞_C^{})=𝒞oker(\delta )`$. This sheaf is supported on the singular points of the curve, let $`x_0`$ be such a point. Then we have
$$𝒯_{C,x_0}^1=\frac{𝒪_{C,x_0}}{\{\{h,f\}|h𝒪_{C,x_0}\}}$$
Now the following equalities hold
$`{\displaystyle \frac{𝒪_{C,x_0}}{\left\{\{h,f\}\right|h𝒪_{C,x_0}\}}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Omega }_{𝕂^2,x_0}^2}{\left\{f\mathrm{\Omega }_{𝕂^2,x_0}^2+\{dfdh|h𝒪_{C,x_0}\}\right\}}}`$
$`=`$ $`{\displaystyle \frac{\mathrm{\Omega }_{𝕂^2,x_0}^2}{\left\{f\mathrm{\Omega }_{𝕂^2,x_0}^2+dfd\mathrm{\Omega }_{𝕂^2,x_0}^0\right\}}}`$
because $`𝒪_{C,x_0}\mathrm{\Omega }_{𝕂^2,x_0}^2/(f\mathrm{\Omega }_{𝕂^2,x_0}^2)`$ and the Poisson bracket of two functions $`f`$ and $`g`$ corresponds under the isomorphism $`𝒪_{𝕂^2,x_0}\mathrm{\Omega }_{𝕂^2,x_0}`$ to the $`2`$-form $`dfdg`$. But it is known (see \[Mal74\]) that the dimension of the last quotient equals $`\mu `$, the Milnor number of the plane curve singularity $`(C,x_0)`$. So the result is:
$$𝒯_C^1=\underset{x_0Sing(C)}{}𝕂^{\mu (C,x_0)}$$
This is remarkable because the usual $`T_C^1`$ has dimension $`\tau `$ (the Tjurina number) which is in general smaller than $`\mu `$. The difference corresponds to the space of deformations of the restriction of the symplectic structure to $`L`$ (see also \[Giv88\]).
Applying lemma 5, we see that the dimension of $`𝒯^1`$ for a surface singularity which is a curve germ, crossed with a smooth factor is also equal to the Milnor number of this curve. This result can also be obtained by a direct calculus, e.g., for a cuspidal edge given in four-space (with coordinates $`A,B,C,D`$ and symplectic form $`dAdC+dBdD`$) by the two equations $`A,B^2C^3`$, we get $`LT^1=𝕂^2`$ and $`LT^2=0`$.
We will proceed with further examples of lagrangian surfaces in $`𝕂^4`$, which satisfy condition P of theorem 2. So there are three strata: one point with embedding dimension four (supposed to be the origin), the singular locus away from this point and the regular locus. In order to simplify the calculation of the cohomology of $`𝒞^{}`$, we will suppose that our varieties are strongly quasi-homogeneous in the sense of \[CJNMM96\], that is, one can choose local coordinates of the ambient space around each point of $`L`$ such that the defining equations become weighted homogeneous with positive weights. In this case, the *de Rham*-complex is a resolution of the constant sheaf as one can see by considering the decomposition of the modules $`\mathrm{\Omega }_L^p`$ into eigenspaces of the Lie-derivative.
###### Lemma 9.
Let $`LM`$ be a strongly quasi-homogeneous lagrangian subvariety. Consider the map $`J:(\mathrm{\Omega }_L^{},d,)(𝒞_L^{},\delta ,)`$ of DGA’s from corollary 1. Denote by $`\stackrel{~}{\mathrm{\Omega }}_L^{}`$ the subcomplex $`m(J)`$ in $`𝒞_L^{}`$. Then $`\stackrel{~}{\mathrm{\Omega }}_L^{}`$ is a resolution of $`𝕂_L`$.
###### Proof.
By the long exact cohomology sequence, it suffices to prove that the complex $`𝒦er(J)`$ is acyclic. This can be done in exactly the same way as for $`\mathrm{\Omega }_L^{}`$ provided that the inner derivative $`i_E`$ ($`E`$ being the quasi-homogeneous Euler vector field) maps $`𝒦er(J)\mathrm{\Omega }_L^p`$ into $`𝒦er(J)\mathrm{\Omega }_L^{p1}`$. But this follows from lemma 1 because if $`\omega `$ is a torsion element than the same holds for $`i_E\omega `$. ∎
###### Corollary 5.
Denote by $`𝒢_L^{}`$ the cokernel of the map $`J`$. Then there is an exact sequence of $`𝒪_L`$-modules
$$0\stackrel{~}{\mathrm{\Omega }}_L^{}𝒞_L^{}𝒢_L^{}0$$
and the long associated long exact sequence gives
$$^i(𝒞_L^{})=^i(𝒢_L^{})$$
for all $`i0`$. In particular, if $`L`$ is of dimension two, then we get
$`^1(𝒞_L^{})`$ $`=`$ $`𝒦er(\delta :𝒢_L^1𝒢_L^2)`$
$`^2(𝒞_L^{})`$ $`=`$ $`𝒞oker(\delta :𝒢_L^1𝒢_L^2)`$
We can thus calculate $`𝒯_L^1`$ and $`𝒯_L^2`$ by computing the induced morphism $`\delta :𝒢_L^1𝒢_L^2`$. As $`J`$ is an isomorphism at smooth points, the sheaves $`𝒢_L^i`$ are supported on the singular locus of $`L`$, which is of dimension one. In a neighborhood of all of its regular points $`q`$ (points with embedding dimension three), the germ is decomposable and the dimension of $`^i(𝒞_L^{})_q`$ is given by lemma 5. So we are only interested in the one special point with maximal embedding dimension. We now choose an element $`p𝒪_L`$ which is finite when restricted to the support of $`𝒢_L^i`$, note that although this is set-theoretically equal to the singular locus of $`L`$, it may have embedded components. We will suppose that $`p`$ maps the origin in $`𝕂^4`$ to the origin in $`𝕂`$. Consider the sheaves $`p_{}𝒢_L^1`$ and $`p_{}𝒢_L^2`$, these are modules over $`𝒪_𝕂`$. Denote by $`\stackrel{~}{E}`$ resp. $`\stackrel{~}{F}`$ the modules of section of $`p_{}𝒢_L^1`$ resp. $`p_{}𝒢_L^2`$ in a small neighborhood of the origin. Then they can be decomposed into torsion and torsion free parts, the former being supported on the origin while the latter is free over $`𝕂\{t\}`$. In practice, this is done as follows: As $`𝒢_L^1`$ and $`𝒢_L^2`$ are graded modules over $`𝒪_L`$ and the map $`\delta :𝒢_L^1𝒢_L^2`$ is homogeneous, we consider the decomposition of these modules into homogeneous parts. The map $`p`$ is finite, so the torsion submodules of $`\stackrel{~}{E}`$ and $`\stackrel{~}{F}`$ corresponds to homogeneous parts of $`𝒢_L^1`$ and $`𝒢_L^2`$ in a finite number of degrees. This yields a decomposition of $`\stackrel{~}{E}`$ and $`\stackrel{~}{F}`$ into $`\stackrel{~}{E}=\widehat{E}E`$ and $`\stackrel{~}{F}=\widehat{F}F`$ such that $`\widehat{E}`$ and $`\widehat{F}`$ are supported on the origin, while $`E`$ and $`F`$ are free. $`\widehat{E}`$ and $`\widehat{F}`$ being artinian, the kernel and cokernel of $`\delta _{|\widehat{E}}`$ can be computed explicitly. The following lemma is used to do this for $`\delta _{|E}`$.
###### Lemma 10.
The rank of $`E`$ and $`F`$ is the Milnor number $`\mu `$ of the transversal curve singularity, i.e. the germ $`(L^{},0)`$ such that $`(L,p)`$ = $`(L^{},0)\times (𝕂,0)`$ for all $`p\text{Sing}(L)\backslash \mathrm{\hspace{0.17em}0}`$. Therefore, $`\delta _{|E}:EF`$ is an $`(E,F)`$-connection in the sense of \[Mal74\].
###### Proof.
This is an explicit calculation involving the definition of the complex $`𝒞_L^{}`$ and the map $`J:\mathrm{\Omega }_L^{}𝒞_L^{}`$. It suffices to calculate the rank of $`(𝒢_L^1)_p`$ and $`(𝒢_L^2)_p`$. So suppose that $`(L,p)`$ is a decomposable germ. We choose coordinates $`(x,y,s,t)𝕂^4`$ (with symplectic form $`\omega =dxdy+dsdt`$) around $`p`$ such that $`L`$ is given as the zero locus of $`s`$ and a function $`f`$ depending only on $`x`$ and $`y`$. Denote the ideal generated by these two functions by $`I`$ and by $`R`$ the stalk of $`𝒪_L`$ at the point $`p`$. Then we can identify $`I/I^2`$ with $`R^2`$, so $`Hom_R(I/I^2,R)`$ is free on the two generators $`n_1`$ and $`n_2`$, where
$$\begin{array}{cc}n_1(f)=1& n_1(s)=0\\ n_2(f)=0& n_2(s)=1\end{array}$$
while $`Hom_R(I/I^2I/I^2,R)`$ is just $`R`$, generated by the homomorphism sending $`fs`$ to $`1`$ in $`R`$. The complex $`𝒞^{}`$ at the point $`p`$ then reads:
$$\begin{array}{ccccc}R& & Rn_1Rn_2& & R\\ h& & (\{h,f\},\{h,s\})& & \\ & & (p,q)& & \{p,s\}+\{f,q\}\end{array}$$
where the pair $`(p,q)R^2=Hom_R(I/I^2,R)`$ denotes the homomorphism sending $`fI/I^2`$ to $`pR`$ and $`sI/I^2`$ to $`qR`$.
Now we have to investigate the modules of differential forms on $`L`$ at $`x`$. In general
$$\mathrm{\Omega }_R^p=\mathrm{\Omega }_S^p/(I\mathrm{\Omega }_S^p+dI\mathrm{\Omega }_S^{p1})$$
where $`S`$ is the ring $`𝕂\{x,y,s,t\}`$. This leads to
$`\mathrm{\Omega }_R^1`$ $`=`$ $`M_1M_2`$
$`\mathrm{\Omega }_R^2`$ $`=`$ $`M_3M_4`$
where we have used the following abbreviations:
$`M_1`$ $`=`$ $`{\displaystyle \frac{RdxRdy}{Rdf}}`$
$`M_2`$ $`=`$ $`Rdt`$
$`M_3`$ $`=`$ $`{\displaystyle \frac{Rdxdy}{RdfdxRdfdy}}`$
$`M_4`$ $`=`$ $`{\displaystyle \frac{RdxdtRdydt}{Rdfdt}}`$
$`J:\mathrm{\Omega }_L^{}𝒞_L^{}`$ can be described as
$`J:M_1`$ $``$ $`Rn_1Rn_2`$
$`dx`$ $``$ $`(\{x,f\},\{x,s\})=(_yf,0)`$
$`dy`$ $``$ $`(\{y,f\},\{y,s\})=(_xf,0)`$
$`J:M_2`$ $``$ $`Rn_1Rn_2`$
$`dt`$ $``$ $`(\{t,f\},\{t,s\})=(0,1)`$
$`J:M_3`$ $``$ $`R`$
$`dxdy`$ $``$ $`J(dx)J(dy)=0`$
$`J:M_4`$ $``$ $`R`$
$`dxdt`$ $``$ $`J(dx)J(dt)=_yf`$
$`dydt`$ $``$ $`J(dx)J(dt)=_xf`$
$`E`$ and $`F`$ are the cokernels of the maps $`J:M_1M_2Rn_1Rn_2`$ and $`J:M_3M_4R`$, respectively. So the result is
$$E=F=R/(_xf,_yf)=𝕂\{t\}𝒪_{L^{},p}/(_xf,_yf)=𝕂\{t\}^\tau $$
As $`L`$ is strongly quasi-homogeneous, we have weighted homogeneous local equations for the transversal slice which gives $`\tau =\mu `$. ∎
Denote $`\delta _{|E}`$ by $`D`$ for short. Then $`D`$ is a first-order differential operator $`D:𝒪_𝕂^\mu 𝒪_𝕂^\mu `$ which respects the grading. So it is of the form
$$D=t_t\mathrm{𝟏}+A$$
where $`A`$ is a constant $`\mu \times \mu `$-matrix. Thus, the second part of the cohomology of $`𝒞_L^{}`$ (i.e. kernel and cokernel of $`\delta `$) can be deduced from the solutions of the differential system given by $`D`$. All explicit calculations have been done using *Macaulay2*.
The first interesting example we are going to study is the so called “open swallowtail”. For details of its definition, see \[Giv82\] and \[Giv83\]. Consider the space of polynomials in one variable of degree $`d:=2k+1`$ with fixed leading coefficient and sum of roots equal to zero, that is, the space
$$\begin{array}{ccccc}\hfill 𝒫_{2k+1}& =& \left\{x^{2k+1}+A_2x^{2k1}+\mathrm{}+A_{2k+1}x^0\right\}& & 𝕂^{2k}\hfill \end{array}$$
which comes equipped with the following symplectic structure
$$\omega =\underset{i=2}{\overset{k+1}{}}\left(2k+1i\right)!\left(i2\right)!(1)^idA_idA_{2k+3i}$$
We will write $`\mathrm{\Sigma }_k`$ for the subspace consisting of those polynomials which have a root of multiplicity greater than $`k`$. This space is obviously of dimension $`k`$ and it can be shown that the form $`\omega `$ vanishes on its regular locus. So we have a lagrangian subvariety in the space $`𝒫_{2k+1}`$, which is called open swallowtail. To get a more concrete impression of how it looks like, we will describe the easiest examples. For $`k=1`$, $`\mathrm{\Sigma }_1𝒫_3`$ is just the ordinary cusp in the plane, this case has already been discussed above. For $`k=2`$, we obtain a surface in the four-dimensional space (see the conceptual figure 1)
$$𝒫_5=\left\{x^5+Ax^3+Bx^2+Cx+D\right|(A,B,C,D)𝕂^4\}$$
(the symplectic form is $`\omega =3dAdD+dCdB`$) consisting of those polynomials $`f`$ with a root of multiplicity at least three. Such a $`f`$ can be written as $`f=(xa)^3(x^2+3ax+b)`$, so there is a normalization of $`\mathrm{\Sigma }_2`$ given by
$`n:𝕂^2`$ $``$ $`𝒫_5=𝕂^4`$
$`(a,b)`$ $``$ $`(b6a^2,8a^33ab,3a^2b3a^4,a^3b)`$
Note that the singular locus of $`\mathrm{\Sigma }_2`$ is a again a cusp as well as the transversal curve singularity.
The space $`\mathrm{\Sigma }_2`$ is our main example, we will describe in some detail how to apply the general results in this case. Using elimination theory, we can calculate the defining equations of $`\mathrm{\Sigma }_2`$ in $`𝕂^4`$. It turns out that the swallowtail is a determinantal variety given by the minors of the matrix
$$\left(\begin{array}{cc}9D& 9B^232AC\\ 3C& 5AB+125D\\ 9B& 45A^2100C\end{array}\right)$$
The ideal which defines $`\mathrm{\Sigma }_2`$ is generated by the following three polynomials
$`f_1`$ $`=`$ $`27B^2C+96AC^245ABD+1125D^2,`$
$`f_2`$ $`=`$ $`81B^3288ABC+405A^2D900CD`$
$`f_3`$ $`=`$ $`45AB^2+135A^2C300C^2+1125BD`$
So $`\mathrm{\Sigma }_2`$ is not a complete intersection but nevertheless Cohen-Macaulay by the Hilbert-Burch theorem. We list the commutators $`\{f_i,f_j\}`$ (for $`1i<j3`$) with respect to the given set of generators (this is a direct proof that $`\mathrm{\Sigma }_2𝕂^4`$ is involutive):
$`\{f_1,f_2\}`$ $`=`$ $`576Af_1+81Bf_296Cf_3`$
$`\{f_1,f_3\}`$ $`=`$ $`15Af_212Bf_3`$
$`\{f_2,f_3\}`$ $`=`$ $`900f_1+18Af_3`$
$`\mathrm{\Sigma }_2`$ is quasi-homogeneous with the weights $`(2,3,4,5)`$ for the variables $`A`$, $`B`$, $`C`$, $`D`$, respectively. We can thus apply the machinery developed above to obtain that $`\left(𝒯_{\mathrm{\Sigma }_2}^1\right)_0=0`$, while $`\left(𝒯_{\mathrm{\Sigma }_2}^2\right)_0=𝕂`$. The operator $`D`$ is in this case
$$t_t\mathbf{\hspace{0.17em}1}+\left(\begin{array}{cccc}11/40& 245/2& 0& 0\\ 33/4000& 109/40& 0& 0\\ 0& 0& 49/15& 59/27\\ 0& 0& 51/100& 11/15\end{array}\right)$$
For $`𝕂=`$, the monodromy of the locally constant sheaf $`𝒯_{|Sing(\mathrm{\Sigma }_2)\backslash 0}^1`$ has the following eigenvalues
$$\frac{8}{10},\frac{13}{10},\frac{22}{10},\frac{27}{10}$$
The second large class of examples are the conormal spaces. Given any submanifold $`Y`$ of an $`n`$-dimensional manifold $`X`$, the total space of the conormal bundle $`T_Y^{}X`$ is always a lagrangian submanifold of $`T^{}X`$. More generally, if $`Y`$ is an analytic subspace, we can take the closure of the space of conormals to all smooth points of $`Y`$. The result (which is called conormal space of $`Y`$ in $`X`$) is still lagrangian, but may have singularities. This is an important class of lagrangian subvarieties, as the characteristic variety of a holonomic $`𝒟_X`$-module is always a finite union of conormal spaces. Obviously, these spaces are conical in the fibers of $`T^{}X`$. If $`X`$ is a plane curve in $`C𝕂^2`$, then the conormal space $`T_C^{}𝕂^2`$ will be a surface in $`𝕂^4`$. Here the results are as follows.
$$\begin{array}{cccc}\hfill \text{equation of }C& \mathrm{𝐋𝐓}^\mathrm{𝟏}\hfill & \mathrm{𝐋𝐓}^\mathrm{𝟐}\hfill & \text{eigenvalues (multiplicity, if }1\text{)}\hfill \\ & & & \\ \hfill y^2x^5& 0\hfill & 0\hfill & \frac{4}{5},\frac{16}{5}\hfill \\ & & & \\ \hfill y^3x^7& 0\hfill & 0\hfill & \frac{37}{7},\frac{61}{7},\frac{69}{7},\frac{85}{7},\frac{93}{7},\frac{117}{7}\hfill \\ & & & \\ \hfill y^5x^7& 0\hfill & 0\hfill & \frac{116}{7},\frac{132}{7},\frac{148}{7},\frac{164}{7},\hfill \\ & & & \\ \hfill y^3x^6& 𝕂\hfill & 𝕂\hfill & \frac{7}{2},\frac{10}{2}^{\left(2\right)},\frac{13}{2}\hfill \\ & & & \\ \hfill xy\left(x+y\right)\left(xy\right)\left(x2y\right)& 𝕂^2\hfill & 𝕂^2\hfill & \hfill \end{array}$$
In the last example, there is only an isolated singularity, so the modules $`𝒢_L^1`$ and $`𝒢_L^2`$ are artinien.
Finally, there is a third class of singular lagrangian subvarieties, these are *completely integrable hamiltonian systems*. Such a system is given in the $`2n`$-dimensional phase space by $`n`$ Poisson-commuting functions. The ideal formed by them then obviously satisfies the involutivity condition. If, additionally, the common zero set of these function is a complete intersection, then it will be lagrangian in our sense. The lagrangian deformation space of such a system is at least $`n`$-dimensional (addition of a constant is flat and the ideal stays involutive).
To get the equations of some interesting examples, we will proceed as follows. Choose coordinates $`(p_1,q_1,p_2,q_2)`$ of $`𝕂^4`$ and set $`z_1=p_1+iq_1`$ and $`z_2=p_2+iq_2`$ (This can obviously be done only in the real case, but it is a formal calculus which works as well for $`𝕂=`$ as for $`𝕂=`$). We can now express functions on $`𝕂^4`$ in the variables $`z_1,z_2,\overline{z_1},\overline{z_2}`$, and the Poisson bracket becomes
$$\{f,g\}=2i\left(_{\overline{z}_1}f_{z_1}g_{\overline{z}_1}g_{z_1}f+_{\overline{z}_2}f_{z_2}g_{\overline{z}_2}g_{z_2}f\right)$$
We want to find functions $`f_1,f_2`$ such that $`\{f_1,f_2\}=0`$. Set, for example $`f=\lambda z_1\overline{z_1}+\mu z_2\overline{z_2}`$ and let us look for a $`g=z_1^\alpha \overline{z_1}^\beta z_2^\gamma \overline{z_2}^\delta `$ for some parameters $`\lambda ,\mu ,\alpha ,\beta ,\gamma ,\delta `$. It can be easily verified that the commuting condition transforms to
$$\lambda (\alpha \beta )\mu (\gamma \delta )=0$$
The following table shows results for some resonance (r) coefficients $`\lambda ,\mu `$ and exponents (e) $`\alpha ,\beta ,\gamma ,\delta `$.
$$\begin{array}{ccccc}\hfill \text{r}& \text{e}& \mathrm{𝐋𝐓}^\mathrm{𝟏}\hfill & \mathrm{𝐋𝐓}^\mathrm{𝟐}\hfill & \text{eigenvalues (multiplicity)}\hfill \\ & & & & \\ \hfill 1,0& 0,0,1,1& 𝕂^2\hfill & 𝕂\hfill & 3^{\left(4\right)}\hfill \\ & & & & \\ \hfill 1,2& 0,2,1,0& 𝕂^3\hfill & 𝕂^2\hfill & \frac{2}{2}^{\left(2\right)},\frac{3}{2}^{\left(2\right)},\frac{4}{2}^{\left(2\right)},\frac{5}{2}^{\left(2\right)},\frac{6}{2}^{\left(2\right)}\hfill \\ & & & & \\ \hfill 1,3& 3,0,0,1& 𝕂^4\hfill & 𝕂^3\hfill & \frac{3}{3}^{\left(2\right)},\frac{5}{3}^{\left(2\right)},\frac{7}{3}^{\left(4\right)},\frac{9}{3}^{\left(4\right)},\frac{11}{3}^{\left(4\right)},\frac{13}{3}^{\left(2\right)},\frac{15}{3}^{\left(2\right)}\hfill \\ & & & & \\ \hfill 1,4& 4,0,0,1& 𝕂^5\hfill & 𝕂^4\hfill & \frac{4}{4}^{\left(2\right)},\frac{7}{4}^{\left(2\right)},\frac{9}{4}^{\left(2\right)},\frac{10}{4}^{\left(2\right)},\frac{12}{4}^{\left(2\right)},\frac{13}{4}^{\left(2\right)},\hfill \\ & & & & \frac{14}{4}^{\left(2\right)},\frac{15}{4}^{\left(2\right)},\frac{16}{4}^{\left(2\right)},\frac{17}{4}^{\left(2\right)},\frac{18}{4}^{\left(2\right)},\frac{19}{4}^{\left(2\right)},\hfill \\ & & & & \frac{20}{4}^{\left(2\right)},\frac{22}{4}^{\left(2\right)},\frac{23}{4}^{\left(2\right)},\frac{25}{4}^{\left(2\right)},\frac{28}{4}^{\left(2\right)}\hfill \end{array}$$
### Remark:
The eigenvalues in all examples have a symmetry property, which we cannot prove at this moment. These eigenvalues looks very similar to the spectrum of an isolated hypersurface singularity. One might speculate that there is a mixed Hodge structure related to this theory and that the eigenvalues share further properties with the spectrum, e.g. the semi-continuity under deformations.
FB 17, Mathematik,
Johannes-Gutenberg-Universität Mainz,
D-55099 Mainz, Germany
straten@mathematik.uni-mainz.de,
sevenhec@mathematik.uni-mainz.de
|
warning/0002/hep-ph0002142.html
|
ar5iv
|
text
|
# Anomalous quartic couplings in six-fermion processes at the Linear Collider
## 1 Introduction
The experimental measurement of the gauge self-couplings is an important test of the Standard Model (SM), that is still at the beginning at present colliders. The trilinear gauge couplings have recently become a subject of studies at LEP II and Tevatron (for recent results, see ), and will also be measured at the Linear Collider (LC) . The quartic couplings are very weakly constrained by present experimental data, through loop diagrams , while the observation of quartic gauge coupling effects at tree-level at LEP is rather difficult when photons are involved and is completely outside the reach of LEP when only massive gauge bosons are present.
The determination of quadrilinear gauge couplings is of particular interest in connection with the problem of the electroweak symmetry breaking. Indeed, if a Higgs boson lighter than 1 TeV does not exist, the tree-level amplitudes for the gauge boson scattering in the Standard Model are known to violate the unitarity limits at tree-level, indicating the existence of strong interactions. This scenario can be the result of different models of symmetry breaking and may be described in the general framework of effective lagrangians, where the electroweak interactions in the low-energy limit are represented by one general parameterization . Different values of the parameters correspond to different models. The effective lagrangian is organized as a power series in $`p/4\pi v`$, where $`p`$ is the scale of momenta of the phenomena under study and $`v246`$ GeV is the scale of symmetry breaking. The lowest-order terms of this expansion are model-independent. The higher order terms involve new gauge coupling structures that give rise to anomalous vertices and contain coefficients that are model-dependent.
It is worth noticing that the lowest-order non-trivial quartic couplings in the effective lagrangian (that are the $`O(p^4)`$ or dimension-four terms) involve only massive gauge bosons and thus their effects can be observed either in loop contributions, or at tree-level in processes with at least six fermions in the final state.
The measurement of the parameters describing the gauge self-couplings, that gives the opportunity of constraining some possible models of new physics, will be an important objective of the LC, where significant improvements will be possible with respect to present colliders for various reasons. In the first place, the c.m. energy is sufficiently high for the production of up to three gauge bosons in the final state, making it possible to study the quadrilinear couplings at tree-level. Moreover, since the deviations from the SM values of these couplings destroy the unitarity cancellations, their effect is enhanced at high energy. Finally, the special features of the LC, such as the high luminosity and the possibility of having polarized $`e^+`$ and $`e^{}`$ beams will provide a very high sensitivity to the anomalous couplings.
Several phenomenological studies have been performed on the possibility of constraining the anomalous quartic couplings at $`e^+e^{}`$ colliders in the approximation of real vector bosons in the final state, both for the couplings including photons and for those involving gauge bosons only . In the latter case, however, for more realistic predictions, as observed above, processes with at least six fermions in the final state have to be examined. This kind of processes also allows for the analysis of final-state distributions that are not accessible in the real approximation, and that can give useful information.
The objective of the present study is to analyse a class of six-fermion ($`6f`$) processes where the genuine quartic gauge couplings of dimension four are involved. The results that will be shown have been obtained by means of complete tree-level calculations in the framework of the chiral approach to electroweak interactions.
In Section 2 the theoretical framework and some technical details of the calculation are explained. In Section 3 the numerical results are presented and discussed, and Section 4 contains the conclusions.
## 2 Theoretical framework and calculation technique
The construction of the effective lagrangian for the electroweak interactions satisfying the requirement of $`SU(2)\times U(1)`$ gauge invariance may be found in the literature . It will be useful to mention here that in this approach the lagrangian can be written in the form
$$=_{YM}+_F+_S,$$
(1)
where $`_{YM}`$ is the Yang-Mills lagrangian,
$$_{YM}=\frac{1}{4}W_{\mu \nu }^iW^{i\mu \nu }\frac{1}{4}B_{\mu \nu }B^{\mu \nu },$$
(2)
$`_F`$ is the fermionic contribution and $`_S`$ is the scalar contribution. The standard couplings between gauge bosons are given by the Yang-Mills lagrangian. The scalar sector is represented by means of the unitary matrix
$$U=\mathrm{exp}\left(i\frac{\tau _i\pi _i}{v}\right),$$
(3)
where $`v=246`$ GeV and $`\pi _i`$ are the would-be Goldstone bosons. The gauge-invariant operators contributing to $`_S`$ can be constructed by taking the traces of the following building blocks:
$`T`$ $`=`$ $`U\tau _3U^{}`$ (4)
$`V_\mu `$ $`=`$ $`(D_\mu U)U^{}`$ (5)
$`W_{\mu \nu }`$ $`=`$ $`_\mu W_\nu _\nu W_\mu +ig[W_\mu ,W_\nu ],`$ (6)
where $`D_\mu U=_\mu U+\frac{ig}{2}\tau _iW_\mu ^iU\frac{ig^{}}{2}B_\mu U\tau _3`$. At leading order, assuming the custodial symmetry, the following contribution to $`_S`$ is found:
$$_0=\frac{v^2}{4}Tr\left((D_\mu U)^{}D^\mu U\right),$$
(7)
that gives the masses to the $`W^\pm `$ and $`Z`$ bosons.
At next-to-leading order, that corresponds to dimension four, new couplings appear, and the independent gauge-invariant structures are multiplied by coefficients that are model-dependent. At this order, trilinear and quadrilinear vertices are present. The operators that give rise to quadrilinear and not to trilinear vertices, with the condition of $`CP`$ conservation, are (using the standard notation):
$`_4`$ $`=`$ $`\alpha _4(Tr(V_\mu V_\nu ))^2`$
$`_5`$ $`=`$ $`\alpha _5(Tr(V_\mu V^\mu ))^2`$
$`_6`$ $`=`$ $`\alpha _6Tr(V_\mu V_\nu )Tr(TV^\mu )Tr(TV^\nu )`$
$`_7`$ $`=`$ $`\alpha _7Tr(V_\mu V^\mu )(Tr(TV_\nu ))^2`$
$`_{10}`$ $`=`$ $`\alpha _{10}{\displaystyle \frac{1}{2}}(Tr(TV_\mu )Tr(TV_\nu ))^2.`$ (8)
The custodial symmetry is respected only by the operators $`_4`$ and $`_5`$, and is violated by the other three operators, due to the presence of $`T`$. In the unitary gauge the above terms are given by:
$`_4`$ $`=`$ $`\alpha _4g^4({\displaystyle \frac{1}{2}}W_\mu ^+W^{+\mu }W_\nu ^{}W^\nu +{\displaystyle \frac{1}{2}}(W_\mu ^+W^\mu )^2`$ (9)
$`+`$ $`{\displaystyle \frac{1}{c_W^2}}W_\mu ^+Z^\mu W_\nu ^{}Z^\nu +{\displaystyle \frac{1}{4c_W^4}}(Z_\mu Z^\mu )^2)`$
$`_5`$ $`=`$ $`\alpha _5((W_\mu ^+W^\mu )^2+{\displaystyle \frac{1}{c_W^2}}W_\mu ^+W^\mu Z_\nu Z^\nu `$ (10)
$`+`$ $`{\displaystyle \frac{1}{4c_W^4}}(Z_\mu Z^\mu )^2)`$
$`_6`$ $`=`$ $`\alpha _6g^4\left({\displaystyle \frac{1}{c_W^2}}W_\mu ^+Z^\mu W_\nu ^{}Z^\nu +{\displaystyle \frac{1}{2c_W^4}}(Z_\mu Z^\mu )^2\right)`$ (11)
$`_7`$ $`=`$ $`\alpha _7g^4\left({\displaystyle \frac{1}{c_W^2}}W_\mu ^+W^\mu Z_\nu Z^\nu +{\displaystyle \frac{1}{2c_W^4}}(Z_\mu Z^\mu )^2\right)`$ (12)
$`_{10}`$ $`=`$ $`\alpha _{10}{\displaystyle \frac{g^4}{2c_W^4}}(Z_\mu Z^\mu )^2.`$ (13)
As can be seen, these operators give the anomalous contributions to the quadrilinear gauge couplings involving the $`W^\pm `$ and $`Z`$ vector bosons, to be added to the standard ones that are contained in the Yang-Mills lagrangian. In particular, it is easy to find, in the above formulas, anomalous contributions to the vertices $`4W`$ ($`_4`$ and $`_5`$), $`WWZZ`$ ($`_4`$, $`_5`$, $`_6`$ and $`_7`$) and $`4Z`$ (all the operators).
In this paper, the numerical results of a phenomenological study on $`6f`$ processes involving these anomalous couplings will be presented. These results have been obtained by means of a computer code already employed in other $`6f`$ analyses . In this code the scattering amplitudes are calculated by the automatic algorithm ALPHA , and the Monte Carlo integration procedure is the result of an adaptation of the four-fermion codes HIGGSPV and WWGENPV to the $`6f`$ calculations. For the present study, the lagrangian in eq. (1) with $`_S=_0+_4+_5+_6+_7+_{10}`$ has been implemented. The case in which the coefficients $`\alpha _i`$ are all equal to zero is considered as a reference model, that represents the limit of infinite Higgs mass in the SM at tree-level, and the deviations from such a model when the coefficients are non-vanishing have been studied.
The input parameters are $`G_\mu `$, $`M_W`$ and $`M_Z`$. The widths of the $`W`$ and $`Z^0`$ bosons and all the couplings are calculated at tree-level. All the fermions are massless. For the propagators of the gauge bosons, the “fixed-width” scheme has been adopted. The reliability of this approach in $`6f`$ calculations at energies of the order of the TeV has been discussed in detail in ref. .
The aim of the present work is to provide an analysis, that has never been done till now, of the impact of a complete electroweak $`6f`$ calculation on the study of anomalous quartic couplings. The numerical results contained in the following section are a first illustrative application of the new implementation of anomalous couplings in the code mentioned above, and the processes considered, although not giving a complete picture of the anomalous quartic coupling phenomenology, are relevant due to their sizeable cross-sections, as can be seen below.
## 3 Physical processes and results
The processes $`e^+e^{}2q+2q^{}\nu _e\overline{\nu }_e`$, with $`q=u,c`$ and $`q^{}=d,s`$, have been considered. This choice is motivated by the two objectives of having contributions from all possible quartic vertices involved in the anomalous terms under consideration, and of having no more than two neutrinos in the final state. The first point can be explained by considering the signature $`u\overline{u}d\overline{d}\nu _e\overline{\nu }_e`$: as can be seen in fig. 1, that shows the diagrams with a quadrilinear vertex, this signature includes contributions from the $`4W`$, $`WWZZ`$ and $`4Z`$ vertices.
From the point of view of the calculation, in order to obtain the correct sum over colours, with the version of ALPHA used here, that does not contain the colour degrees of freedom, it is necessary to apply a procedure analogous to the one already illustrated in ref. . This procedure consists of the combination, with proper weights, of the results from the different processes and is valid in the approximation of massless quarks and unit CKM matrix. The signatures $`u\overline{d}\overline{c}s\nu _e\overline{\nu }_e`$, $`u\overline{u}d\overline{d}\nu _e\overline{\nu }_e`$ and $`u\overline{u}s\overline{s}\nu _e\overline{\nu }_e`$ must be combined according to the following formula:
$$\sigma _{uudd}+\sigma _{udcs}+\sigma _{uuss}=N_c(\widehat{\sigma }_{uudd}+(2N_c1)(\widehat{\sigma }_{udcs}+\widehat{\sigma }_{uuss})),$$
(14)
where the results directly provided by ALPHA are denoted by $`\widehat{\sigma }`$, while the cross-sections including the colour degrees of freedom are denoted by $`\sigma `$. The interested reader is referred to for more details.
The signatures $`c\overline{s}\overline{u}d\nu _e\overline{\nu }_e`$, $`c\overline{c}s\overline{s}\nu _e\overline{\nu }_e`$ and $`c\overline{c}d\overline{d}\nu _e\overline{\nu }_e`$ are obtained from the previous ones by means of the simultaneous exchanges $`uc`$ and $`ds`$, and, given the set of parameters adopted here, they give exactly the same result, so that they can be easily taken into account by including a factor of 2 in eq. (14).
In principle, assuming that the $`b`$-tagging technique is applied, and thus the $`b`$ quark can be identified, a realistic calculation should take into account all the possible combinations of the remaining flavours, $`u`$, $`d`$, $`c`$ and $`s`$ that cannot be distinguished. These combinations include, in addition to those mentioned above, the final states $`u\overline{u}c\overline{c}\nu _e\overline{\nu }_e`$, $`u\overline{u}u\overline{u}\nu _e\overline{\nu }_e`$ $`d\overline{d}s\overline{s}\nu _e\overline{\nu }_e`$ $`d\overline{d}d\overline{d}\nu _e\overline{\nu }_e`$ and those obtained from these with the exchanges $`uc`$ and $`ds`$. Moreover a sum over the neutrino flavours should be made. Such signatures have been neglected in this first analysis. Nevertheless, it will be possible to examine the most important qualitative features of the phenomenology of anomalous gauge couplings in $`6f`$ processes, while referring to further developments for a better accuracy from a quantitative point of view.
In the following, the results of complete electroweak calculations at tree level are presented, where the above specified class of processes is considered.
As can be seen in the scattering of real gauge bosons, as a consequence of the absence of the Higgs boson and of the gauge cancellations that ensure unitarity, the cross-sections at high energy can be expected to increase rapidly and to violate the unitarity bounds. The energy at which unitarity is violated depends on the values of the parameters $`\alpha _i`$: these should then be taken within some bounds if the unitarity condition has to be respected at a given energy. However, it is useful to make a first rough analysis of the dependence of the cross-sections on the anomalous parameters in a wide range, while the sensitivity to the anomalous couplings in a smaller range will be studied in a second step. Thus, in the first set of results, the cross-section has been evaluated at the two c.m. energies of 500 GeV and 1 TeV, and allowing each of the coefficients $`\alpha _i`$ to vary in the range $`(0.2,0.2)`$, while keeping all the others equal to zero. A simple set of kinematical constraints has been adopted, by requiring the invariant masses of the “up-anti-up” and “down-anti-down” quark pairs to be greater than 70 GeV (where “up” stands for $`u`$ and $`c`$ and “down” for $`d`$ and $`s`$), so as to eliminate the soft photon contributions to pair production.
In the plots of fig. 2 the cross-section is shown as a function of the various parameters $`\alpha _i`$ at the two energies of 500 GeV and 1 TeV. First of all, it should be observed that the effects of the anomalous couplings are greater at 1 TeV than at 500 GeV, as expected, according to the above arguments on unitarity violation. As a consequence, the cross-section turns out to have a very strong growth with energy for large values of the anomalous coefficients. The growth of the cross-section with energy is in agreement with results for the scattering of real longitudinal gauge bosons obtained by means of the equivalence theorem . Moreover it can be observed that the greatest effects are given by the coefficients $`\alpha _4`$ and $`\alpha _5`$, while the coefficient $`\alpha _{10}`$ gives the smallest effect. To understand this fact it is useful to examine the expressions of the operators in eqs. (913), where it can be seen that $`_4`$ and $`_5`$ are the only terms where the vertex $`4W`$ appears. This vertex gives a greater enhancement with respect to the others due to the fact that the couplings of the $`W`$ boson to the fermions are stronger than those of the $`Z`$ boson. For the same reason, the operator $`_{10}`$ has the smallest effect, since it contributes only to the $`4Z`$ vertex. Moreover the $`4Z`$ vertex, as can be seen in fig. 1, can appear only in $`s`$-channel diagrams, and has not the advantage of the $`t`$-channel growth at high energy.
The parameters $`\alpha _6`$, $`\alpha _7`$ and $`\alpha _{10}`$, that induce violation of the custodial symmetry, are more constrained by the radiative corrections to the $`\rho `$ parameter with respect to $`\alpha _4`$ and $`\alpha _5`$. For this reason, the latter are more interesting and the remaining part of the study will be restricted to them.
The behaviour of the cross-section as a function of the c.m. energy is considered in fig. 3, where the energy range between 500 GeV and 1 TeV is studied for the case without anomalous couplings and for two choices of the anomalous parameters taken as examples, $`\alpha _4=0.05`$ and $`\alpha _5=0.05`$. The invariant masses of all the possible quark pairs are required to be above 60 GeV. The solid curves include initial-state radiation (ISR) and beamstrahlung (BS), while the dashed ones are in the Born approximation. The growth with energy is already present when the parameters $`\alpha _i`$ are equal to zero, and is enhanced when they are different from zero. The effect of ISR and BS is to reduce the effective c.m. energy, and this explains the observed lowering of $`1520\%`$ in the cross-sections with respect to the Born approximation.
In order to study the sensitivity of some differential distributions to the anomalous couplings, two samples of events have then been generated, one with all the $`\alpha _i`$ equal to zero and the other with $`\alpha _4=0.1`$. The c.m. energy is 1 TeV and the numbers of events correspond to a luminosity of 1000 fb<sup>-1</sup>. As in fig. 3, all the quark pairs have an invariant mass greater than 60 GeV. A relatively large value of $`\alpha _4`$ has been chosen, so as to have sizeable effects: indeed, the objective of this analysis is to obtain clear indications on the cuts to be applied for enhancing the sensitivity to the anomalous couplings under study. Among the variables considered, the ones that turn out to be most sensitive to the parameter $`\alpha _4`$ are defined in the following way. The invariant mass of the system of four jets is indicated as $`M(WW)`$. This variable does not require any identification procedure. The other variables that have been studied are based instead on simple identification algorithms, to take into account the fact that the quark flavours are not distinguishable. The $`W`$ bosons are reconstructed as the pairs of quarks $`q_iq_j`$ and $`q_kq_l`$ such that the quantity $`|m(q_iq_j)^2M_W^2|+|m(q_kq_l)^2M_W^2|`$ is minimized. The angle of one reconstructed $`W`$ boson with respect to the beam axis is indicated as $`\theta (W)`$. For each event, the four jets are ordered in transverse momentum and are labelled with $`j_1,\mathrm{},j_4`$, where $`j_1`$ is the one with greatest $`p_t`$. The invariant masses of pairs of jets are then considered, and one in particular, $`M(j_3j_4)`$, that is the invariant mass of the pair of jets with lowest $`p_t`$, is found to be most sensitive to the anomalous couplings under study. In fig. 4 the variables defined above are shown for the sample with $`\alpha _4=0.1`$ (solid histograms) in comparison with the sample where all the anomalous parameters are set to zero (dashed histograms). In these plots, where the different total numbers of events for the two samples, due to the differences of the cross-sections, must be taken into account, significant deviations are found in the shapes of the distributions. This is due to the Lorentz structures of the anomalous vertices involved, that tend to populate the region of phase-space with high transverse momenta for the gauge bosons.
In view of a realistic simulation, these distributions include the effects of ISR and BS. It has been verified that these effects do not introduce significant variations with respect to the Born approximation, as can be expected, since the variables considered do not involve the momenta of the neutrinos.
On the contrary, the distribution of the missing mass, defined as $`M_{miss}=\sqrt{P_{miss}^2}`$, where $`P_{miss}`$ is the missing momentum, is strongly affected by ISR and BS. This variable, that in the Born approximation is the invariant mass of the neutrino pair, is shown in fig. 5: in the upper plot the case with $`\alpha _i=0`$ is considered, while the lower plot refers to $`\alpha _4=0.1`$. The solid histograms include ISR and BS, the dashed histograms are in the Born approximation. In both plots the peaks corresponding to the $`Z`$ mass are almost cancelled by the effects of ISR and BS.
Results very similar to those illustrated above are obtained for the coupling $`\alpha _5`$. On the basis of this analysis of distributions, suitable cuts have then been determined to enhance the sensitivity to the anomalous couplings $`\alpha _4`$ and $`\alpha _5`$. The results are given in fig. 6, where the cross-section in the Born approximation is shown as a function of the parameters $`\alpha _4`$ (upper plot) and $`\alpha _5`$ (lower plot) in the range $`(0.01,0.01)`$.
The cuts are defined as follows: $`M(WW)`$ is greater than 420 GeV, the invariant mass of the two jets with the lowest transverse momenta, $`m(j_3j_4)`$, is greater than 80 GeV and the angle of one reconstructed $`W`$ boson with respect to the beam axis satisfies $`\mathrm{cos}\theta (W)<0.7`$. Moreover, the invariant mass of all the quark pairs is greater than 60 GeV. It has been found that neither the addition of a cut on the variable $`p_t(W)`$ nor its substitution to one of the above cuts can improve the results obtained. In the same plots, the $`1\sigma `$ limits around the value of the cross-section for $`\alpha _i=0`$, with an experimental error corresponding to a luminosity of 1000 fb<sup>-1</sup>, are shown. It can be seen that the sensitivity to the parameters $`\alpha _4`$ and $`\alpha _5`$ turns out to be in the range between $`10^3`$ and $`10^2`$. These conclusions are not modified by the introduction of ISR and BS, since the variables used in the cuts are not sensitive to these effects.
## 4 Conclusions
The analysis of anomalous dimension-four quartic gauge couplings involves processes with at least six fermions in the final state. A set of processes of this kind has been considered in this work. By using a Monte Carlo event generator that has already been employed for other phenomenological studies on $`6f`$ physics, complete tree-level results have been obtained, in the context of the electroweak chiral lagrangian formalism. The five dimension-four operators giving genuine quartic anomalous couplings have been considered, and a special attention has been devoted to the two custodial-symmetry conserving ones, usually indicated by $`_4`$ and $`_5`$ and involving the parameters $`\alpha _4`$ and $`\alpha _5`$. After studying the energy dependence of the cross-section and the effects of initial-state-radiation and beamstrahlung, a set of kinematical cuts has been considered in order to enhance the signals of the anomalous couplings. A sensitivity to the parameters $`\alpha _4`$ and $`\alpha _5`$ in the range between $`10^3`$ and $`10^2`$ has been found at 1 TeV.
The results discussed above can be seen as an example of the possibilities of study of these phenomena through a complete tree-level simulation of $`6f`$ processes. In order to make the calculations quantitatively more accurate, some simplifications that have been adopted here should be eliminated: namely the class of signatures to be taken into account should include a full sum over the indistinguishable neutrino and quark flavour combinations in the final state, and the effect of QCD backgrounds should be considered. In particular the latter improvement can be achieved by means of the last version of ALPHA , that includes the QCD lagrangian. Moreover, the role of polarization of the initial electrons and positrons could be investigated. These points will be the the subject of future developments.
Acknowledgements
The author would like to thank G. Montagna, M. Moretti, O. Nicrosini and F. Piccinini for stimulating discussions and useful comments on the manuscript. The author also thanks the INFN, Sezione di Pavia, for the use of computing facilities.
|
warning/0002/quant-ph0002035.html
|
ar5iv
|
text
|
# Controlling Decoherence in Bose-Einstein Condensation by Light Scattering far off Resonance
Recently, some important developments in Bose Einstein condensates (BEC) of trapped atomic vapors have been achieved in connection with light scattering of BEC, such as the Bragg scattering and Rayleigh scattering in various cases where the atoms interact only with far off-resonant optical fields. The super-radiant Rayleigh scattering by the BEC was observed in a recent experiment by Ketterle and his co-workers at MIT. It was also demonstrated that, when the far off-resonant laser light is scattered into the vacuum modes of the electro-magnetic field, the dominant two-photon interaction can optically manipulate matter-wave coherence properties to generate a quai-CW atomic laser and the so called atomic four-wave mixing from BEC . In this paper we will probe into the problem whether the far off-resonant pump light scattering from a classical pumped mode to many-vacuum modes can generate an ideal entanglement between atomic and optical fields and then demolish the quantum coherence of the scattered condensate. Especially we will consider how to use the pump laser to control the quantum decoherence process of BEC.
Usually, an interaction between a quantum system and an environment can cause two types of quantum irreversible effects: quantum dissipation indicating the loss of the system energy, and quantum decoherence indicating the leaking out of coherent information while the system energy is conserved. In the study of irreversible process in the trapped BEC, progress has been made by adopting particular forms of coupling between BEC atoms and the environment,or in other words,by modeling the decoherence. For BEC system, the natural origin of environment coupling is the interaction between BEC atoms and the background electromagnetic field or the thermal atomic cloud. It is noticed that this kind of interaction can also change the energy of BEC and thus can also lead to a quantum dissipation. So a pure decoherence without dissipation can not be modeled directly based on the usual electromagnetic coupling. To detect the quantum decoherence phenomenon clearly in a practical experiment, it is essential to find a really-physical environment to produce quantum decoherence purely without quantum dissipation. Fortunately, the above mentioned far off-resonant light scattering on BEC can just induce such an environment coupling to BEC system purely with decoherence. Since the effective coupling coefficients are determined mainly by the Rabi frequency of the pump laser, the quantum decoherence process can be controlled by adjusting the intensity of the pump laser. To study the physical effect of decoherence in the BEC, we consider how the coherent tunneling of BEC in a well-separated tight double wall is suppressed by the effectively -entangled vacuum modes.
Following Moore and Meystre et al , we consider the Schreodinger fields of two-level atoms coupled to a classical pump laser field far-off resonant as well as to the quantized vacuum modes of the electromagnetic field, via the electric-dipole interaction. For the large detuning between the pump frequency and the atomic transition frequency, the excited state field can be eliminated adiabatically and then all atoms can be described by a scalar field $`\widehat{\mathrm{\Psi }}_g(𝐫)`$ of the ground state. The corresponding effective Hamiltonian is given by
$`\widehat{}`$ $`=`$ $`{\displaystyle d^3𝐫\widehat{\mathrm{\Psi }}_g^{}(𝐫)H_A\widehat{\mathrm{\Psi }}_g(𝐫)}+{\displaystyle \underset{k}{}}\mathrm{}\omega _ka_k^{}a_k+H_{AA}`$ (2)
$`+({\displaystyle \underset{k}{}}\mathrm{}g_k{\displaystyle }d^3𝐫\widehat{\mathrm{\Psi }}_g^{}(𝐫)\widehat{\mathrm{\Psi }}_g(𝐫)a_ke^{i(𝐤𝐤_0)𝐫}+H.c),`$
where $`H_{AA}=\mathrm{}\xi d^3𝐫[\widehat{\mathrm{\Psi }}_g^{}\widehat{\mathrm{\Psi }}_g^{}\widehat{\mathrm{\Psi }}_g\widehat{\mathrm{\Psi }}_g]`$ denotes two -body interatomic interaction measured by the scattering length $`a`$ of $`swave`$ and $`\xi =\frac{4\pi \mathrm{}a}{M}`$; $`H_A`$ =$`\frac{\mathrm{}^2}{2M}^2+V(𝐫)`$ is the free Hamiltonian for single atom of mass $`M`$ in an effective trap $`V(𝐫)=V_g(𝐫)+\frac{\mathrm{}R^2}{2\mathrm{\Delta }}`$; $`a_k^{}`$ and $`a_k`$ are the creation and annihilation operators for the vacuum electromagnetic field with frequency $`\omega _k=ck\omega _0`$ for $`k=|𝐤|`$, $`c`$ being the speed of light. The effective trap $`V(𝐫)`$ was obtained from the trap potential $`V_g(𝐫)`$ for the ground state with modification by the coupling strength $`R`$ (Rabi frequency ) between the atoms under consideration and the classical pump filed of frequency $`\omega _0=ck_0`$, and $`R(\sqrt{I})`$depends on the pump intensity $`I`$. With respect to the atomic transition frequency $`\omega _a`$, the detuning $`\mathrm{\Delta }`$ =$`\omega _0\omega _a`$ is very large. Particularly, it is noticed that the coupling coefficient $`g_k\frac{|R|\sqrt{k}}{2|\mathrm{\Delta }|}\frac{\sqrt{Ik}}{2|\mathrm{\Delta }|}`$ . So, in principle, one can control the quantum dynamic processes of BEC resulting from the coupling to the vacuum, such as the quantum decoherence.
When the BEC happens in a single trap $`V_g(𝐫)`$, the atomic field $`\widehat{\mathrm{\Psi }}_g(𝐫)`$ can be approximately quantized as $`\widehat{\mathrm{\Psi }}_g(𝐫)b_0\varphi _0`$ where $`\varphi _{0\text{ }}`$is the wave function of the ground state of energy $`E_0=\mathrm{}\mathrm{\Omega }`$ and $`b_0`$ is its annihilation operator obeying $`[b_0,b_0^{}]=1`$. Then, the effective Hamiltonian takes the form
$`\widehat{_e}`$ $`=`$ $`\mathrm{}\mathrm{\Omega }b_0^{}b_0+\mathrm{}\kappa b_0^2b_0^2+`$ (4)
$`{\displaystyle \underset{𝐤}{}}\mathrm{}\omega _𝐤a_𝐤^{}a_𝐤+b_0^{}b_0({\displaystyle \underset{𝐤}{}}\mathrm{}\eta _𝐤a_𝐤+H.c),`$
where $`\eta _𝐤=\mathrm{}g_kd^3𝐫\varphi _{0\text{ }}^{}(𝐫)\varphi _{0\text{ }}(𝐫)e^{i(𝐤𝐤_0)𝐫}`$ and $`\kappa =\xi d^3𝐫|\varphi _{0\text{ }}(𝐫)|^4`$.The above Hamiltonian describes a typical quantum decoherence process since the effective interaction $`H_I=b_0^{}b_0[_𝐤\mathrm{}\eta _𝐤a_𝐤+H.c]`$ commutes with the free BEC Hamiltonian $`H_0=\mathrm{}\mathrm{\Omega }b_0^{}b_0+\mathrm{}\kappa b_0^2b_0^2`$ , i.e., $`[H_I,H_0]=0`$. However, $`H_I`$ can still induce a virtual transition, in which the atomic internal state remains unchanged. But it should be noticed that since $`[H_I,H_L]0`$, this transition may result in a back-action on the vacuum light field of the free Hamiltonian $`H_L=_𝐤\mathrm{}\omega _𝐤a_𝐤^{}a_𝐤`$ .
For example, when an atom in the Fock state $`|n=\frac{1}{\sqrt{n!}}b_0^n|n`$ absorbs a pump photon and then emits a pump photon, the vacuum light field experiences different recoil kicks described by $`H_n=n(_𝐤\mathrm{}\eta _𝐤a_𝐤+H.c)`$ where $`n`$ is the atom number. After time $`t`$ , this external forced term will drive the vacuum light field to evolve into a product state $`|v_n=\mathrm{\Pi }_𝐤e^{i\gamma _{nk}(t)}|\alpha _𝐤^n`$ of coherent states $`|\alpha _𝐤^n(t)=\mathrm{exp}[\alpha _𝐤^n(t)a_𝐤^{}H.c]|0`$ where
$`\alpha _𝐤^n(t)`$ $`=`$ $`{\displaystyle \frac{ng_𝐤}{i\omega _𝐤}}(e^{i\omega _𝐤t}1])`$ (5)
$`\gamma _{n𝐤}`$ $`=`$ $`{\displaystyle \frac{n^2|g_𝐤|^2}{\omega _𝐤^2}}[\omega _𝐤t\mathrm{sin}(\omega _𝐤t)]`$ (6)
Therefore, the total system formed by the BEC plus the vacuum field can evolve from a factorized initial state $`|\mathrm{\Psi }(0)=_{n=0}c_n|n|0`$ to an entangled state
$$|\mathrm{\Psi }(t)=\underset{n=0}{}c_ne^{iϵ(n)t}|n|v_n$$
(7)
where $`ϵ(n)=n(\mathrm{\Omega }_0\kappa )+\kappa n^2`$ is the Hartree-Fock energy of the BEC. The overlaps $`O_{mn}=v_m|v_n`$, which are called decoherence factors, can be calculated directly as follows
$`O_{mn}`$ $`=`$ $`{\displaystyle \underset{𝐤}{}}\alpha _𝐤^m|\alpha _𝐤^n`$ (8)
$`=`$ $`\mathrm{exp}[{\displaystyle \underset{𝐤}{}}(mn)^2{\displaystyle \frac{g_𝐤^2}{\omega _𝐤^2}}\mathrm{sin}^2({\displaystyle \frac{\omega _𝐤t}{2}})]\times `$ (10)
$`\mathrm{exp}[{\displaystyle \underset{𝐤}{}}i(m^2n^2){\displaystyle \frac{g_𝐤^2}{\omega _𝐤^2}}(\omega _𝐤t\mathrm{sin}(\omega _𝐤t))]`$
It characterizes the extent of entanglement and has a factorized structure with respect to the individual mode $`𝐤`$ of the ”environment”. All necessary information about the decoherence effects of the vacuum environment on the BEC are contained in the decoherence factor. The zero overlap means an ideal entanglement and thus leads to a complete decoherence. In fact, if $`v_m|v_n=0`$ for $`mn,`$the reduced density $`\rho `$ matrix of atoms have the vanishing off-diagonal elements $`\rho _{mn}=e^{i[ϵ(m)ϵ(n)]t}v_m|v_n0`$.
Using the explicit expression of $`g_𝐤`$ given in ref., we replace the sum $`_𝐤`$ $`(\mathrm{})`$ in Eq.(6) with an integral $``$ $`\mu (k)(\mathrm{})k^2\mathrm{sin}\theta d\theta d\varphi dk`$ where$`\theta `$ and $`\varphi `$ are the polar variables of the wave vector $`𝐤`$. For the isotropic spectral distribution (effective mode density) of the vacuum field, we can explicitly calculate the norm of the decoherence factors:
$`|O_{mn}(t)|`$ $`=`$ $`\mathrm{exp}[(mn)^2{\displaystyle \frac{\pi R^2cd^2}{8(2\pi )^3\mathrm{}\mathrm{\Delta }^2}}\times `$ (12)
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\mathrm{sin}^2(\frac{t}{2}[ck\omega _0])k^3}{(ck\omega _0)^2}}\mu (k)dk]\times `$
In the free space with $`\mu (k)=1`$, the integral in the above Eq.(6) diverges to positive infinity and thus $`O_{mn}(t)`$ approaches zero. This leads to the instantaneous quantum decoherence of the BEC. However, if we put the system into a micro-cavity, the effective mode density $`\mu (k)`$ will change singularly. In this case $`\mu (k)`$ may take certain special forms so that the integral does not diverge to positive infinity. For instance, in a special case where $`\mu (k)=\frac{\xi }{k^3},`$
$$|O_{mn}|=e^{\lambda _{mn}[\pi t2+2\omega _0^1\mathrm{cos}(\omega _0t)+2Si\left(\omega _0t\right)t]}$$
(13)
where $`\lambda _{mn}=\frac{\xi (mn)^2R^2d^2}{256\pi ^2\mathrm{}\mathrm{\Delta }^2}`$ and $`Si(z)=_0^z`$ $`\frac{\mathrm{sin}x}{x}dx`$ is a special function. Figure.1(a-b) illustrates the decoherence processes for different $`\lambda _{mn}`$ and $`\omega _0.`$What is of high interest is the fact that the damping rate of quantum coherence described by $`\rho _{mn}`$ $`O_{mn}`$ depends on both the frequency $`\omega _0`$ of the pump light field and its coupling strength $`R`$ to the BEC. Thus there arises a possibility for one to control the happening of the quantum decoherence.
On the other hand, the leading damping term $`\lambda _{mn}\pi `$=$`\frac{\xi (mn)^2R^2d^2}{256\pi \mathrm{}\mathrm{\Delta }^2}`$ , which plays a role in the decoherence process for much longer time , does not depends on $`\omega _0`$, so in a long time scale only different $`R`$ (or $`\lambda _{mn}`$) can notably change the decaying behavior of quantum coherence. Figure.1a shows this result qualitatively. It is clearly seen in Figure .1b that the curves $`|O_{mn}(t)|`$ almost remain unchanged for different $`\omega _0`$. Therefore, one can artificially control the quantum dynamic process of decoherence for the BEC by adjusting the physical parameter, namely, the intensity of the classical pump laser.
As shown above quantum decoherence can only happen for the superposition of the Fock states of different atomic numbers. If we understand BEC state as a Fock state with definite atomic number $`N`$, no quantum decoherence happens and the BEC is certainly robust. However, there exists a different view: an assembly of the BEC atoms must be assigned a definite phase and so the BEC is not a number state though it has an average atomic number $`N`$. From this viewpoint a good pure state description for BEC is obtained by using the coherent state,which can survive in an usual open environment much longer than the number state does. Since the BEC coherent state is a superposition of atomic number states, the quantum decoherence can be induced by the Rayleigh scattering far-off-resonance. By testing the different effects of decoherence for the number state and the coherent state, we can judge which of the two views concerning BEC reflects the physical reality better.
To further analyses the influences of decoherence on BEC, we consider atomic tunneling in the BEC formed in a symmetric double-well atomic trap, whose potential $`V(𝐫)`$ has two well separated minima at $`x=\pm \frac{a}{2}`$. Assume the potential is such that the two lowest states $`\varphi _0(𝐫)`$ and $`\varphi _1(𝐫)`$ with even and odd parities are closely spaced and well separated from higher levels. Their energy difference $`\mathrm{}\delta \frac{1}{M\xi }e^{2\xi a}`$ where $`\xi =\sqrt{2MV_0}`$depends on the height of maxima of the double-well potential. In this case with very large $`\xi ,`$ the interactions with the atoms in higher levels do not significantly change the dynamics of atoms in the two lowest states. Thus, a two-mode approximation is permitted for the many-body description of the BEC system. In this way we can quantize the atomic fields as
$$\widehat{\mathrm{\Psi }}_g(𝐫)b_0\varphi _0+b_1\varphi _1=b_r\varphi _r+b_l\varphi _l$$
(14)
where we have introduced the annihilation operators $`b_{r,l}`$ =$`\frac{1}{\sqrt{2}}(b_0\pm `$ $`b_1)`$ for the right and the left local modes $`\varphi _{r,l}(𝐫)=\frac{1}{\sqrt{2}}[\varphi _0(𝐫)`$ $`\pm \varphi _1(𝐫)]`$. If the position uncertainty in the states $`\varphi _l(𝐫)`$ and $`\varphi _r(𝐫)`$ is much less than the separation of the minima of the global potential $`V(𝐫)`$ , the local modes may be treated as orthogonal modes and $`[b_r,b_l^{}]0`$. Then, neglecting the interatomic interaction we can write down the effective Hamiltonian
$`\widehat{_e}`$ $`=`$ $`\mathrm{}\mathrm{\Omega }𝐍+\mathrm{}\delta 𝐓+{\displaystyle \underset{𝐤}{}}\mathrm{}\omega _𝐤a_𝐤^{}a_𝐤+`$ (16)
$`({\displaystyle \underset{𝐤}{}}\mathrm{}(\mu _𝐤𝐍+\zeta _k𝐓)a_𝐤+H.c),`$
for the double-well BEC scattered by a far-off resonance light. Here, $`𝐍=b_l^{}b_l+b_r^{}b_r`$ is the total atomic number and $`𝐓=b_lb_r^{}+b_rb_l^{}`$ is the tunnelling operator between the two minima. The corresponding tunneling frequency is just the transition frequency $`\delta `$ of the two non-local motional modes $`\varphi _0`$ and $`\varphi _1`$. Here, the effective coupling constants $`\mu _𝐤`$ and $`\zeta _k`$ are defined by $`\mu _𝐤=\mathrm{}g_kd^3𝐫\varphi _{l\text{ }}^{}(𝐫)\varphi _{l\text{ }}(𝐫)e^{i(𝐤𝐤_0)𝐫}`$ and $`\zeta _k=\mathrm{}g_kd^3𝐫\varphi _{l\text{ }}^{}(𝐫)\varphi _{r\text{ }}(𝐫)e^{i(𝐤𝐤_0)𝐫}`$ .
If the BEC system is initially prepared in the left local mode, we describe it by a coherent state $`|\alpha _l=_{n=0}^{\mathrm{}}c_n(0)|n_l`$where $`c_n=e^{\frac{1}{2}|\alpha |^2}\frac{\alpha ^n}{\sqrt{n!}}`$. Interacting with the vacuum modes of the electromagnetic fields, the BEC system plus the vacuum field will be driven from the factorized state $`|\mathrm{\Psi }(0)=_{n=0}c_n|n_l|0_r|0`$ ($`|0_r`$ denotes the vacuum of the right local mode in the double well) into an entangled state
$`|\mathrm{\Psi }(t)={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{n}{}}}c_ne^{in(\mathrm{\Omega }+\delta )t}f_m^n(t)|nm,m|v_m^n(t)`$
where
$`|m,k={\displaystyle \frac{1}{\sqrt{m!k!}}}b_1^mb_0^k|0`$
$$f_m^n(t)=\frac{(1)^m}{(\sqrt{2})^n}\sqrt{\frac{n!}{(nm)!m!}}e^{2ik\delta t}$$
(17)
$`|v_m^n(t)={\displaystyle \underset{k}{}}e^{it[\omega _𝐤a_𝐤^{}a_𝐤[\mu _𝐤n+\zeta _k(n2m)a_𝐤+H.c)]}|0`$
Based on the above exact solution, we now consider the atomic tunneling between the two condensates in the presence of decoherence. The atomic tunneling is usually characterized by the population difference between two condensates
$`p(t)`$ $`=`$ $`b_r^{}b_rb_l^{}b_l=2Re(b_1^{}b_0)`$ (18)
$`=`$ $`2Re\{{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{m=0}{\overset{n1}{}}}|c_n|^2f_m^nf_{m+1}^n\sqrt{(m+1)(nm)}O_m\}`$ (19)
where $`O_m=v_{m+1}^n(t)|v_m^n(t)`$is the decoherence factor for the entangling vacuum fields. According to Eq.(9), we can roughly write $`O_mJe^{imS}`$ for the real time-dependent functions $`J(t)`$ and $`R(t)`$ and then calculate out the population difference in a compact form
$`p(t)`$ $`=`$ $`Re(J\alpha ^2e^{\frac{1}{2}\left(1e^{iR}\right)\alpha ^2}e^{2i\delta t})`$ (20)
$`=`$ $`J\alpha ^2\mathrm{cos}(2\delta t)Ree^{\frac{1}{2}\left(1e^{iS}\right)\alpha ^2}+`$ (22)
$`J\alpha ^2\mathrm{sin}(2\delta t)Re(ie^{\frac{1}{2}\left(1e^{iS}\right)\alpha ^2})`$
where $`J(t)=|O_m|`$ is a damping factor similar to that in Eq.(6).
The above result answers the following natural question: what is the effect of decoherence on the quantum coherent atomic tunneling ? It indicates that the decoherence always tends to suppress the atomic tunneling current between the two condensates since $`J(t)`$ is decaying with time $`t.`$ The similar observation was even made by Kuang et.al most recently, but our result seems to be more clearly. Without quantum decoherence ( $`O_m=1)`$ the atomic tunneling manifests a simple harmonic oscillation, $`p(t)=\alpha ^2\mathrm{cos}(2\delta t).`$ Generally it is modified with a decaying factor $`J(t)`$ and a phase shift $`\theta `$ defined by
$$\mathrm{tan}\theta =\frac{Ree^{\frac{1}{2}\left(1e^{iS}\right)\alpha ^2}}{Re(ie^{\frac{1}{2}\left(1e^{iS}\right)\alpha ^2})}$$
(23)
In Figure 2 we show this modification by plotting the time-dependent curves of $`p(t)`$ for different $`\theta `$ and $`J(t)`$.
We have shown that a strong far off-resonant pump laser applied to a BEC can produce quantum entanglement between the BEC and the many-mode quantized vacuum field. It thus can result in quantum decoherence in the BEC. One of the most interesting indications of this observation is the possibility to control the dynamic processes of the BEC decoherence in the BEC by engineering the coupling between the BEC and its environment. We notice the effective coupling coefficients $`g_𝐤`$ or $`\mu _𝐤`$ and $`\zeta _k`$ are determined by the Rabi frequency $`R`$ of the pump laser and the pump frequency $`\omega _0.`$ Therefore, the quantum decoherence process can be controlled by adjusting the intensity of the pump laser and the pump frequency $`\omega _0`$. It can be seen from Eq.(6) that the norm of the decoherence factor is more sensitive to the intensity of the pump laser than to the pump frequency $`\omega _0`$. So the dominant element governing the induced quantum decoherence of the BEC is the intensity of the pump laser rather than the pump frequency $`\omega _0`$. The above discussion in this paper is echoed by current experiments. For instance, a recent experiment by Ketterle’s group studied a condensate driven by a far-off resonant pump laser. Its generalization maybe clearly demonstrate many aspects of the present theoretical investigation for engineering the decoherence theory of BEC. In fact, certain engineered environments have been implemented recently to observe the decoherence of ion and cooled atom systems quantitatively.
There also exists another way to engineer the induced environment around the BEC so that the quantum decoherence can be controlled effectively, that is, to put the system in a an optical micro-cavity so that the spectral density $`\mu (k)`$ of the vacuum fields can be changed dramatically by the boundary of cavity. In fact, as a classic aspect of cavity quantum electromagnetic theory, the suppression and enhancement of spontaneous emission of atom within the cavity has been studied in many theoretical and experimental investigations. Since the quantum dissipation caused by spontaneous emission can be controlled by the cavity, it is natural to expect that quantum decoherence,which is another quantum irreversible process, can also be controlled . We will study,in the future research, the quantum decoherence of the BEC system under the influence of cavity field with various spectral density in details.
###### Acknowledgements.
This work is supported in part by the National Foundation of Natural Science of China.
|
warning/0002/quant-ph0002042.html
|
ar5iv
|
text
|
# References
ON THE NON-ORTHONORMALITY OF LIPPMANN-SCHWINGER-LOW STATES
V. J. Menon<sup>1</sup> and B. K. Patra<sup>2</sup>
<sup>1</sup> Department of Physics, Banaras Hindu University, Varanasi 221 005, India
<sup>2</sup> Variable Energy Cyclotron Centre, 1/AF Bidhan Nagar, Calcutta 700 064, India
Abstract
It is pointed out that for a general short-ranged potential the Lippmann-Schwinger-Low scattering state $`|\psi _k^L`$ does not strictly satisfy the Schrodinger eigen equation, and the pair $`|\psi _n^L`$, $`|\psi _k^L`$ is mutually nonorthogonal if $`E_n=E_k`$. For this purpose, we carefully use an infinitesimal adiabatic parameter $`ϵ`$, a nonlinear relation among transition amplitudes, and a separable interaction as illustration.
PACS : 03.65.Nk, 03.80. + r
Introduction
The Lippmann-Schwinger-Low (LSL) integral equations for state vectors and transition matrices form the backbone of quantum scattering theory . They provide the basis for deriving the Born series in wave mechanics , reaction amplitudes in rearrangement collisions , Dyson’s perturbation expansion in the Dirac picture , and various cross sections in old-fashioned quantum electrodynamics . The aim of the present paper is to examine some features of the LSL equations which have not been treated adequately in the existing literature. To be more precise, Lemmas A, B, C and D below answer the following four questions : (i) Are the LSL representations strictly equivalent to the underlying Schrodinger eigen equations? (ii) What is a general off/on energy-shell unitarity-like relation obeyed by the LSL transition amplitudes? (iii) Do various LSL state vectors accurately satisfy the orthonormality relations mentioned by Goldberger-Watson ? (iv) Can we confirm the results explicitly in the case of a separable potential for which the LSL solutions can be obtained in closed form ?
Preliminaries
We denote the free and full Hamiltonian operators by $`H^o`$ and $`HH^o+V`$ respectively with $`V`$ being a short-range interaction. Their continuum eigenkets obey the Schrodinger (superscript S) equations
$`(E_kH^o)|k`$ $`=`$ $`0`$ (1)
$`(E_kH)|\psi _k^S`$ $`=`$ $`0`$ (2)
where the masses are assumed to be renormalized so that energies do not shift. For later convenience we also introduce the free resolvent $`G_k^o`$, the complex projector $`\eta _k^o`$ onto free states of energy $`E_k`$, $`\pi `$ times a Dirac delta $`D_k^o`$, related functions $`\mu _{nk}`$ and $`d_{nk}`$ along with a useful identity via
$`G_k^o={\displaystyle \frac{1}{E_kH^o+iϵ}};\eta _k^o=iϵG_k^o;\mu _{nk}={\displaystyle \frac{iϵ}{E_kE_n+iϵ}}`$ (3)
$`D_k^o=\pi \delta (E_kH^o)=ϵG_k^oG_k^o;d_{nk}={\displaystyle \frac{ϵ}{(E_kE_n)^2+ϵ^2}}`$ (4)
$`{\displaystyle \frac{G_n^oG_k^o}{E_kE_n+iϵ}}={\displaystyle \frac{E_kE_n+2iϵ}{E_kE_n+iϵ}}G_n^oG_k^o=\left(1+\mu _{nk}\right)G_n^oG_k^o`$ (5)
where $`ϵ+0`$ is an adiabatic parameter, and $`\mu _{nk}`$ and $`d_{nk}`$ vanish if $`E_nE_k`$. It is customary to replace Eq.(1) $`\&`$ (2) by the LSL representations (labeled by the superscript $`L`$)
$`|\psi _k^L`$ $`=`$ $`|k+G_k^oV|\psi _k^L:LS`$ (6)
$`=`$ $`|k+(E_kH+iϵ)^1V|k:Low`$ (7)
obeying plane $`+`$ outgoing boundary conditions. Our objective is to propose a few Lemmas on some algebraic properties of $`|\psi _k^L`$ below by paying careful attention to the $`ϵ`$ factors.
LEMMA A (COMPARISON WITH SCHRODINGER) :
“In sharp contrast to the underlying Eq.(1) the LSL states satisfy
$`(E_kH+iϵ)|\psi _k^L`$ $`=`$ $`iϵ|k,`$ (8)
or equivalently
$`(E_kH)|\psi _k^L`$ $`=`$ $`\eta _k^oV|\psi _k^L\mathrm{"}`$ (9)
Proof
Eq.(8) follows from the application of the operator $`(E_kH^o+iϵ)`$ on Eq.(6) or $`(E_kH+iϵ)`$ on Eq.(7). It suggests that $`|\psi _k^L`$ is not a strict eigenket of $`H`$ for any nonzero infinitesimal $`ϵ`$. Eq.(9) is an outcome of the fact that $`iϵ|\psi _k^L=\eta _k^oV|\psi _k^L`$ is generally a nonzero ket. Indeed, the matrix element $`n|\eta _k^oV|\psi _k^L=\mu _{nk}n|V|\psi _k^L`$ becomes the on-shell transition amplitude if $`E_n=E_k`$.
LEMMA B (NONLINEAR RELATION FOR T-MATRIX) :
“The amplitudes $`T_{nk}^Ln|V|\psi _k^L`$ fulfill a nonlinear relation
$`(T_{nk}^LT_{kn}^L^{})/(E_kE_n+iϵ)=(1+\mu _{nk})C_{nk}^L`$ (10)
$`C_{nk}^L=\psi _n^L|VG_n^oG_k^oV|\psi _k\mathrm{"}`$ (11)
Proof
From Eq.(6) we first obtain $`n|`$ and thereby write
$`T_{nk}^L=\psi _n^L|V|\psi _k^L\psi _n^L|VG_n^oV|\psi _k^L`$ (12)
Subtracting a similar expression for $`T_{kn}^L^{}\psi _n^L|V|k`$ and employing the identity (5) the desired Lemma follows.
Incidentally, in the special case of $`E_n=E_k`$ our Eqs.(10), (11) reduce to the usual on-shell unitarity relation \[2-5\] viz.
$`\left[T_{nk}^LT_{kn}^L^{}\right]_{E_n=E_k}=2iA_{nk}^L`$ (13)
$`A_{nk}^L=\left[ϵC_{nk}^L\right]_{E_n=E_k}=\psi _n^L|VD_k^oV|\psi _k^L`$ (14)
LEMMA C (NONORTHONORMALITY) :
“Consider the overlap $`I_{nk}^L\psi _n^L|\psi _k^L`$ between two arbitrary outgoing LSL states. In sharp contrast to the conventional erroneous value $`n|k`$ for the overlap its correct value is
$`I_{nk}^L=n|kd_{nk}A_{nk}^L\mathrm{"}`$ (15)
with $`d_{nk}`$ given by Eq.(4) and $`A_{nk}^L`$ by Eq.(14).
Proof
Upon using the Low form for $`\psi _n^L|`$ and the LS form for $`|\psi _k^L`$ (cfs. Eqs. 6,7) one finds
$`\psi _n^L|\psi _k^L=n|\psi _k^L+n|V(E_nHiϵ)^1|\psi _k^L`$ (16)
In the usual Goldberger-Watson treatment (labeled by the superscript G) one erroneously assumes that $`H|\psi _k^L=E_k|\psi _k^L`$ and reduces Eq.(16) into
$`I_{nk}^G=n|k+n|V\left({\displaystyle \frac{1}{E_kE_n+iϵ}}+{\displaystyle \frac{1}{E_nE_kiϵ}}\right)|\psi _k^L=n|k`$ (17)
In our opinion the use of Eqs. (8), (9) as eigenket statement is quite risky and it is much safer to employ the LS representations (6) for both $`\psi _n^L|`$ and $`|\psi _k^L`$. Then
$`I_{nk}^L`$ $`=`$ $`n|k+n|G_k^oV|\psi _k^L+\psi _n^L|VG_n^o|k`$ (18)
$`+\psi _n^L|VG_n^oG_k^oV|\psi _k^L`$
which is readily shown to coincide with the Lemma (15) in view of the properties (Eq.(10)) and (Eq.(14)). The fact that $`I_{nk}^L`$ reduces to $`n|k`$ if $`E_nE_k`$ but fails to do so if $`E_n=E_k`$ is very disturbing because it implies that the set of LSL states $`|\psi _n^L`$ which are degenerate at a given collision energy $`E_k`$ are mutually nonorthogonal.
LEMMA D (ILLUSTRATION) :
“Consider a rank $`1`$ separable potential $`V=\lambda |gg|`$ with $`\lambda `$ being a real coupling and $`|g`$ a wave packet. Then, the overlap $`\psi _n^L|\psi _k^L`$ can be independently shown to be
$`I_{nk}^L=n|kd_{nk}\lambda ^2g_ng_k^{}{\displaystyle \frac{g|D_k^o|g}{\mathrm{\Delta }_n^{}\mathrm{\Delta }_k}}`$ (19)
where the form factor $`g_k`$ and Fredholm determinant $`\mathrm{\Delta }_k`$ are defined by
$`g_k=k|g;\mathrm{\Delta }_k=1\lambda g|G_k^o|g\mathrm{"}`$ (20)
Proof
With $`V=\lambda |gg|`$, Eq.(6) is readily solved in closed form as
$`|\psi _k^L`$ $`=`$ $`|k+G_k^o|g\left(\lambda g_k^{}/\mathrm{\Delta }_k\right)`$
$`\psi _n^L|`$ $`=`$ $`n|+\left(\lambda g_n/\mathrm{\Delta }_n^{}\right)g|G_n^o`$ (21)
Then it is straightforward to compute
$`I_{nk}^L=n|k{\displaystyle \frac{\lambda ^2g_ng_k^{}}{\mathrm{\Delta }_n^{}\mathrm{\Delta }_k}}\left[{\displaystyle \frac{\mathrm{\Delta }_k\mathrm{\Delta }_n^{}}{\lambda (E_kE_n+iϵ)}}g|G_n^oG_k^o|g\right]`$ (22)
which coincides with the stated lemma in view of the useful identity (5). Of course, the illustrative Eq.(19) and the general result Eq.(15) are in complete agreement although they were derived by different methods.
CONCLUSIONS
The main findings of the present paper are contained in Lemmas A, B, C, and D. The nonorthogonality of the LSL states (for $`E_n=E_k`$, $`nk`$) implies that, even in the absence of bound states, the Moller operator connecting $`|k`$ to $`|\psi _k^L`$ may be nonunitary and $`_k|\psi _k^L\psi _k^L|`$ may loose its interpretation as the unit matrix. Several standard results of scattering perturbation theory \[1-7\] based on the LSL states may require re-examination. Before ending, it may be added that the present work is not concerned with another peculiarity of the LS equation - the Faddeev ambiguity - arising from the noncompactness of the kernel. We also believe that the time-dependence of the LSL states will be much richer than the standard Schrodinger kets $`|\psi _k^S(t)`$ but this aspect will be dealt-with in a future communication.
ACKNOWLEDGEMENTS : VJM thanks the UGC, Goverment of India, New Delhi for financial support.
|
warning/0002/cond-mat0002083.html
|
ar5iv
|
text
|
# Diffusion limited aggregation as a Markovian process. Part I: bond-sticking conditions
## I Introduction
Diffusion limited aggregation (DLA) has been the subject of extensive study since it was first introduced. This model exhibits a growth process that produces highly ramified self similar patterns, which are believed to be fractals . It seems that DLA captures the essential mechanism in many natural growth processes, such as viscous fingering , dielectric breakdown , etc. It is now understood that the Laplace equation, which is common to all of these processes and to DLA, has a major role in the resemblance between them. One of the interesting features of DLA is that there are no parameters to fine-tune in order to obtain a fractal. It thus differs from ordinary critical phenomena, and belongs to the class of self organized criticality (SOC) . In spite of the apparent simplicity of the model, an analytic solution is still unavailable. Particularly, the exact value of the fractal dimension is not known.
In DLA there is a seed cluster of particles fixed somewhere. A particle is released at a distance from the cluster, and performs a random walk until it attempts to penetrate the fixed cluster, in which case it sticks. Then the next particle is released and so on. There are two common types of sticking conditions. The sticking condition described above is called “bond-DLA”, because it occurs when a particle goes into a perimeter bond. In “site-DLA”, sticking occurs as soon as the particle arrives in a perimeter site. This paper deals with bond-DLA, whereas part II deals with site-DLA. The large scale structure of DLA is not sensitive to the type of sticking conditions used .
It has been shown that bond-DLA is equivalent to the dielectric breakdown model (DBM) with $`\eta =1`$ . DBM is a cellular automaton that is defined on a lattice. It consists of the following steps: one starts with a seed cluster of connected sites and with a boundary surface far away from it. A field $`\mathrm{\Phi }`$, which corresponds to the electrostatic potential, is found by solving the discrete Laplace equation on a lattice,
$$^2\mathrm{\Phi }=0,$$
(1)
with the following boundary conditions: the aggregate is considered to have a constant potential that is usually set to $`0`$, and the potential gradient on the distant boundary is set to $`1`$ in some arbitrary units (some use a constant potential on the distant boundary instead). In this paper we set the distant boundary at infinity, and ignore the exponentially small finite size corrections. After solving the discrete Laplace equation (1), the field $`\mathrm{\Phi }`$ determines the growth probabilities per perimeter bond. More specifically, the growth probabilities are proportional to the electric field to some power $`\eta `$. The electric field is simply equal to the potential difference across each bond. Because the potential is set to $`0`$ on the aggregate, the electric-field is equal to the potential value at the sites lying across the perimeter bonds. Thus,
$$P_b=\frac{|\mathrm{\Phi }_b|^\eta }{_b|\mathrm{\Phi }_b|^\eta }.$$
(2)
Here, $`b`$ is the bond index.
DLA and DBM can be grown in various geometries. By geometry we refer to the dimensionality of the lattice, to the shapes of the boundaries and to the details of the seed for growth (usually a point or a line for two dimensional growth). For instance, the case in which the distant boundary is a sphere is called radial boundary conditions, and the case in which the boundary is a distant plane at the top, while the seed cluster is a parallel plane at the bottom, with periodic boundary conditions on the sides, is called cylindrical boundary conditions. In this paper we only consider the cylindrical case, with a relatively short period length (width), from $`N=2`$ to about $`N=7`$, although the method described here could also be used for wider cases.
Recently we published an exact solution to DLA in cylindrical geometry of width $`N=2`$ . The present paper generalizes and extends that solution. Our approach follows the dynamics of the interface. The interface alone determines the growth probabilities at each time step, and whatever lies behind it is irrelevant. This is because the solution to the Laplace equation is unique, provided that the boundary conditions are well defined. We now give a brief summary of Ref. . The characterization of the interface for $`N=2`$ is simple; The interface is fully characterized by a single parameter (usually denoted by $`i`$ or $`j`$), which corresponds to the height difference between the two columns. This height difference, referred to as the step size, can be infinitely large; see Fig. 1. If the interface is flat ($`j=0`$), one can assume that the next particle will always stick on the right side, without limiting the generality of this discussion. This means that the step size can always be considered as nonnegative. The Markovian dynamics is then presented using the Master equation,
$$P_i(t+1)=\underset{j=0}{\overset{\mathrm{}}{}}E_{i,j}P_j(t),$$
(3)
where $`P_j(t)`$ is the probability that the step size is $`j`$ at time $`t`$, and $`E_{i,j}`$ is the time independent conditional probability that an initial step size $`j`$ will become $`i`$ after the next growth process. An example with several possible transitions is shown in Fig. 2. $`𝐏(t)`$ is called the state vector and $`𝐄`$ is called the evolution matrix. In principle, a similar Master equation can be written for more complex growth situations, provided the various configurations can be indexed with a single index $`j`$. Being made out of conditional probabilities, the elements of the evolution matrix obey that,
$`0`$ $`E_{i,j}1,i,j=0,\mathrm{},\mathrm{},`$ (4)
$`{\displaystyle }`$ $`{}_{i=0}{}^{\mathrm{}}E_{i,j}^{}=1,j=0,\mathrm{},\mathrm{}.`$ (5)
After many iterations of Eq. (3) the system converges to a fixed point $`𝐏^{}`$, also called the steady state, which represents the asymptotic time distribution of the step sizes. From the steady state and the evolution matrix we are able to extract the average upward growth probability $`p_{\mathrm{up}}^{}`$, the average density $`\rho `$ and the fractal dimension $`D`$.
In order to obtain an analytic expression for the elements of the evolution matrix, one must first solve the Laplace equation. Having found the solutions $`\mathrm{\Phi }(m,n)`$, the growth probabilities are found from Eq. (2). The denominator there, which comes from the normalization, is particularly simple for the special case of $`\eta =1`$, where the discrete version of the divergence theorem implies that
$$\underset{b}{}\mathrm{\Phi }_b=N.$$
(6)
The actual growth probability into a site is then found from
$$p_{\mathrm{site}}=\underset{\mathrm{bonds}\mathrm{into}\mathrm{site}}{}p_b.$$
(7)
The solution of the Laplace equation is now divided into two parts. In the first part, we solve the Laplace equation for the ’upper’ part of space, which starts just above the highest particle of the aggregate and continues upwards to infinity. In the example of Fig. 1, this part contains all the rows with $`m>0`$. As we explain below, this solution is completely determined by the boundary conditions and by the values of the potential at the row with $`m=0`$, i.e. $`\{\mathrm{\Phi }(0,n)\}`$. We then solve the Laplace equation for the ’lower’ part ($`m0`$ in Fig. 1), and find the values of $`\{\mathrm{\Phi }(0,n)\}`$ from matching the two regimes. The solution in the ’upper’ part is given as a combination of solutions of the form
$$\mathrm{\Phi }(m,n)=e^{\kappa m+ikn},$$
(8)
with the dispersion relation
$$\mathrm{sinh}(\frac{\kappa }{2})=\pm \mathrm{sin}(\frac{k}{2})$$
(9)
and with the discrete set of allowed values $`k_l=\frac{2\pi }{N}l`$, which follow from the periodic lateral boundary conditions, which require that $`e^{ikN}=1`$. The boundary conditions at infinity have a uniform gradient, i.e.,
$$\underset{m\mathrm{}}{lim}(\mathrm{\Phi }(m+1,n)\mathrm{\Phi }(m,n))=1,n=0,\mathrm{},N1.$$
(10)
Given the arbitrarily set of values $`\mathrm{\Phi }(0,n)`$, the solution for the row $`m=1`$ is
$$\mathrm{\Phi }(1,n)=1+\underset{n^{}=0}{\overset{N1}{}}\mathrm{\Phi }(0,n^{})g_N(|nn^{}|),$$
(11)
where
$$g_N(n)\frac{1}{N}\underset{l=0}{\overset{N1}{}}e^{\kappa _l}\mathrm{cos}(k_ln),n=0,\mathrm{},N1,$$
(12)
is the boundary Green’s function, and $`\kappa _l`$ corresponds to $`k_l`$ via the dispersion relation (9). The solution is given only for $`m=1`$, because we are only interested in the potential at sites near the interface. Note that the Green’s function has the general property
$$\underset{n=0}{\overset{N1}{}}g_N(n)=1$$
(13)
. It is therefore good practice to check this normalization for each of the calculations presented below. Indeed, all our results obey this rule.
In general, the solution in the ’lower’ regime is complicated by the variety of configurations. However, this solution is very simple for $`N=2`$, when $`\mathrm{\Phi }(m,0)`$ is a linear combination of $`e^{\kappa _fm}`$ and $`e^{\kappa _fm}`$. Since $`\mathrm{\Phi }(j,0)=0`$, one is left with one unknown $`\mathrm{\Phi }(0,0)`$, to be determined by the matching at row $`0`$.
For the special case $`N=2`$, the above procedure has led to the exact solution
$`E_{i,j}`$ $`=`$ $`\{\begin{array}{ccc}y(\mathrm{})e^{\kappa _fi}\frac{1e^{2\kappa _f(ji)}}{1+\beta e^{2\kappa _fj}}& ,\hfill & 0ij2\hfill \\ \frac{3}{2}y(\mathrm{})e^{\kappa _f(j1)}\frac{1e^{2\kappa _f}}{1+\beta e^{2\kappa _fj}}& ,\hfill & i=j1\hfill \\ E_{\mathrm{}+1,\mathrm{}}\left(1\alpha \frac{e^{2\kappa _fj}}{1+\beta e^{2\kappa _fj}}\right)& ,\hfill & i=j+1\hfill \\ 0& & \mathrm{otherwise}\hfill \end{array},`$ (19)
$`j1,`$
where
$$E_{\mathrm{}+1,\mathrm{}}=\underset{j\mathrm{}}{lim}E_{j+1,j}=\frac{1+g_2(1)y(\mathrm{})}{2}=0.5658\mathrm{},$$
(20)
$`y(\mathrm{})=\sqrt{3}\sqrt{2}=0.3178\mathrm{}`$, $`e^{\kappa _f}=2\sqrt{3}=0.2679\mathrm{}`$, $`\alpha =(1+\beta )g_2(1)y(\mathrm{})/(2E_{\mathrm{}+1,\mathrm{}})=0.1281\mathrm{}`$ and $`\beta =5\sqrt{24}=0.1010\mathrm{}`$. For $`j=0`$, the interface will transform into a step of size $`j=1`$ with probability $`1`$, hence $`E_{1,0}=1`$ and $`E_{i,0}=0`$ for $`i1`$. The values of $`E_{i,j}`$ for $`0i,j4`$, up to the fourth decimal digit, are
$$𝐄=\left[\begin{array}{cccccc}0& 0.4393& 0.3160& 0.3177& 0.3178& \mathrm{}\\ 1& 0& 0.1185& 0.0847& 0.0851& \\ 0& 0.5607& 0& 0.0318& 0.0227& \\ 0& 0& 0.5655& 0& 0.0085& \\ 0& 0& 0& 0.5658& 0& \\ \mathrm{}& & & & & \mathrm{}\end{array}\right].$$
(21)
The first diagonal below the main, which represents the probabilities for the step to grow larger by one, $`E_{j+1,j}`$, approaches its asymptotic value of $`E_{\mathrm{}+1,\mathrm{}}=0.5658\mathrm{}`$ exponentially, as the third row of Eq. (19) indicates. The diagonal above the main represents the probabilities for growths at the bottom of the fjord, $`E_{j1,j}`$, and corresponds to the second row in Eq. (19). These probabilities decay exponentially as the step size $`j`$ grows. According to the first row in Eq. (19), the elements $`E_{i,j}`$ converge exponentially for large $`j`$’s to a simple exponential function:
$$E_{i,\mathrm{}}=\underset{j\mathrm{}}{lim}E_{i,j}=y(\mathrm{})e^{\kappa _fi}.$$
(22)
These probabilities relate to the transition from a very large step to a step of size $`i`$. Next, the steady state vector $`𝐏^{}`$ is computed and used to evaluate the average upward growth probability $`p_{\mathrm{up}}^{}`$, which in turn, determines the average density $`\rho `$ and the fractal dimension $`D`$. These computations are explained later in Sec. II.
Our previous paper does not specify details concerning the manner in which the system converges to the steady state in time. Besides addressing this issue, our present paper also treats DLA grown in wider geometrical periods (still in cylindrical geometry). The basic approach is the same, i.e., we try to characterize the possible configurations of the interface for wider periods, and then write the evolution matrix, which is composed of the growth probabilities, which are computed from the Laplace potential, after proper normalization. The first difficulty encountered is in the characterization. For example, already for a width of $`N=3`$ one cannot characterize the interface using a single parameter as in the case $`N=2`$, nor is it easy doing so using $`2`$ parameters, or more. Instead, we make a manual list of possible configurations of the interface, which we then order according to the difference in height between the highest and lowest points on the interface. This difference is denoted by $`\mathrm{\Delta }m`$. Our order-$`O`$ approximation includes only the configurations with $`\mathrm{\Delta }mO`$. In our approximation, some of these configurations represent many other (excluded) configurations, in the sense that they have very similar growth probabilities, especially upward. This is because of the screening quality of the Laplace equation, which causes the potential to decay exponentially inside fjords. Thus, the deeper parts of the interface have a very small effect on the upward growth probability. The finite list of configurations is indexed arbitrarily, with an index usually denoted by $`i`$ or $`j`$. Our experience shows that accurate results are obtained, only when the order of approximation $`O`$ is comparable to the width of the cylinder $`N`$. Thus, for wide periods, a high order calculation is called for. This causes the method to be ineffective for very wide periods, because the number of configurations grows exponentially with the order of approximation. We conducted calculation up to $`N=7`$.
After selecting the finite list of configurations and obtaining the finite evolution matrix, we compute the steady state vector, which is the fixed point of the matrix (the normalized eigenvector with an eigenvalue of $`1`$). For each configuration, we identify the upward growth processes (when the newly attached particle is higher than the rest). We then calculate the average upward growth probability $`p_{\mathrm{up}}^{}`$ as a weighted average over the configurations. ¿From $`p_{\mathrm{up}}^{}`$ we calculate the average density $`\rho `$ and the fractal dimension $`D`$. The computed values of $`p_{\mathrm{up}}^{}`$, from different orders of the approximation, are compared with numerical simulations in Table I.
In Sec. II we introduce a simple Markov process, called the “frustrated climber”, which we solve exactly. A slight modification of the model is equivalent to site-DLA with a period of $`N=2`$, which is presented in part II of this paper . We then show a way of successively generalizing the model to approximate bond-DLA with a period of $`N=2`$ and with increasing orders $`O`$. We are able to check the approximations by comparing with the exact results of Ref. . This model also enables us to investigate the rate of convergence to the steady state. In this context we describe the convergence in terms of other eigenvectors, with eigenvalues whose absolute values are smaller than $`1`$, and in terms of the infinite shift down operator. We show that the average upward growth probability converges exponentially in time to its steady state value, with a characteristic time constant on the order of unity. In Sec. III we generalize our method to cylindrical DLA with $`N>2`$. We present in detail the calculations for $`N=3`$ with $`O=1`$ and $`O=2`$, and for $`N=4`$ with $`O=1`$. Next we report on numerical results for wider periods and higher orders. In the final section we review the results and summarize.
## II The frustrated climber model
Consider someone trying to climb up a slippery infinite ladder. At each time step the climber climbs up one step with probability $`0p1`$, or falls all the way down with probability $`q1p`$. We call the climber “frustrated”, because the probability to get very high is exponentially small. We wish to compute the probability $`P_i(t)`$ for the climber to be at height $`i`$ after $`t`$ time steps, for $`i=0,\mathrm{},\mathrm{}`$. The Master equation for this problem is $`𝐏(t+1)=\mathrm{𝐄𝐏}(t)`$, where the matrix element $`E_{i,j}`$ is the conditional probability that the climber moves from height $`j`$ to $`i`$ in a single time step. The rules of the model imply that
$$E_{i,j}=\left\{\begin{array}{ccc}p& ,& i=j+1\hfill \\ q& ,& i=0\hfill \\ 0& ,& \text{otherwise}\hfill \end{array}\right\},j0,$$
(23)
so the matrix looks like this:
$$𝐄=\left[\begin{array}{ccccc}q& q& q& q& \mathrm{}\\ p& 0& 0& 0& \\ 0& p& 0& 0& \\ 0& 0& p& 0& \\ \mathrm{}& & & & \mathrm{}\end{array}\right].$$
(24)
This presentation helps us see the resemblance to the dynamics of DLA with $`N=2`$ in Eqs. (19, 21): Eq. (24) would approximate these equations if we were to replace $`E_{j+1,j}`$ by $`pE_{\mathrm{}+1,\mathrm{}}`$ and $`E_{0,j}`$ by $`q`$ for all $`j`$, and neglect all other growth probabilities, which are indeed smaller. We shall discuss this and better approximations for DLA in the next subsections. Because the Markovian matrices for the two cases are similar for large $`j`$’s, we expect that some of the dynamical features are similar as well. We therefore present here an exact solution for the frustrated climber model, and then try to draw conclusions for generalized models which represent successive approximations for DLA. The advantage is that in the simple model of the frustrated climber it is possible to derive a simple analytic expression for the steady state and a complete description of the temporal convergence.
The steady state equations for the frustrated climber model are
$`P^{}`$ $`{}_{i+1}{}^{}={\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}E_{i+1,j}P_j^{}=pP_i^{},i0,`$ (25)
$``$ $`P`$ $`{}_{j}{}^{}=qp^j,j0.`$ (26)
One can easily check that this steady state is normalized,
$$\underset{j=0}{\overset{\mathrm{}}{}}P_j^{}=\underset{j=0}{\overset{\mathrm{}}{}}qp^j=\frac{q}{1p}=1.$$
(27)
The average upward growth probability in the steady state is
$$p_{\mathrm{up}}^{}=\underset{j=0}{\overset{\mathrm{}}{}}P_j^{}p_{\mathrm{up}}(j)=\underset{j=0}{\overset{\mathrm{}}{}}P_jp=p,$$
(28)
where $`p_{\mathrm{up}}(j)`$ stands for the probability to move upwards when the height of the climber is $`j`$. In this simple model $`p_{\mathrm{up}}(j)=p`$ for all $`j`$’s.
We now investigate the temporal convergence to the steady state. We define the vector $`𝐯(t)`$ by
$$𝐏(t)=𝐏^{}+𝐯(t).$$
(29)
Because $`𝐏^{}`$ and $`𝐏(t)`$ are probability vectors, $`_{j=0}^{\mathrm{}}P_j^{}=_{j=0}^{\mathrm{}}P_j(t)=1`$, for any $`t`$, hence
$$\underset{j=0}{\overset{\mathrm{}}{}}v_j(t)=0.$$
(30)
We substitute $`𝐯`$ into the dynamical equation and obtain
$`𝐏`$ $`(t+1)=\mathrm{𝐄𝐏}(t)=𝐏^{}+\mathrm{𝐄𝐯}(t),`$ (31)
$``$ $`𝐯`$ $`(t+1)=\mathrm{𝐄𝐯}(t).`$ (32)
Next, we look for the rest of the eigenvectors of the evolution matrix (any eigenvector $`𝐯`$ with an eigenvalue $`\lambda 1`$, has to obey Eq. (30)). Surprisingly, there are no eigenvectors besides the steady state in this case. The eigenvector equations are
$`\lambda v_0=q{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}v_j=0,`$ (33)
$`\lambda v_{i+1}=pv_i(t),i0.`$ (34)
The first equation implies that either $`\lambda =0`$ or $`v_0=0`$. In both cases, the last equation implies that v=0.
We next introduce the infinite shift-down operator:
$$𝐒\left[\begin{array}{ccccc}0& 0& 0& 0& \mathrm{}\\ 1& 0& 0& 0& \\ 0& 1& 0& 0& \\ 0& 0& 1& 0& \\ \mathrm{}& & & & \mathrm{}\end{array}\right].$$
(35)
This operator causes a vector to “slide down” and inserts a zero at the evacuated component at the top. S has no eigenvectors at all, not even a fixed point (in spite of the fact that $`_{i=0}^{\mathrm{}}S_{i,j}=1`$ for $`j=0,\mathrm{},\mathrm{}`$). In fact, $`\mathrm{𝐄𝐯}=p\mathrm{𝐒𝐯}`$ for all vectors $`𝐯`$ with $`_{j=0}^{\mathrm{}}v_j=0`$.
Nevertheless, the convergence of $`𝐏(t)`$ to $`𝐏^{}`$ is simple. Starting from any initial state vector $`𝐏(t=0)`$, the first application of $`𝐄`$ causes the first component to be set to its steady state value $`P_0(t=1)=q`$. At each subsequent iteration another components is set permanently: $`P_1(t=2)=qp`$, $`P_2(t=3)=qp^2`$, etc. $`P_j`$ becomes equal to $`P_j^{}`$ after no more than $`j+1`$ time steps. The context we are interested in is wider. We wish to compute the convergence of “observables”, i.e., the average of an arbitrary function $`a(j)`$ over configurations. We compute the average at time $`t`$
$$a(t)\underset{j=0}{\overset{\mathrm{}}{}}a(j)P_j(t)=a^{}+\underset{j=0}{\overset{\mathrm{}}{}}a(j)v_j(t),$$
(36)
where $`a^{}_{j=0}^{\mathrm{}}a(j)P_j^{}`$ is the steady state average. Starting from an initial deviation from the steady state $`𝐯(0)`$, each iteration causes a down shift and a multiplication by $`p`$, hence
$$a(t)=a^{}+p^t\underset{j=0}{\overset{\mathrm{}}{}}a(j+t)v_j(0).$$
(37)
Equation (37) is the analogue of the standard eigenvector description. We can also identify here the exponential decay of the factor $`p^t`$. For example, the function $`a(j)=\delta _{j,j_0}`$ “measures” the probability of the climber to be at height $`j_0`$ (at any time). At time $`t`$ the observed average probability is
$$a(t)=P_{j_0}^{}+p^tv_{j_0t}(0),$$
(38)
for $`tj_0`$, and $`a(t)=P_{j_0}^{}`$ for $`t>j_0`$ .
### A First-order approximation for $`N=2`$
We now return to Eq. (19), and try to approximate it by a sequence of models which are related to the frustrated climber model. The simplest approximation would follow if we do not let the particle penetrate into the fjord at all. This is equivalent to setting $`\kappa _f=\mathrm{}`$ in Eq. (19). According to these simplified rules, the particle can either stick at $`(0,0)`$ and create a flat step of $`i=0`$, or it can stick at $`(1,1)`$ and increase the step height by $`1`$. Let us denote the probability for the former event by $`q`$ and the latter by $`p`$. In the first-order approximation we take $`p`$ and $`q`$ to be independent of the initial step size $`j`$, unless $`j=0`$, in which case the step size increases with probability $`1`$. The Markovian matrix E for this case is almost identical to the case of the frustrated climber,
$$𝐄=\left[\begin{array}{ccccc}q_0& q& q& q& \mathrm{}\\ p_0& 0& 0& 0& \\ 0& p& 0& 0& \\ 0& 0& p& 0& \\ \mathrm{}& & & & \mathrm{}\end{array}\right],$$
(39)
the only difference being in the first column, where we denote $`q_0=0`$ and $`p_0=1`$. In part II of this paper we show that this model is exact for the case of site-sticking DLA for $`N=2`$ .
The solution to this problem is very similar to that of the frustrated climber, with small modifications. The steady state is
$$P_j^{}=P_0^{}p_0p^{j1},j1,$$
(40)
where $`P_0^{}`$ can be determined using the normalization condition
$`1={\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}P_j^{}=P_0^{}(1+p_0{\displaystyle \underset{j=0}{\overset{\mathrm{}}{}}}p^j),`$ (41)
$`P_0^{}={\displaystyle \frac{1p}{1p+p_0}}.`$ (42)
The average upward growth probability is evaluated by
$$p_{\mathrm{up}}^{(1)}^{}=P_0^{}p_0+(1P_0^{})p=\frac{p_0}{1p+p_0}.$$
(43)
The superscript $`(1)`$ appears because it is the first-order approximation. We now need to choose $`p`$. One possible choice would be to take $`p=E_{\mathrm{}+1,\mathrm{}}=0.5658`$, because this is the asymptotic upward growth probability, and then set $`q=1p`$. This would give $`p_{\mathrm{up}}^{(1)}^{}=0.6973`$, to be compared with the exact value $`0.6812`$ . An alternative approximation would return to Eq. (1.13), but replace $`y(\mathrm{})`$ by $`q`$, and then find $`q`$ by solving $`1=p+q=[1+g_2(1)q]/2+q`$. This yields $`p=1q=2\sqrt{2}=0.5858`$, and therefore $`p_{\mathrm{up}}^{(1)}^{}=\sqrt{2}/2=0.7071`$.
We next calculate the average density and the fractal dimensionality. Similar to the argument used by Turkevich and Scher , we consider a large number of growth processes $`n`$ in the steady state. During this growth the aggregate would grow higher by $`h=p_{\mathrm{up}}^{}n`$. The total volume covered by the new growth is $`hN^{d1}`$, where $`d=2`$ is the Euclidean dimension. Thus, for $`N=2`$ and for our first approximation the density is
$$\rho =\frac{n}{hN^{d1}}=\frac{n}{p_{\mathrm{up}}^{}nN^{d1}}=\frac{1}{p_{\mathrm{up}}^{}N^{d1}}=0.7171,$$
(44)
to be compared with the exact value $`\rho =0.7340`$. Although our model does not really produce fractal structures (due to the narrow width of our space), we can make an estimate of the fractal dimension in the same way Pietronero $`etal.`$ estimated it in . For a self similar fractal structure, one expects that a change of scale by a factor $`N`$ will change the average mass (number of occupied sites) of a $`N\times N`$ cut by a factor $`N^D`$, where $`D`$ is the fractal dimension. Assuming that the above procedure represents a coarse graining of the sites into $`N\times N`$ cells, we conclude that asymptotically
$$\rho =N^{Dd},$$
(45)
and this means that
$$D=d+\frac{\mathrm{ln}(\rho )}{\mathrm{ln}(N)}=1\frac{\mathrm{ln}(p_{\mathrm{up}}^{})}{\mathrm{ln}(N)}=1.5202.$$
(46)
In Sec. IV we suggest a modified estimate of the fractal dimension, allowing for corrections to the asymptotic form (45).
The study of the convergence to the steady state is again limited to the subspace of vectors $`𝐯`$ with $`_{j=0}^{\mathrm{}}v_j=0`$. The dynamic equation for $`i=0`$ is,
$`v`$ $`{}_{0}{}^{}(t+1)=q_0v_0(t)+{\displaystyle \underset{j=1}{\overset{\mathrm{}}{}}}qv_j(t)=(q_0q)v_0(t),`$ (47)
$``$ $`v`$ $`{}_{0}{}^{}(t)=(q_0q)^tv_0(0).`$ (48)
Since $`q_0=0`$, the exponentiated prefactor is negative, and therefore $`v_0(t)`$ is oscillating during its decay. After the first iteration $`v_1(1)=p_0v_0(0)`$, regardless of its initial value. Afterwards it continues to follow $`v_0`$, i.e., $`v_1(t)=p_0(q_0q)^{t1}v_0(0)`$. After the second iteration $`v_2(2)=p_0pv_0(0)`$, and it also starts to decay exponentially with the factor $`(q_0q)`$. This happens for any $`j>1`$; After more than $`j`$ time steps ($`t>j`$) one has,
$$v_j(t)=p_0p^{j1}(q_0q)^{tj}v_0(0).$$
(49)
For short times and large indices $`t<j`$, the dynamics is governed by the shift down operator:
$$𝐯(t)=v_0(0)(q_0q)^t𝐡+p^t\underset{j=1}{\overset{\mathrm{}}{}}c_j𝐞^{(j+t)},$$
(50)
where $`𝐞^{(j)}`$ are the standard basis vectors, the components of the vector $`𝐡`$ are,
$`h_0`$ $``$ $`1,`$ (51)
$`h_j`$ $``$ $`{\displaystyle \frac{p_0}{p}}\left({\displaystyle \frac{p}{q_0q}}\right)^j,j1,`$ (52)
and the constants $`c_j`$ are determined by the initial conditions,
$$c_j=v_j(0)v_0(0)h_j,j=1,2,\mathrm{}$$
(53)
For $`p>0.5`$ the components of $`𝐡`$ explode exponentially. However, $`_{j=0}^{\mathrm{}}v_j(0)=0`$ and therefore $`lim_j\mathrm{}v_j(0)=0`$. Thus, in order to cancel the divergence of the $`h_j`$’s, the $`c_j`$’s must also explode exponentially and have an opposite sign. We note that because of this divergence $`𝐡`$ does not have a finite $`L_1`$ norm and thus does not belong to the domain of $`𝐄`$. Therefore it is not an eigenvector.
### B Higher-order approximations for $`N=2`$
As mentioned earlier, the frustrated climber model resembles the bond-DLA evolution matrix (19, 21). In this section we approximate the full dynamics using increasingly more complex matrices. By doing so we do not improve on the accuracy of our previously published results , but rather learn about the rate of convergence to the steady state. The method used in this section is generalized and applied to cylindrical DLA with wider periods in the next section. The case $`N=2`$ is the simplest demonstration of this approach.
The second-order approximation is to allow also transitions of the kind $`j1`$ for $`j1`$. We also allow having arbitrary values in the top left $`2\times 2`$ corner of the matrix, which we copy from the original matrix of Eq. (21), i.e.,
$$𝐄=\left[\begin{array}{cccccc}q_0& q_1& q& q& q& \mathrm{}\\ r_0& r_1& r& r& r& \\ 0& p_1& 0& 0& 0& \\ 0& 0& p& 0& 0& \\ 0& 0& 0& p& 0& \\ \mathrm{}& & & & & \mathrm{}\end{array}\right],$$
(54)
We still require that the sum of the elements in each column be equal to $`1`$, i.e.,
$`q_0+r_0=1,`$ (55)
$`q_1+r_1+p_1=1,`$ (56)
$`q+r+p=1.`$ (57)
In terms of standard DLA this means that we allow the particle to penetrate two sites into the fjord, but no more. Indeed it is exponentially improbable to penetrate deep into the fjord. This fact suggests a controlled approximation for DLA. In each order of the approximation we allow the depth of penetration into the fjord to grow by $`1`$. This is done by copying the $`(O+1)\times O`$ upper left block of the original matrix (19, 21), where $`O`$ is the order of approximation. Asymptotic values are used outside this block, i.e.,
$`E_{j+1,j}=E_{\mathrm{}+1,\mathrm{}},jO,`$ (58)
$`E_{i,j}=y(\mathrm{})e^{\kappa _fi},jO,iO2,`$ (59)
$`E_{n1,j}=1{\displaystyle \underset{i=0}{\overset{n2}{}}}y(\mathrm{})e^{\kappa _fi}E_{\mathrm{}+1,\mathrm{}}`$ (60)
$`=y(\mathrm{}){\displaystyle \frac{e^{\kappa _f(n1)}}{1e^{\kappa _f}}},jO,`$ (61)
and the rest of the matrix elements are null. For example, in our case, $`O=2`$, the constants in the matrix (54) are
$`q_0=0,`$ (62)
$`r_0=1,`$ (63)
$`q_1={\displaystyle \frac{63\sqrt{2}}{4}}=0.4393,`$ (64)
$`r_1=0,`$ (65)
$`p_1={\displaystyle \frac{3\sqrt{2}2}{4}}=0.5607,`$ (66)
$`q=y(\mathrm{})=\sqrt{3}\sqrt{2}=0.3178,`$ (67)
$`p=E_{\mathrm{}+1,\mathrm{}}=0.5658,`$ (68)
$`r=y(\mathrm{}){\displaystyle \frac{e^{\kappa _f}}{1e^{\kappa _f}}}=0.1163.`$ (69)
First, the steady state is found by solving $`𝐏^{}=\mathrm{𝐄𝐏}^{}`$, i.e.,
$`P_0^{}=q_0P_0^{}+q_1P_1^{}+q{\displaystyle \underset{j=2}{\overset{\mathrm{}}{}}}P_j^{},`$ (70)
$`P_1^{}=q_0P_0^{}+q_1P_1^{}+r{\displaystyle \underset{j=2}{\overset{\mathrm{}}{}}}P_j^{},`$ (71)
$`P_2^{}=p_1P_1^{},`$ (72)
$`P_{j+1}^{}=pP_j^{},j2.`$ (73)
The solution to the last equation is
$$P_j^{}=P_2^{}p^{j2},j2.$$
(74)
Keeping this in mind it is possible to exchange the two last equations of the set (73) with
$$\underset{j=2}{\overset{\mathrm{}}{}}P_j^{}=p_1P_1^{}+p\underset{j=2}{\overset{\mathrm{}}{}}P_j^{}.$$
(75)
Thus we obtain an autonomous and finite set of $`3`$ equations for $`3`$ unknowns, namely, $`P_0^{}`$, $`P_1^{}`$ and $`\stackrel{~}{P}_2^{}_{j=2}^{\mathrm{}}P_j^{}`$. The third parameter, $`\stackrel{~}{P}_2^{}`$, represents the total probability for the infinitely many configurations with $`j2`$. This reduction of the problem to three parameters became possible because all of the configurations with $`j2`$ have exactly the same transition probabilities to the configurations $`j=0`$ and $`j=1`$, and because they have exactly the same upward growth probability. Thus we obtain a fixed point equation for a $`3\times 3`$ matrix,
$$\left[\begin{array}{c}P_0^{}\\ P_1^{}\\ \stackrel{~}{P}_2^{}\end{array}\right]=\left[\begin{array}{ccc}q_0& q_1& q\\ r_0& r_1& r\\ 0& p_1& p\end{array}\right]\left[\begin{array}{c}P_0^{}\\ P_1^{}\\ \stackrel{~}{P}_2^{}\end{array}\right].$$
(76)
It is guaranteed that a nontrivial solution exists, because the sum of the terms in each column of the finite matrix equals $`1`$. Using the constants from Eqs. (69), the normalized solution obtained is,
$`P_0^{(2)}=0.2705,(0.2696),`$ (77)
$`P_1^{(2)}=0.3184,(0.3113),`$ (78)
$`\stackrel{~}{P}_2^{(2)}=0.4111,(0.4191),`$ (79)
where the superscript denotes the order of approximation and a comparison is drawn to the exact values in parentheses. By “exact” we refer to very high order calculations, or to values from simulations (which are the same up to the presented accuracy of $`10^4`$) . The elements $`P_j^{}`$ for $`j2`$ are evaluated using
$$P_j^{(2)}=(1p)\stackrel{~}{P}_2^{(2)}p^{j2},j2.$$
(80)
It is now possible to evaluate the average upward growth probability
$$p_{\mathrm{up}}^{(2)}^{}=P_0^{}r_0+P_1^{}p_1+\stackrel{~}{P}_2^{}p=0.6816,$$
(81)
where the exact value is $`0.6812`$. The fractal dimension is evaluated as in Eq. (46),
$$D^{(2)}=1.5530,$$
(82)
compared to the exact value $`1.5538`$.
The temporal convergence to the steady state in the second-order approximation can be analyzed using both the shift down operator and eigenvectors. The first eigenvector of the matrix in Eq. (76) is the fixed point solution, which we denote by $`\stackrel{~}{𝐏}^{}`$. Let us denote the other two (three-components) eigenvectors by $`\stackrel{~}{𝐡}`$ and $`\stackrel{~}{𝐠}`$, and their corresponding eigenvalues by $`|\lambda _0||\lambda _1|`$. After $`t`$ iterations of the evolution matrix we have
$$\stackrel{~}{𝐏}(t)=\stackrel{~}{𝐏}^{}+c_0\lambda _0^t\stackrel{~}{𝐡}+c_1\lambda _1^t\stackrel{~}{𝐠},$$
(83)
where $`c_0`$ and $`c_1`$ are constants determined by the initial conditions. The configurational average of some function $`a(j)`$ with $`a(j)=a(2)`$ for $`j>2`$, can be expressed in terms of these eigenvalues only,
$$a(t)=a^{}+k_0\lambda _0^t+k_1\lambda _1^t,$$
(84)
where $`k_0`$ and $`k_1`$ are some other constants. A special function of this type is the upward growth probability, $`p_{\mathrm{up}}(j)=(r_0,p_1,p,p,p,\mathrm{})`$. The eigenvalue with the largest absolute value other than $`1`$, $`\lambda _0`$, makes the largest contribution to the deviation from the steady state values, and thus controls the temporal convergence. The characteristic time constant for the exponential convergence is,
$$\tau =\frac{1}{\mathrm{ln}(|\lambda _0|)}.$$
(85)
The eigenvalues obtained are $`\lambda _0^{(2)}=0.5599`$ and $`\lambda _1^{(2)}=0.1257`$, using the constants of Eqs. (69). Hence, $`\tau ^{(2)}=1.7`$. In order to describe the convergence of $`P_j(t)`$ for $`j2`$ we use the vector $`𝐯(t)=𝐏(t)𝐏^{}`$, once more, and we perform a decomposition similar to Eq. (50):
$$𝐯(t)=c_0\lambda _0^t𝐡+c_1\lambda _1^t𝐠+p^t\underset{j=2}{\overset{\mathrm{}}{}}c_j𝐞^{(j+t)},$$
(86)
where $`c_0`$ and $`c_1`$ are the same as in Eq. (83) and the constants $`c_j`$ for $`j2`$ are determined by the initial condition $`𝐯(0)`$. The vectors $`𝐡`$ and $`𝐠`$ are infinite generalizations of the finite vectors $`\stackrel{~}{𝐡}`$ and $`\stackrel{~}{𝐠}`$, according to
$$\begin{array}{ccc}h_j=\stackrel{~}{h}_j,\hfill & g_j=\stackrel{~}{g}_j,\hfill & j=0,1,\hfill \\ h_2=p_1\stackrel{~}{h}_1,\hfill & g_2=p_1\stackrel{~}{g}_1,\hfill & j=2,\hfill \\ h_j=h_2\left(\frac{p}{\lambda _0}\right)^{j2},\hfill & g_j=g_2\left(\frac{p}{\lambda _1}\right)^{j2},\hfill & j2.\hfill \end{array}$$
(87)
Because $`p=E_{\mathrm{}+1,\mathrm{}}>|\lambda _0|,|\lambda _1|`$, it is apparent that the components $`h_j`$ and $`g_j`$ diverge exponentially for large $`j`$’s. This means that these vectors do not have a finite $`L_1`$ norm, and that they do not belong to the domain of $`𝐄`$. Therefore, they are not eigenvectors, and $`\lambda _0`$ and $`\lambda _1`$ are not eigenvalues of $`𝐄`$. Nevertheless, Eq. (86) is still true. The effect of the shift down operator is manifested in the sum $`p^t_{j=2}^{\mathrm{}}c_j𝐞^{(j+t)}`$.
Using the same method it is possible to make higher order calculations. The steady state quantities resulting from the third order approximation are presented in Table II, in comparison with exact results. The eigenvalue with the largest absolute value is $`\lambda _0^{(3)}=0.5687`$, which has a greater absolute value than $`E_{\mathrm{}+1,\mathrm{}}=0.5658`$. This means that a legitimate eigenvector exists for the infinite matrix. In the fourth and fifth order approximation we get $`\lambda _0^{(4,5)}0.5688`$. This suggests that the higher the order the more accurate is the evaluation of $`\lambda _0`$ and that the accuracy obtained is better than $`10^4`$. The typical time needed to settle in the steady state from any initial condition is therefore as short as
$$\tau =1.8.$$
(88)
## III DLA with $`N>2`$
The generalization of the exact methods from Ref. to $`N>2`$ is not straightforward. Trying to proceed along a similar line, one would try to parameterize the interface with a parameter $`i=1,2,\mathrm{},\mathrm{}`$, and write the Master equation $`P_i(t+1)=_{j=1}^{\mathrm{}}E_{i,j}P_j(t)`$. Unlike the case $`N=2`$, the parameterization for $`N>2`$ is very complicated. For instance, for the case $`N=3`$ it is reasonable to try using two parameters, which indicate the height of two columns relative to the highest (or lowest) third column. However, this is insufficient because complex fjords (involving overhangs) might occur, as shown in Fig. 3. Instead of achieving a perfect parameterization, we adopt the approximate approach of Sec. II B, i.e., we take into account only a finite number of interface configurations. These configurations are classified according to the maximum height difference between the highest and lowest particles on the interface $`\mathrm{\Delta }m`$. In the $`O`$th-order approximation all the configurations with $`\mathrm{\Delta }mO`$ are included. The excluded configurations with $`\mathrm{\Delta }m>O`$ are transformed into a configuration with $`\mathrm{\Delta }m=O`$, by filling in the $`(O+1)`$th row below the highest particle; see Fig. 4. This transformation does not change the growth probabilities considerably. Especially, the upward growth probability would hardly change for large $`O`$. The variable $`P_i(t)`$, where $`i`$ corresponds to a configuration with $`\mathrm{\Delta }m=O`$, actually represents the sum of probabilities of all the configurations with $`\mathrm{\Delta }mO`$, that have the same $`O`$ uppermost rows, rather than represent the probability of the configuration $`i`$ alone. This is analogous to $`\stackrel{~}{P}_2^{}`$ in the example above, see Sec. II B. After the finite set of configurations is chosen, the configurations are indexed with arbitrary consecutive numbers. Then, the growth probabilities for each configuration are computed by solving the Laplace equation and by taking into account the bond multiplicity. Each growth process results in a different final configuration, which must be identified with one of the configurations in the finite set. Special attention is required for the upward growth processes, which might result in configurations with $`\mathrm{\Delta }m>O`$, which do not belong to the finite set. This is rectified by truncating the bottom row of the interface (considering it as fully occupied). The total upward probability for each configuration is added up and stored in a function $`p_{\mathrm{up}}(i)`$, later to be averaged over the steady state distribution of configurations. The growth probabilities are arranged in the evolution matrix, $`𝐄`$, whose fixed point corresponds to the steady state distribution of configurations, which is required for evaluating $`p_{\mathrm{up}}^{}`$, $`\rho `$ and $`D`$. Because the matrix is finite, the existence of at least one fixed point is guaranteed. The other eigenvectors describe the rate of convergence to the steady state.
The best way to demonstrate this approach is by showing a few sample calculations. The easiest ones are the first and second order approximation for $`N=3`$ and the first order approximation for $`N=4`$. After that we explain the general algorithm for higher orders and widths, and report the results obtained numerically.
### A First order approximation for $`N=3`$
In the first order approximation we only look at the top row of the aggregate. For $`N=3`$ there are only $`3`$ possible configurations (up to symmetry), with the top row occupied by $`1`$, $`2`$ or $`3`$ particles. Each configuration is indexed and for each configuration we identify the growth processes and the final configurations resulting from them; see Fig. 5. In part II of this paper we show that the calculation presented in this section can be used to solve exactly (no approximations) the case of site-DLA with $`N=3`$.
The first configuration $`(j=1)`$ grows upward with probability $`1`$, thus $`p_{\mathrm{up}}(1)=1`$. The resulting configuration is $`i=2`$, thus $`E_{2,1}=1`$ and $`E_{i,1}=0`$ for $`i2`$. This concludes the construction of the first column of the evolution matrix.
In order to obtain the other growth probabilities we have to solve the relevant Laplace problems, for which we need the Green’s function according to Eq. (12). For $`N=3`$ we have $`k_l=\frac{2\pi }{3}l`$ for $`l=0,1,2`$. We recall that $`e^{\kappa (k)}=q\sqrt{q^21}`$, where $`q2\mathrm{cos}(k)`$ and find that
$`e^{\kappa _0}=1,`$ (89)
$`e^{\kappa _1}=e^{\kappa _2}={\displaystyle \frac{5\sqrt{21}}{2}},`$ (90)
and thus
$`g_3(0)={\displaystyle \frac{1}{3}}\left(1+2{\displaystyle \frac{5\sqrt{21}}{2}}\right)={\displaystyle \frac{6\sqrt{21}}{3}},`$ (91)
$`g_3(1)=g_3(2)={\displaystyle \frac{1g_3(0)}{2}}={\displaystyle \frac{\sqrt{21}3}{6}}.`$ (92)
These values obey the normalization condition (13).
Because of the symmetry of the configuration $`j=2`$, the potential can be expressed in terms of one variable $`x\mathrm{\Phi }(0,0)=\mathrm{\Phi }(0,2)`$, as shown in Fig. 6. This kind of figure demonstrates the distribution of the potential $`\mathrm{\Phi }(m,n)`$ over the lattice, and thus we call it a “potential diagram”. The potentials $`\mathrm{\Phi }(1,0)=\mathrm{\Phi }(1,2)=1+(1g_3(1))x`$ do not correspond to a growth process, but are important for solving for $`x`$. The potential $`\mathrm{\Phi }(1,1)=1+2xg_3(1)`$ corresponds to the upward growth process. The Laplace equation for $`x`$ is
$`4x`$ $`=`$ $`x+(1g_3(1))x+1,`$ (93)
$`x`$ $`=`$ $`{\displaystyle \frac{9\sqrt{21}}{10}}=0.4417.`$ (94)
Growth in both sites $`(0,0)`$ and $`(0,2)`$ results in configuration $`i=3`$, hence
$$E_{3,2}=\frac{4}{3}x=\frac{182\sqrt{21}}{15}=0.5890,$$
(95)
where the numerator, $`4`$, is inserted because there are $`2`$ bonds for each of the $`2`$ growth sites, and the denominator is the normalization factor $`N=3`$. A growth process in site $`(1,1)`$ results in an interface that does not belong to our finite set. In this approximation we only take into account the top most row of the interface, and therefore this interface is identified with configuration $`i=2`$, i.e.,
$$E_{2,2}=\frac{2xg_3(1)+1}{3}=\frac{2\sqrt{21}3}{15}=0.4110.$$
(96)
The transition to $`i=1`$ is impossible, hence, $`E_{1,2}=0`$. It is easy to check that the second column of the matrix is normalized, i.e., $`_{i=1}^3E_{i,2}=1`$. The total upward growth probability for this configuration is
$$p_{\mathrm{up}}(2)=E_{2,2}=0.4110.$$
(97)
The potentials of configuration $`j=3`$ are described in terms of $`x=\mathrm{\Phi }(0,1)`$, as in Fig. 7. The Laplace equation is
$`4x`$ $`=`$ $`g_3(0)x+1,`$ (98)
$`x`$ $`=`$ $`{\displaystyle \frac{6\sqrt{21}}{5}}=0.2835.`$ (99)
There are $`3`$ bonds leading to growth in site $`(1,0)`$, which results in the configuration $`i=1`$, hence
$$E_{1,3}=\frac{3}{3}x=0.2835.$$
(100)
The upward growth process results in $`i=2`$ after truncation, and has probability
$$p_{\mathrm{up}}(3)=E_{2,3}=\frac{2}{3}(1+g_3(1)x)=0.7165.$$
(101)
The third element in the column is $`E_{3,3}=0`$, which concludes the calculation of the elements of the evolution matrix,
$$𝐄^{(3,1)}=\left[\begin{array}{ccc}0& 0& 0.2835\\ 1& 0.4110& 0.7165\\ 0& 0.5890& 0\end{array}\right],$$
(102)
where the superscript indicates that it is the first-order approximation for $`N=3`$. The upward growth probabilities series is
$$p_{\mathrm{up}}=(1,0.4110,0.7165),$$
(103)
which happens to be equal to the second row of the matrix.
The normalized fixed point of the matrix is $`P_1^{}=0.0951`$, $`P_2^{}=0.5695`$ and $`P_3^{}=0.3354`$. The average upward growth probability is
$$p_{\mathrm{up}}^{}=\underset{i=1}{\overset{3}{}}P_i^{}p_{\mathrm{up}}(i)=0.5695.$$
(104)
We have performed some DLA simulations in the cylindrical geometry for several values of $`N`$ and measured $`p_{\mathrm{up}}^{}`$ . The value obtained from simulations for $`N=3`$ is $`0.5462`$. The typical accuracy is on the order of $`10^4`$. The steady state average density and fractal dimension are evaluated using Eqs. (44) and (46),
$`\rho ={\displaystyle \frac{1}{3p_{\mathrm{up}}^{}}}=0.5853,(0.6103),`$ (105)
$`D=1{\displaystyle \frac{\mathrm{ln}(p_{\mathrm{up}}^{})}{\mathrm{ln}(3)}}=1.5125,(1.5506).`$ (106)
The values in parentheses are obtained from the same formulae, using the simulation value of $`p_{\mathrm{up}}^{}`$. The two other eigenvalues are complex, $`\lambda _{0,1}=0.29\pm 0.28i`$, so according to Eq. (85) $`\tau =1.10`$.
### B Higher-order approximations for $`N=3`$
The possible configurations of the interface in the second-order approximation are listed and indexed in Fig. 8. The growth probabilities for the first $`3`$ configurations were already computed in the previous section, but a rearrangement of the upward growths is required in the evolution matrix. Now, the upward growth from configuration $`j=2`$ no longer stays at $`i=2`$, but rather makes a transition to $`i=4`$, and the upward growth from $`j=3`$ results in $`i=5`$ instead of $`i=2`$. Thus, we copy the previous evolution matrix $`𝐄^{(3,1)}`$ into the upper left corner of the new matrix $`𝐄^{(3,2)}`$ with the replacements: $`E_{2,2}^{(3,2)}=0`$, $`E_{4,2}^{(3,2)}=E_{2,2}^{(3,1)}`$, $`E_{2,3}^{(3,2)}=0`$, and $`E_{5,3}^{(3,2)}=E_{2,3}^{(3,1)}`$. The unspecified elements in the first three columns are all equal to zero.
The next step is to go over each of the remaining configurations $`i=4,\mathrm{},7`$, and compute their probabilities, which are inserted into the evolution matrix according to the final configuration in which the relevant growth process results. Configuration $`4`$ is shown in Fig. 9. The Laplace equation is
$`4y`$ $`=`$ $`y+x,`$ (107)
$`4x`$ $`=`$ $`x+y+1+(1g_3(1))x,`$ (108)
$`x`$ $`=`$ $`{\displaystyle \frac{3}{14}}(7\sqrt{21})=0.5180,`$ (109)
$`y`$ $`=`$ $`x/3=0.1727.`$ (110)
The growth probabilities are
$`E_{6,4}`$ $`=`$ $`{\displaystyle \frac{2}{3}}x=0.3453,`$ (111)
$`E_{5,4}`$ $`=`$ $`{\displaystyle \frac{4}{3}}y={\displaystyle \frac{4}{9}}x=0.2302,`$ (112)
$`E_{4,4}`$ $`=`$ $`{\displaystyle \frac{1+2xg_3(1)}{3}}=0.4244.`$ (113)
The upward growth probability is $`p_{\mathrm{up}}(4)=E_{4,4}=0.4244`$.
Configuration $`5`$ is shown in Fig. 10. The Laplace equations are
$`4y=y/4+x+xg_3(1)+yg_3(0)+1,`$ (114)
$`4x=y+g_3(0)x+g_3(1)y+1,`$ (115)
$``$ (116)
$`y=0.4808,`$ (117)
$`x=0.4557.`$ (118)
The growth probabilities are
$`E_{7,5}`$ $`=`$ $`{\displaystyle \frac{2}{3}}x=0.3038,`$ (119)
$`E_{3,5}`$ $`=`$ $`{\displaystyle \frac{y}{3}}=0.1603,`$ (120)
$`E_{2,5}`$ $`=`$ $`{\displaystyle \frac{3}{3}}y/4=0.1202,`$ (121)
$`E_{4,5}`$ $`=`$ $`{\displaystyle \frac{1+g_3(1)(x+y)}{3}}=0.4157.`$ (122)
The upward growth probability is $`p_{\mathrm{up}}(5)=E_{4,5}=0.4157`$.
Configuration $`6`$ is shown in Fig. 11. The Laplace equations are
$`4y=y/4+x,`$ (123)
$`4x=y+g_3(0)x+1,`$ (124)
$``$ (125)
$`x={\displaystyle \frac{15}{151}}(265\sqrt{21})=0.3067,`$ (126)
$`y={\displaystyle \frac{4}{15}}x=0.0818.`$ (127)
The growth probabilities are
$`E_{1,6}`$ $`=`$ $`{\displaystyle \frac{2}{3}}x=0.2044,`$ (128)
$`E_{3,6}`$ $`=`$ $`{\displaystyle \frac{2}{3}}y={\displaystyle \frac{8}{45}}x=0.0545,`$ (129)
$`E_{7,6}`$ $`=`$ $`{\displaystyle \frac{3}{3}}y/4=x/15=0.0204,`$ (130)
$`E_{5,6}`$ $`=`$ $`{\displaystyle \frac{2}{3}}(1+g_3(1)x)=0.7206.`$ (131)
The upward growth probability is $`p_{\mathrm{up}}(6)=E_{5,6}=0.7206`$.
Configuration $`7`$ is shown in Fig. 12. The Laplace equations are
$`4x=x/4+g_3(0)x+1,`$ (132)
$``$ (133)
$`x={\displaystyle \frac{12}{105}}(214\sqrt{21})=0.3051.`$ (134)
The growth probabilities are
$`E_{1,7}`$ $`=`$ $`{\displaystyle \frac{2}{3}}x=0.2304,`$ (135)
$`E_{3,7}`$ $`=`$ $`{\displaystyle \frac{3}{3}}x/4=0.0763,`$ (136)
$`E_{5,7}`$ $`=`$ $`{\displaystyle \frac{2}{3}}(1+g_3(1)x)=0.7203.`$ (137)
The upward growth probability is $`p_{\mathrm{up}}(7)=E_{5,7}=0.7203`$.
In summary,
$`𝐄^{(3,2)}=`$ (138)
$`\left[\begin{array}{ccccccc}0& 0& 0.2835& 0& 0& 0.2044& 0.2034\\ 1& 0& 0& 0& 0.1202& 0& 0\\ 0& 0.5890& 0& 0& 0.1603& 0.0545& 0.0763\\ 0& 0.4110& 0& 0.4244& 0.4157& 0& 0\\ 0& 0& 0.7165& 0.2302& 0& 0.7206& 0.7203\\ 0& 0& 0& 0.3453& 0& 0& 0\\ 0& 0& 0& 0& 0.3038& 0.0204& 0\end{array}\right],`$ (146)
$`p_{\mathrm{up}}=`$ (148)
$`\left(\begin{array}{ccccccc}1,& 0.4110,& 0.7165,& 0.4244,& 0.4157,& 0.7206,& 0.7203\end{array}\right).`$ (150)
One can check that elements in each column of the matrix sum up to $`1`$. Note that the majority of the elements are null. The normalized fixed point is,
$`𝐏^{}=`$ $`(\begin{array}{cccc}0.0685,& 0.1011,& 0.1145,& 0.2680,\end{array}`$ (153)
$`\begin{array}{ccc}0.2711,& 0.0925,& 0.0843\end{array})`$ (155)
with which we compute some steady state quantities,
$`p_{\mathrm{up}}^{}={\displaystyle \underset{j=1}{\overset{7}{}}}P_j^{}p_{\mathrm{up}}(j)=0.5459,(0.5462),`$ (156)
$`\rho ={\displaystyle \frac{1}{3p_{\mathrm{up}}^{}}}=0.6106,(0.6103),`$ (157)
$`D=1{\displaystyle \frac{\mathrm{ln}(p_{\mathrm{up}}^{})}{\mathrm{ln}(3)}}=1.5510,(1.5506),`$ (158)
where once again, the values from simulation are shown in parentheses. It is apparent that the addition of configurations increases the accuracy of the results. The eigenvalues with the largest absolute values (except for $`1`$) are $`\lambda _{0,1}=0.34\pm 0.40i`$, hence $`\tau =1.6`$.
The third-order approximation yields $`17`$ configurations. The final results are
$`p_{\mathrm{up}}^{}={\displaystyle \underset{j=1}{\overset{17}{}}}P_j^{}p_{\mathrm{up}}(j)=0.5460,(0.5462),`$ (159)
$`\rho ={\displaystyle \frac{1}{3p_{\mathrm{up}}^{}}}=0.6104,(0.6103),`$ (160)
$`D=1{\displaystyle \frac{\mathrm{ln}(p_{\mathrm{up}}^{})}{\mathrm{ln}(3)}}=1.5507,(1.5506).`$ (161)
The eigenvalues with the largest absolute values (except for $`1`$) are $`\lambda _{0,1}=0.34\pm 0.40i`$, hence $`\tau =1.6`$.
It is interesting to inspect the histogram of the distribution of $`p_{\mathrm{up}}(j)`$, illustrated in Fig. 13. One immediately observes that the upward growth probabilities are clustered in three groups: the top one at $`1`$, the second just above $`0.7`$ and the third, just above $`0.4`$. It is easy to check that the top one corresponds to the configuration $`i=1`$, the middle group corresponds to configurations that have two particles at the top row, and the bottom group corresponds to configurations with one particle at the top row. This suggests, that perhaps $`17`$ different configurations are excessive, and the real number of effective configurations is around $`3`$. An interesting question is whether it is possible to further reduce the number of configurations in higher-order approximations by including only “effective” ones.
### C First-order approximation for $`N=4`$
Our last example is the case $`N=4`$, for which we present the first-order calculation. First, we calculate the Green’s function $`g_4(n)`$ according to Eq. (12). For $`N=4`$, there are four possible values for $`k`$ and $`\kappa `$, namely, $`k_l=\frac{2\pi }{N}l=0,\frac{\pi }{2},\pi ,\frac{3}{2}\pi `$, $`e^{\kappa _0}=1`$, $`e^{\kappa _1}=e^{\kappa _3}=2\sqrt{3}`$, and $`e^{\kappa _2}=3\sqrt{8}`$. Hence,
$`g_4(0)`$ $`=`$ $`{\displaystyle \frac{1+2(2\sqrt{3})+3\sqrt{8}}{4}}`$ (162)
$`=`$ $`2{\displaystyle \frac{\sqrt{3}+\sqrt{2}}{2}}=0.4269,`$ (163)
$`g_4(1)`$ $`=`$ $`g_4(3)={\displaystyle \frac{13+\sqrt{8}}{4}}={\displaystyle \frac{\sqrt{2}1}{2}}=0.2071,`$ (164)
$`g_4(2)`$ $`=`$ $`{\displaystyle \frac{12(2\sqrt{3})+3\sqrt{8}}{4}}={\displaystyle \frac{\sqrt{3}\sqrt{2}}{2}}=0.1589.`$ (165)
Once again, Eq. (13) is obeyed.
Figure 14 displays the relevant configurations. Configuration $`j=1`$ grows into configuration $`i=2`$ with probability $`1`$, thus $`E_{2,1}=1`$ and $`E_{i,1}=0`$ for $`i2`$. Also, $`p_{\mathrm{up}}(1)=1`$.
Configuration $`j=2`$ is shown in Fig. 15. The Laplace equations are
$`4x=y+g_4(1)y+\left(g_4\left(0\right)+g_4\left(2\right)\right)x+1,`$ (167)
$`4y=2x+g_4(0)y+2g_4(1)x+1,`$ (168)
$``$ (169)
$`x=0.5148,`$ (170)
$`y=0.6277.`$ (171)
The nonzero growth probabilities in the second column are $`E_{3,2}=\frac{4}{3}x=0.5148`$, $`E_{4,2}=\frac{1}{4}y=0.1569`$, and $`E_{2,2}=\frac{1}{4}\left(1+2g_4\left(1\right)x+g_4\left(2\right)y\right)=0.3283=p_{\mathrm{up}}(2)`$.
Configuration $`j=3`$ is presented in Fig. 16. The Laplace equation is
$`4x`$ $`=`$ $`x+\left(g_4\left(0\right)+g_4\left(1\right)\right)x+1,`$ (172)
$`x`$ $`=`$ $`0.4226.`$ (173)
The nonzero growth probabilities in the third column are $`E_{5,3}=\frac{4}{4}x=0.4226`$ and $`E_{2,3}=\frac{2}{4}\left[1+\left(g_4\left(1\right)+g_4\left(2\right)\right)x\right]=0.5774=p_{\mathrm{up}}(3)`$.
Configuration $`j=4`$ is shown in Fig. 17. The Laplace equation is
$`4x`$ $`=`$ $`\left(g_4\left(0\right)+g_4\left(2\right)\right)x+1,`$ (174)
$`x`$ $`=`$ $`0.2929.`$ (175)
The nonzero growth probabilities in the fourth column are $`E_{5,4}=\frac{6}{4}x=0.4393`$ and $`E_{2,4}=\frac{2}{4}\left(1+2g_4\left(1\right)x\right)=0.5607=p_{\mathrm{up}}(4)`$. Note that this configuration already appeared for $`N=2`$.
The last configuration is shown in Fig. 18. The Laplace equation is
$`4x`$ $`=`$ $`1=g_4(0)x,`$ (176)
$`x`$ $`=`$ $`0.2799.`$ (177)
The nonzero growth probabilities in the fifth column are $`E_{1,5}=\frac{3}{4}x=0.2099`$ and $`E_{2,5}=\frac{1}{4}\left[3+\left(g_4\left(2\right)+2g_4\left(1\right)\right)x\right]=0.7901=p_{\mathrm{up}}(5)`$. This concludes the calculation of the $`5\times 5`$ evolution matrix $`𝐄^{(4,1)}`$.
The steady-state vector is
$$𝐏^{}=\left(\begin{array}{ccccc}0.0298,& 0.4954,& 0.2551,& 0.0777,& 0.1420\end{array}\right).$$
(178)
It enables to calculate the following steady-state quantities:
$`p_{\mathrm{up}}^{}=P_2^{}=0.4954,(0.4657),`$ (179)
$`\rho ={\displaystyle \frac{1}{4p_{\mathrm{up}}^{}}}=0.5046,(0.5368)`$ (180)
$`D=1{\displaystyle \frac{\mathrm{ln}(p_{\mathrm{up}}^{})}{\mathrm{ln}(4)}}=1.5066,(1.5512),`$ (181)
where again, the values in parentheses are from simulation. The eigenvalues with the largest absolute value after $`1`$ are $`\lambda _{0,1}=0.16\pm 0.38i`$, hence $`\tau =1.1`$.
It is also possible to conduct these calculations using different boundary conditions at the bottom; rather than assuming that there is a filled row of occupied sites below the configuration, it is possible to assume that each unoccupied site at the lowest row of the configuration is above an infinite fjord that extends all the way below. The two possibilities are explained in Fig. 19. Performing the calculations with infinite fjords is a bit simpler, because there are less configurations, e.g., the configuration $`i=4`$ would not appear in the first-order approximation for $`N=4`$ .
### D Higher order computations
As one increases $`N`$ and the order of approximation $`O`$, the number of configurations increases exponentially, and it becomes harder to go over all of them manually. However, it is possible to construct a computer algorithm to perform the procedure described here. The main challenges are the automatic configuration recognition and automatic computation of the exact growth probabilities per configuration. In this section we explain the algorithm and report some of the important results.
The algorithm follows the method outlined in the examples of the previous sections, i.e., it goes over all the possible configurations of the interface. In the sample calculations we have initially made a list of all the possible configurations, called the index. Instead of doing this, the program starts with only one configuration, namely the flat one (all the sites of the top row of the aggregate are occupied), which is indexed by $`j=1`$. This configuration grows with probability $`1`$ to a new configuration that has one particle at the top row, while the row below it is fully occupied. This new configuration is inserted into the list of configurations with an index $`j=2`$. Therefore, the program sets $`E_{2,1}=1`$ and $`p_{\mathrm{up}}(1)=1`$. Then the program continues by handling the next configuration in the list, namely $`j=2`$. For each configuration, it solves the Laplace equations and calculates the growth probabilities. Each growth process may create a new configuration. The resulting configuration is first checked for consistency with the desired order $`O`$; configurations which have $`\mathrm{\Delta }m>O`$ are truncated, as in Fig. 4. One then compares each ’new’ configuration with the existing list of configurations. If it does not exist in that list it is added at the end of the list, and indexed consecutively. If the index of the configuration that results from the growth process is $`i`$ and the index of the initial configuration is $`j`$ then the growth probability is inserted into the matrix element $`E_{i,j}`$. The total sum of all the upward growth probabilities of the initial configuration $`j`$ is stored in $`p_{\mathrm{up}}(j)`$. The main loop stops when the program finishes to process the last configuration in the index list. At this stage the Markovian evolution matrix $`𝐄`$ is irreducible and closed, i.e., $`_iE_{i,j}=1`$ for every $`j`$. Then the fixed point $`𝐏^{}`$ is calculated, by taking an initial vector and iterating $`𝐄`$ on it many times until it converges (for very large matrices this is much faster than using any of the MATLAB library functions). The average upward growth probability is calculated using
$$p_{\mathrm{up}}^{}=\underset{j}{}P_j^{}p_{\mathrm{up}}(j),$$
(182)
the average density and the fractal dimension are then computed using the left hand side of Eq. (44) and Eq. (46).
One of the challenges of the computer algorithm is the recognition of configurations. This recognition is important so that each growth process will be inserted into the evolution matrix $`E_{i,j}`$ with the correct index $`i`$ ($`j`$ is the index of the configuration before growth). The recognition maybe difficult because configurations that seem different may actually be equivalent. By equivalent we mean that they have the exact same set of transition (growth) probabilities. The solution to the Laplace equations is determined uniquely by the shape of the interface, therefore all of the configurations with the same external interface are equivalent. The description of the interface is not a trivial task though. We find that an efficient way to characterize an interface is by the set of empty sites that are connected to infinity. Of course, it is sufficient to specify only empty sites that are not higher than the highest particle in the aggregate, because all of the empty sites above it are connected to infinity. Figure 20 shows an example of two configurations that are not identical, but they have the same exterior contour. Both of them have a single empty site that is connected to infinity.
In order to reduce greatly the number of configurations it is advisable to take symmetry into account, i.e., all the configurations which can be obtained from one another using a rotation around the axis of the cylinder have the same growth probabilities and the same steady state weights. The same is true for mirror images. Instead of taking all of them into account, we choose one as a canonical representative of the whole set of symmetric configurations.
The results are summarized in Table I. By comparing the approximations to accurate results from simulations, it seems that in order to obtain a relative accuracy of about $`10^3`$ one has to use at least an order of approximation of $`O=N2`$ (except for $`N=3`$, where one still has to use the second-order approximation). This becomes very difficult already for $`N=6`$, where in the fourth-order calculation there are $`49678`$ different configurations up to symmetry.
## IV Discussion
This paper treats DLA as a Markov process. The Markov states are the possible shapes of the interface, and the Markovian evolution matrix $`𝐄`$ is calculated analytically using exact solutions of the Laplace equations, with proper normalizations. We propose a truncation scheme that takes into account only a finite number of states. The states are ordered according to the maximal difference in height between the highest and lowest points on the interface, $`\mathrm{\Delta }m`$, and in each order of truncation $`O`$, only the states with $`\mathrm{\Delta }mO`$ are included. We justify this approach by the fact that the potential $`\mathrm{\Phi }`$ decays exponentially in deep fjords, and thus the shape of the interface in its deeper parts has very little effect on the growth probabilities. We perform this calculation for $`N=2`$, and verify that indeed it converges to the known analytic solution. We adopt the same approach for higher values of the width $`N`$, between $`3`$ and $`7`$, and calculate the average density $`\rho `$ in good agreement with simulations. The fact that the number of configurations grows exponentially with $`N`$ and with $`O`$, makes the computation less effective than simulation for large $`N`$.
We observe that the method converges as a function of $`O`$, also for higher values of $`N`$. Let us denote the calculated average steady-state density of an aggregate of width $`N`$ in the $`O`$’th-order approximation by $`\rho _c(N,O)`$. We observe that $`\rho _c(N,O)`$ converges to a finite limit very rapidly as a function of $`O`$. In fact, a relative accuracy of $`10^3`$ is achieved for $`O=N2`$ (except for $`N=3`$). This enables us to obtain accurate results for $`3N6`$. The drawback of this method is that the number of configurations diverges exponentially with $`O`$ and $`N`$, and therefore it is possible to perform the calculations only for relatively low $`N`$’s and $`O`$’s. Our computer was strong enough to perform the calculation only in the third-order approximation for $`N=7`$, and therefore the result for $`N=7`$ is not very accurate. One would hope that it may be possible to perform low-order approximations for large $`N`$’s and then extrapolate, in order to estimate the results for large $`O`$’s. Indeed, it is reasonable to conjecture the scaling law $`\rho _c(N,O)=\rho (N)f(N/O)`$, where $`\rho (N)`$ is the exact ($`O\mathrm{}`$) density, as a function of $`N`$, and $`f(N/O)`$ is a universal scaling function that obeys $`lim_{x0}f(x)=1`$. Our investigation shows that in spite of the fact that the conjecture is not very accurate for $`O=1`$ and $`O=2`$, it is quite good for higher values of $`O`$, and presumably also for higher values of $`N`$. This scaling relation may help to perform the extrapolation $`O\mathrm{}`$ for higher values of $`N`$. Paradoxically, it is very hard to obtain data points for large $`N`$’s and $`O`$’s, and thus to extract the scaling function accurately. Thus we are unable to make the extrapolation even for $`N=7`$, and we estimate $`\rho (N)`$ by the highest-order approximation available. However, we suggest an alternative way to obtain $`\rho _c(O,N)`$, namely by simulation: it is possible to perform a regular DLA simulation in cylindrical geometry, only that one has to keep the $`O`$’th row below the highest particle in the aggregate constantly filled. Measuring the average density of the aggregate in such a simulation would approximate $`\rho _c(N,O)`$. This simulation would be faster than a regular simulation, because particles would stick faster, due to the fact that they have less free space. This study would perhaps yield the scaling function $`f(N/O)`$, and enable extrapolation of lower order approximations for higher $`N`$’s, should anyone venture to perform them on more powerful computers. In light of this discussion we suggest a more efficient way to perform DLA simulations in cylindrical geometry. We argue that one can obtain a relative accuracy of $`10^3`$ if one follows just the $`N2`$ top most rows of the aggregate. This should save some time, because the diffusing particle would stick faster, and it would also require less memory. This is not to say that it is sufficient to grow the aggregate until it reaches a height of $`N2`$, but rather, to perform many more growth processes, and each time the aggregate reaches a height of $`N1`$, truncate the bottom row.
We also discuss the temporal rate of convergence of the system to its steady state. In this context we find that there is an exponential convergence to the steady state, and we calculate the characteristic time constant $`\tau `$. This is demonstrated using the simple model of the frustrated climber. The convergence is described in terms of the eigenvalues of the Markovian matrix, and in terms of the infinite shift-down operator.
Considering the fractal dimension, Pietronero et al. suggested that $`\rho (N)=N^{Dd}`$, as mentioned in Eq. (45). In principle, one should always include an amplitude and finite size corrections of the form
$$\rho (N)=AN^\alpha \left(1+B/N+\mathrm{}\right),$$
(183)
where $`\alpha =dD`$, and $`A`$ and $`B`$ are constants. The second term appearing in Eq. (183) is a correction to scaling term. Generally, there is an infinite sum of such terms with higher negative powers of $`N`$. Because we have data only for small values of $`N`$, these correction terms may be large, but since we have only a few accurate data points ($`\rho (N)`$ for $`N=2,3,\mathrm{},6`$), we try to extract the parameters $`\alpha `$, $`A`$ and $`B`$ only, and not higher order terms. Using the three results for $`N=4,5,6`$, we determine the three unkown parameters to be $`A=0.82`$, $`B=0.35`$ and $`\alpha =0.362`$, hence $`D=1.64`$. The deviation from the well know value of $`D=1.66`$ can be attributed to systematic error due to the omission of higher order finite size correction terms. We fit simulation data for $`N=3`$, $`4`$, $`5`$, $`6`$, $`7`$, $`32`$, $`48`$, $`64`$, $`96`$, $`128`$, to a higher-order approximation $`\rho (N)=AN^\alpha \left(1+B/N+C/N^2\right)`$, and find that $`C=0.205`$, $`B=0.561`$, $`A=0.761`$ and $`\alpha =0.339`$, which means that $`D=1.661`$. The maximum relative error of the fit is $`1.2\times 10^3`$, and the average relative error is $`1.0\times 10^3`$, which is in good agreement with estimated accuracy of the simulations.
###### Acknowledgements.
We wish to thank Barak Kol and A. Vespignani for helpful discussions. We also wish to thank Yiftah Navot for helping with the computer program, by suggesting more efficient data structures and algorithms. We thank Nadav Schnerb for offering the frustrated climber metaphor. This work was supported by a grant from the German-Israeli Foundation (GIF).
|
warning/0002/astro-ph0002456.html
|
ar5iv
|
text
|
# ISO-SWS Observations of OMC-1: H2 and Fine Structure Lines Based on observations with ISO, an ESA project with instruments funded by ESA Member States (especially the PI countries: France, Germany, The Netherlands, and the United Kingdom) and with the participation of ISAS and NASA.
## 1 Introduction
The Orion molecular cloud, OMC-1, located behind the Orion M42 Nebula at a distance of $``$450 pc (Genzel & Stutzki gen89 (1989)), is the best-studied massive star forming region. This cloud embeds a spectacular outflow arising from some embedded young stellar object, which can possibly be identified as the radio source “I” 0.49 arcsec south of the infrared source IRc2-A (Menten & Reid men95 (1995); Dougados et al. dou93 (1993)). The outflow shocks the surrounding molecular gas, thereby giving rise to the strongest H<sub>2</sub> infrared line emission appearing in the sky (Fig. 1). Peak 1 (Beckwith et al. bec78 (1978)) is the brighter of the two H<sub>2</sub> emission lobes of the outflow. Although the outflow has been studied extensively for nearly two decades, the nature of the excitation mechanism remains unclear.
Molecular hydrogen, through its infrared rotational and rotation-vibrational transitions, is an important coolant in shocks and photodissociation regions, and thereby a particularly well suited tracer of the flourescently- and/or shock-excited gas. The Short Wavelength Spectrometer (SWS, de Graauw et al. deg96 (1996)) aboard the Infrared Space Observatory (ISO, Kessler et al. kes96 (1996)) offered the first opportunity to observe pure rotational and rotation-vibrational H<sub>2</sub> lines from 2.4 $`\mu `$m to 28 $`\mu `$m with one instrument, unhindered by the Earth’s atmosphere.
In this paper, we present a comprehensive set of intensities for 56 H<sub>2</sub> near- and mid-infrared lines we observed with the ISO-SWS. These observations trace populations of energy levels ranging from $`E/k=`$1015 K to 43 000 K. The redundancy of the H<sub>2</sub> level determinations provides information on the average gas excitation along the line of sight over an unprecedented range. This sheds new light on the possible excitation mechanisms in the IRc2 outflow.
We here concentrate on the interpretation of the $`\mathrm{H}_2`$ and the atomic and ionic fine structure line emission, whereas a detailed discussion of the CO and $`\mathrm{H}_2\mathrm{O}`$ lines will be presented in a separate paper (Boonman et al. boo00 (2000)). In a related paper we already discussed the detection of HD toward Orion Peak 1 (Bertoldi et al. ber99 (1999)).
## 2 Observation
We observed OMC-1 in the SWS 01 ($`2.445\mu \mathrm{m}`$ grating scan) and SWS 07 (Fabry-Pérot) modes of the short wavelength spectrometer (de Graauw et al. deg96 (1996)) on board ISO on October 3, 1997, and in the SWS 02 ($`0.01\lambda `$ range grating scan) mode on September 20, 1997 and February 15, 1998. Fig. 1 illustrates the various aperture orientations with respect to the $`\mathrm{H}_2`$ 1-0 S(1) emission at 2.12 $`\mu \mathrm{m}`$ observed with NICMOS on the HST (Schultz et al. sch99 (1999)).
For the 2.4–45 $`\mu `$m spectrum of Peak 1 we used the slowest speed, highest spectral resolution full scan observing mode of the SWS. Data reduction was carried out using standard Off Line Processing (OLP) routines up to the Standard Processed Data (SPD) stage within the SWS Interactive Analysis (IA) system. Between the SPD and Auto Analysis Result (AAR) stages, a combination of standard OLP and in-house routines were used to extract the individual scans as well as for the removal of fringes. The flux calibration errors range from 5% at 2.4 $`\mu `$m to 30% at 45$`\mu `$m (SWS Instrument & Data Manual, Issue 1.0). The statistical uncertainties derived from the line’s signal to noise ratio are for most detected lines smaller than the systematic errors due to flux calibration uncertainties.
## 3 Results and Discussion
Fig. 2 shows the full SWS 01 spectrum. Most of the observed continuum flux is probably coming from the strong Becklin Neugebauer (BN) source near the edge of the aperture. Aperture size changes from one band to another then cause changes in the intercepted continuum which are not simply proportional to the aperture size. In addition, there is extended continuum emission all over the outflow.
We normalized the line and continuum fluxes by the aperture size, assuming that there is uniform surface brightness at least for the line emission. The exact aperture profiles for the various wavelength bands is yet to be determined. Assuming an effective aperture resembling those shown in Fig. 1 is approximate. The error from this assumption, and the nonuniform continuum surface brightness cause additional relative offsets in the continuum fluxes of neighboring bands of –10% for the 7–12 $`\mu `$m band, $`30\%`$ for the 12–16 $`\mu `$m band, +15% at 16–19.5 $`\mu `$m, -5% at 19.5–27.5 $`\mu `$m, -7.5% at 27.5–29.5 $`\mu `$m, and +5% above 29.5 $`\mu `$m.
Since the observed H<sub>2</sub> line intensities result in a smooth distribution of column densities in the excitation diagram, Fig. 8, the line intensities appear not to be affected much by the uncertainty of the aperture. Fig. 3 shows the SWS 01 spectrum in more detail. Fig. 4 shows selected lines at higher spectral resolution from line scan observation in the SWS 02 and SWS 07/SWS 06 modes. The SWS 06 grating spectra were simultaneously recorded with the SWS 07 Fabry-Perot spectra.
The Peak 1 spectrum is dominated by a large number of rotational and ro-vibrational $`\mathrm{H}_2`$ lines. The pure rotational lines arise from levels with energies ranging from $`E/k=`$1015 K for the 0-0 S(1) line to $`E/k=\mathrm{42\hspace{0.17em}515}`$ K for the 0-0 S(25) line. They represent gas with excitation temperatures ranging from 600 K for the low energy levels to over 3000 K for level energies $`E/k>\mathrm{14\hspace{0.17em}000}`$ K. The observed fluxes of the identified H<sub>2</sub> lines are listed in Table 3. The spectrum is rich also in H i recombination lines and atomic and ionic fine structure lines. Between 4.5 to 5 $`\mu `$m, we find a forest of gaseous 1-0 ro-vibrational CO emission, possibly mixed with absorption of solid CO (van Dishoeck et al. dis98 (1998)). Gaseous water is seen in emission through the $`\nu _2`$ bending mode between 6.3 and 7 $`\mu `$m and several lines between 30 and 45 $`\mu `$m, and gaseous $`\mathrm{CO}_2`$ is detected at 15 $`\mu `$m. PAH features dominate the emission between 6 and 12 $`\mu \mathrm{m}`$. Absorption features of water ice are seen at 3.1 $`\mu `$m, of $`\mathrm{CO}_2`$ ice at 4.25 $`\mu \mathrm{m}`$, and of silicate at 9.7 $`\mu \mathrm{m}`$.
The various observed lines and features probe different regions – both within the SWS aperture and along the line of sight. The H ii region in the foreground of OMC-1 contributes to the H recombination and ionic fine structure emission, whereas the PAH (Verstraete et al. ver96 (1996)) emission and a large fraction of the \[Si ii\]34.8$`\mu \mathrm{m}`$ emission (Haas et al. haa91 (1991)) originate in the PDR between the H ii region and the molecular cloud which embeds OMC-1. The PDR also contributes to the $`\mathrm{H}_2`$ emission. The shocks dominate the emission of H<sub>2</sub>, H<sub>2</sub>O and CO, and may make a minor contribution to the H recombination and most fine structure emission.
### 3.1 Observed H<sub>2</sub> level column densities
All molecular hydrogen lines, due to the small radiative transition probabilities, remain optically thin. Therefore the corresponding “observed” upper level column density can be computed from the observed line flux,
$$N_{\mathrm{obs}}(v,J)=\frac{4\pi \lambda }{hc}\frac{I_{\mathrm{obs}}(v,Jv^{},J^{})}{A(v,Jv^{},J^{})},$$
(1)
where $`I_{\mathrm{obs}}(v,Jv^{},J^{})`$ and $`A(v,Jv^{},J^{})`$ are the observed line flux and the Einstein-$`A`$ radiative transition probability of the transition from level $`(v,J)`$ to $`(v^{},J^{})`$, respectively. The Einstein coefficients are adopted from Turner et al. (tur77 (1977)) and Wolniewicz et al. (wol98 (1998)). The transition energies we computed from level energies kindly provided by E. Roueff (1992, private communication).
A convenient way to visualize the level column densities is to divide them by the level degeneracy $`g_J`$, and plot this against the upper level energy $`E_\mathrm{u}(v,J)/k`$; the degeneracy $`g_Jg_s(2J+1)`$, where $`g_s=3`$ for ortho (odd $`J`$) H<sub>2</sub> and $`g_s=1`$ for para (even $`J`$) H<sub>2</sub>. For the lines we observed toward Peak 1 we found that in such a “Boltzmann diagram” Fig. 5 the level columns show a smooth distribution, where the level columns line up irrespective of their quantum numbers. There is no sign of fluorescent excitation or of a deviation from the ortho-to-para H<sub>2</sub> ratio of three.
Fluorescently excited $`\mathrm{H}_2`$, as seen in photodissociation regions (Timmermann et al. 1996a ) would produce a level distribution in which the “rotational temperature” derived from levels at given $`v`$ is lower than the “vibrational temperature” derived from levels of the same $`J`$ (e.g. Draine & Bertoldi dra96 (1996)). Fluorescent excitation therefore shows a characteristic jigsaw distribution of the $`v>1`$ levels, unlike the smooth line-up we observed here, where $`N/g`$ appears not to depend on the state quantum number. Furthermore, fluorescently excited gas usually shows ortho-to-para ratios in vibrationally excited levels smaller than the total ortho-to-para ratio of the gas along the line of sight. This is due to the enhanced self-shielding, and therefore reduced excitation rate, of the more abundant ortho-H<sub>2</sub> (Sternberg & Neufeld ste99 (1999)).
### 3.2 Extinction
The shocks emitting the strong infrared lines in the OMC-1 outflow are deeply embedded in the molecular cloud, so that the emerging radiation suffers significant extinction. To correct the column densities derived from the H<sub>2</sub> emission line intensities, $`N_{\mathrm{obs}}(v,J)`$, for this extinction, we need to know the proper extinction correction as a function of wavelength. However, this interstellar infrared extinction curve is not well determined. Especially the depth and width of the silicate absorption features, centered at 9.7 $`\mu `$m and 18 $`\mu \mathrm{m}`$ are uncertain, and they could vary from region to region (Draine dra89 (1989)). A further complication arises from the mixing of the emitting and absorbing gas, which we might expect in the outflow regions considering the complex spatial variation of the near-IR emission mapped in OMC-1 (Fig. 1), or in similar outflows such as DR21 and Cep A (Davis & Smith dav96 (1996); Goetz et al. goe98 (1998)).
With enough redundancy in the information provided by the molecular and atomic lines in Peak 1, we are in principle able to estimate the average extinction along our line of sight as a function of wavelength. However, the H i recombination lines and the H<sub>2</sub> emission lines may not be tracing the same regions, and might therefore be subject to differing extinction. We therefore treat them separately. To correct the $`\mathrm{H}_2`$ line fluxes for extinction, we tried to derive the extinction from the $`\mathrm{H}_2`$ lines directly (Bertoldi et al. ber99 (1999)).
An inspection of the excitation diagram Fig. 5 derived from the line intensities (uncorrected for extinction) shows that the column densities follow a smooth distribution, with no dependence on vibrational quantum number, and no sign of an ortho-to-para column density ratio different from three. Transitions from a given state to different lower states produce lines with different wavelengths which suffer extinction. Deviations from the expected line ratios of lines from the same state therefore yield the difference in extinction between the corresponding wavelengths of the lines. More generally, we can use this to estimate the extinction as a function of wavelength for a large set of lines, by minimizing the dispersion in the excitation diagram around a least-squares fit to the level columns. We thereby assume that the dispersion in the column densities is partly due to extinction.
We constructed an extinction curve (Fig. 6) with four free parameters: the absolute normalization for a $`A_\lambda \lambda ^{1.7}`$ power law extinction curve from $`2.4\mu \mathrm{m}`$ to $`6\mu \mathrm{m}`$, the width and depth of the water ice absorption feature at 3.1$`\mu `$m, and the depth of the $`9.7\mu `$m silicate absorption feature. We fixed the width of the 9.7 and $`18\mu \mathrm{m}`$ silicate features from calculations by Draine & Lee (dra84 (1984)). The depth of the 18$`\mu `$m feature was taken to be 0.44 times that of the 9.7$`\mu \mathrm{m}`$ feature, based on an average of previous estimates (Draine & Lee dra84 (1984); Pegourie & Papoular peg85 (1985); Volk & Kwok vol88 (1988); Bertoldi et al. in prep.). Calibration uncertainties and a small contribution of foreground fluorescently excited H<sub>2</sub> may give rise to a dispersion in the derived extinction corrections that is not due to extinction. The derived curve should therefore be considered as very approximate. The extinction curve minimizing the dispersion in the excitation curve is shown in Fig. 6. Explicitly it can be written
$`A(\lambda )`$ $`=`$ $`A_\mathrm{K}(\lambda /2.12)^{1.7}+0.58e^{22(\lambda 3.05)^2}`$
$`+(1.350.08A_\mathrm{K})\{e^{[c_1\mathrm{log}(\lambda /9.66)]^2}`$
$`+0.44e^{[c_2\mathrm{log}(\lambda /19)]^2}\},`$
where $`A_\mathrm{K}=(1.0\pm 0.1)`$ mag is the implied extinction at 2.12 $`\mu `$m, and $`c_1=14.3`$ for $`\lambda <9.7`$, $`c_1=9.8`$ for $`\lambda >9.7`$, $`c_2=7.5`$ for $`\lambda <19`$, $`c_2=4.8`$ for $`\lambda >19`$, and $`\lambda `$ is given here in $`\mu `$m. The depth of the extinction minimum at 6.5 $`\mu `$m is very uncertain, since it is not constrained by the inconsistent corrections derived from the four lines between 5.8 and 7.3 $`\mu `$m. There is an indication that the minimum is at least as deep as our simple curve shows, but a much more careful analysis of the line fluxes would be necessary to reach a firm conclusion.
Atomic hydrogen recombination lines could offer another means to trace the extinction as a function of wavelength. We detected seventeen H i recombination lines ranging in wavelength from 2.6 to 19 $`\mu \mathrm{m}`$. Since the relative emissivities are known from theory (Storey & Hummer sto95 (1995)) and depend only mildly on the gas temperature and density, a comparison of the observed line intensities divided by their respective case B emissivities yields a measure for the differential extinction between the respective lines’ wavelengths.
In Fig. 7 we plot against wavelength the observed line intensities, divided by their emissivities and normalized to this ratio for the H i 8–5 transition line. The data points scatter around unity, which means that there is little if any differential extinction over this wavelength range. A comparison with the distribution of intensities expected for an extinction curve with the shape we found from the H<sub>2</sub> lines reveals none of the prominent extinction features. A reasonable explanation is that the total extinction to the H i emission region is very low, $`A_\mathrm{K}<0.3`$ mag.
Although the errors are too large to constrain the exact value of the extinction, it is obvious that the H recombination lines are much less attenuated than the H<sub>2</sub> lines. This suggests that the bulk of the atomic hydrogen emission arises in the foreground H ii region, whereas the $`\mathrm{H}_2`$ emitting region is more deeply embedded in the molecular cloud.
This conclusion agrees with the assessment of Everett et al. (eve95 (1995)), who obtained $`A_\mathrm{J}=(0.38\pm 0.09)`$ mag for the extinction shown by H recombination lines, but found $`A_\mathrm{J}=(2.15\pm 0.26)`$ mag from the H<sub>2</sub> lines. With a $`\lambda ^{1.7}`$ extinction law this corresponds to $`A_\mathrm{K}=(0.15\pm 0.04)`$ mag and $`A_\mathrm{K}=(0.9\pm 0.1)`$ mag, in good agreement with our results.
### 3.3 Fine structure lines
The observed atomic and ionic fine structure lines are valuable diagnostics. If highly-ionized species are found toward the shocked region – which is well shielded from the ionizing radiation of the Trapezium stars – they would indicate the presence of fast, ionizing J-shocks. It is therefore interesting to disentangle the respective contributions to the fine structure line emission of the foreground H ii region/PDR and the shocked gas of the OMC-1 outflow.
Table 1 lists the observed intensities of a number of fine structure lines we searched for toward Peak 1.
A predominantly ionized medium is traced by species with ionization potentials larger than 13.6 eV. From such ions a number of lines, \[Ar ii\]6.9$`\mu `$m, \[Ar ii\]8.99$`\mu `$m, \[Ne ii\]12.8$`\mu `$m, \[Ne iii\]15.5$`\mu `$m, \[Ne iii\]36$`\mu `$m, \[S iii\]18.7$`\mu `$m, and \[S iv\]10.5$`\mu `$m, are found in our ISO-SWS spectra, and their intensities can be compared with H ii region models such as those computed by Rubin et al. (rub91 (1991)). A comparison of the line intensities and their ratios to a blister H ii region model with a star of $`T_{\mathrm{eff}}=\mathrm{37\hspace{0.17em}000}`$ K and $`\mathrm{log}g=4.0`$, shows good agreement of all line intensities, except for that the models overestimate the \[S iii\]18.7$`\mu `$m/\[S iii\]33.5$`\mu `$m ratio by a factor 1.8. The good agreement indicates that these lines may be predominantly produced in the foreground H ii region, although a shock contribution of up to 30% cannot be excluded.
We can compare the Peak 1 fine structure line emission also with that seen toward the Orion Bar photodissociation region and ionization front, which is also irradiated by the Trapezium stars. We know that here no fast shocks should contribute to the emission, and that the emission should be similar to that coming from the PDR in front of the OMC-1 outflow (Herrmann et al. her97 (1997)). In Table 2 we list line intensities we observed toward two positions on the Orion Bar: toward the ionization front at $`5^\mathrm{h}35^\mathrm{m}19.31^\mathrm{s}`$, $`5\mathrm{°}24\mathrm{}59.9\mathrm{}`$ (J2000), and toward the peak of the H<sub>2</sub> 1-0 S(1) emission at $`5^\mathrm{h}35^\mathrm{m}20.31^\mathrm{s}`$, $`5\mathrm{°}25\mathrm{}19.9\mathrm{}`$. A comparison with the Peak 1 intensities shows that the intensities of most lines agree within a factor of a few, suggesting that ionic emission indeed arises in the foreground H ii region.
The \[P iii\]17.9$`\mu `$m, \[Fe iii\]22.9$`\mu `$m, \[Fe ii\]26$`\mu `$m, and the \[S iii\]33.5$`\mu `$m lines were not included in the Rubin models, but their intensities are very similar toward the outflow and the Bar. It is unclear where the \[Fe ii\]26$`\mu `$m emission comes from, though. It could be produced either in the PDR or in the ionization front. A detailed analysis of the \[Fe ii\] emission will be subject of a subsequent publication (Bertoldi et al., in prep.).
##### Silicon:
Haas et al. (haa91 (1991)) observed \[Si ii\]34.8$`\mu `$m strip maps across the OMC-1 outflow. From the apparent peak of emission near IRc2 they concluded that about half of this emission must be due to the production and excitation of gas phase silicon in shocks. In their preliminary reduction of a $`6\mathrm{}`$ square map of \[Si ii\], Stacey et al. (sta95 (1995)) also find that the emission peaks toward the OMC-1 outflow, and this excess is consistent with that observed by Haas et al. (haa91 (1991)). Haas et al. find a surface flux density of $`6\times 10^3\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1\mathrm{sr}^1`$ toward Peak 1, of which they attribute about half to an extended component, which most likely arises from the PDR lining the foreground H ii region. Haas et al. find that the flux is similar toward Peak 1 and the Orion Bar, which may well be due to limb brightening by a factor two of the PDR component at the Bar. Our observations of \[Si ii\]34.8$`\mu `$m toward the Bar and Peak 1 yield fluxes twice as high as those of Haas. This could be due either to a calibration error, or to beam dilution in the Haas et al. measurements. Either way, it seems that both the PDR and the shocks give rise to strong silicon emission, which for the PDR at least requires Si gas phase abundances of order 10% solar: the PDR models by Tielens & Hollenbach (1985b ) adopt a 2.2% solar gas phase Si abundance and predict about a quarter of the flux we could attribute to the PDR. The silicon abundance can be enhanced in shocks by sputtering (Martin-Pintado et al. mar92 (1992); Caselli et al. cas97 (1997); Bachiller & Perez-Gutierrez bac97 (1997)). Large gas phase silicon abundances are also found in other PDRs such as NGC 7023 (Fuente et al. fue99 (1999)). The mechanism by which the abundance is enhanced in PDRs is still unclear, although photodesorption has been suggested (Walmsley et al. wal99 (1999)). Strong silicate emission is however not a universal feature of PDRs: based on ISO observations of \[Si ii\]34.8$`\mu `$m toward NGC 2023 and a comparison with model calculations, Draine & Bertoldi (2000) report Si to be quite highly depleted in the NGC 2023 PDR.
##### Other lines:
Of the other fine structure lines seen toward Peak 1, \[Ni ii\]6.6$`\mu `$m and \[S i\]25$`\mu `$m are not detected toward the Orion Bar. The \[Ni ii\]6.6$`\mu `$m line is confused with a water line, making it difficult to detect. Although the \[Fe ii\]18$`\mu `$m and \[Fe ii\]36$`\mu `$m lines are marginally detected in both objects, they both appear to be an order of magnitude fainter toward the Bar than toward Peak 1. This suggests that shocks are more efficient in producing and exciting gas phase iron.
##### Sulfur:
The strong \[S i\]25$`\mu `$m line emission is probably shock-excited. Burton et al. (1990a ) computed the \[S i\] intensity in their PDR model for densities of $`10^3`$ to $`10^5\mathrm{cm}^3`$ and radiation fields of $`10^3`$ to $`10^5`$ times the ambient interstellar field as $`10^5\mathrm{erg}\mathrm{s}^1\mathrm{cm}^2\mathrm{sr}^1`$, which is three orders of magnitude below our observed \[S i\] intensity. Both a J- or C-type shock could account for the \[S i\] emission. But only a J-shock is able to produce both the \[S i\] and the \[Si ii\] line emission.
We compared the estimated shock contribution to the observed \[Si ii\]34.8$`\mu `$m flux of $`7\times 10^3`$ $`\mathrm{erg}\mathrm{s}^1\mathrm{cm}^2\mathrm{sr}^1`$ (Haas et al. haa91 (1991)) and the \[S i\]25$`\mu `$m line flux to the J-shock model of Hollenbach & McKee (hol89 (1989)). Both the relative and absolute \[S i\] and \[Si ii\] fluxes could be explained by shocks of high velocities, $`v_\mathrm{s}=(85\pm 10)\mathrm{km}\mathrm{s}^1`$, a pre-shock hydrogen nuclei density $`n_\mathrm{H}=(10^510^6)\mathrm{cm}^3`$, and a beam filling factor $`\varphi 34`$. A beam-filling planar shock results in $`\varphi =1`$, and a beam-filling spherical shock in $`\varphi =4`$. A shock contribution of 10 to 30% to the \[Ne ii\]12.8$`\mu `$m flux would also explain the observed \[Ne ii\]/\[Si ii\] and \[Ne ii\]/\[S i\] flux ratios.
### 3.4 Molecular hydrogen
In the spectra shown in Figs. 2, 3, and 4, we detected 56 different $`\mathrm{H}_2`$ lines of pure rotational and rotation-vibrational transitions (Table 3). Pure rotational lines were detected ranging from the 0-0 S(1) to 0-0 S(25) transitions, which correspond to upper level energies $`E(v,J)/k`$ ranging from 1015 K to $`\mathrm{42\hspace{0.17em}500}`$ K. Adding a large number of vibration-rotational transition lines, we are able to study the excitation of the gas within the ISO aperture over an unprecedented range.
The H<sub>2</sub> 0-0 S(0) transition line was not detected from our observations with the medium resolution grating modes (SWS 01 and SWS 02, $`R10002000`$). Unfortunately, our observation with the Fabry-Perot did not cover a spectral range wide enough to detect a line with the expected width of $`60\mathrm{km}\mathrm{s}^1`$ (Nadeau & Geballe nad79 (1979); Brand et al. 1989b , Moorhouse et al. moo90 (1990); Chrysostomou et al. chr97 (1997)). However, the FP spectrum, shown in Fig. 4, shows a line-like feature with a a narrow width of 12 km s<sup>-1</sup>, comparable to the spectral resolution in this observing mode. This feature could be emission arising in the foreground photodissociation region bounding the Orion Nebula and the dense molecular cloud embedding the outflow.
#### 3.4.1 Contribution from the foreground PDR
The line emission toward Peak 1 must include some contribution from the photodissociation region bordering the foreground Orion Nebula H ii region. Garden (gar86 (1986)) produced a $`\mathrm{H}_2`$ 1-0 S(1) map which covers OMC-1, the Trapezium, and the Orion Bar PDR. Following Burton & Puxley (1990b ), we estimate that the extended fluorescent $`\mathrm{H}_2`$ flux should amount to about 5% of the total $`\mathrm{H}_2`$ emission toward Peak 1 over the SWS aperture.
For an additional estimate of the expected PDR contribution we can compare the total H<sub>2</sub> luminosity toward Peak 1 to that toward the Orion Bar, a PDR observed nearly edge-on south-east of the Trapezium. We did observe the Bar with the ISO-SWS (Bertoldi et al., in prep.), and find that the total H<sub>2</sub> emission here amounts to 0.008 $`\mathrm{erg}\mathrm{s}^1\mathrm{cm}^2\mathrm{sr}^1`$, compared to the 0.28 $`\mathrm{erg}\mathrm{s}^1\mathrm{cm}^2\mathrm{sr}^1`$ toward Peak 1. Since the Bar is the brightest PDR emission peak in the Orion Nebula, we see that the $`\mathrm{H}_2`$ emission from the PDR toward Peak 1 is probably small compared with the emission arising from the deeply embedded outflow.
#### 3.4.2 Excitation of molecular hydrogen
From the line intensities we derived observed column densities of the levels from which these transitions arise (Eq.1). We correct these values for extinction with the curve we derived in Sec. 3.2 (Fig. 6), to obtain the inherent level column densities
$$N(v,J)=N_{\mathrm{obs}}(v,J)10^{0.4A(\lambda )}.$$
(3)
The resulting excitation (Boltzmann) diagram is shown in Fig. 8.
The lack of signs of fluorescent excitation in the level columns suggest that the molecules might be mostly thermally excited. An H<sub>2</sub> column $`N_{\mathrm{H}_2,\mathrm{tot}}`$ in statistical (thermodynamic) equilibrium at a single kinetic temperature $`T`$ would yield a level distribution
$$\frac{N(v,J)}{g_J}=N_{\mathrm{H}_2,\mathrm{tot}}\frac{e^{E(v,J)/kT}}{_{v^{},J^{}}g_J^{}e^{E(v^{},J^{})/kT}},$$
(4)
which produces a straight line in the Boltzmann (excitation) diagram Fig. 8. An excitation temperature function, $`T_{\mathrm{ex}}(E)`$, can be assigned to the level distributions at each level energy, $`E(v,J)`$, by computing the inverse of the derivative of the line which best fits $`\mathrm{ln}[N(v,J)/g_J]`$ as a function of $`E(v,J)/k`$. Near the lowest energy levels, $`T_{\mathrm{ex}}600`$ K, whereas at $`E(v,J)/k\mathrm{14\hspace{0.17em}000}`$ K, the excitation temperature rises to $`3200`$ K.
To describe the range of excitation temperatures, we decomposed the distribution of column densities to a sum of five Boltzmann distributions of different excitation temperatures:
$$N(v,J)/g_J=\underset{i=1}{\overset{5}{}}C_ie^{E(v,J)/kT_{\mathrm{ex},i}},$$
(5)
where we chose $`T_{\mathrm{ex},i}=(628,800,1200,1800,3226)`$ K, and the $`C_i`$ (see Table 4) were determined by a least-squares-fit to the observed level columns. In Fig. 8 the dotted line shows the five-component fit. From this fit we can also compute the total warm $`\mathrm{H}_2`$ column density, by summing the column densities over all levels following the interpolated level column distribution:
$`N_{\mathrm{H}_2,\mathrm{tot}}`$ $`=`$ $`{\displaystyle \underset{v,J}{}}\left[{\displaystyle \frac{N(v,J)}{g_J}}\right]g_J`$ (6)
$`=`$ $`{\displaystyle \underset{v,J}{}}{\displaystyle \underset{i=1}{\overset{5}{}}}g_JC_ie^{E(v,J)/kT_{\mathrm{ex},i}}`$
$`=`$ $`(1.9\pm 0.5)\times 10^{21}\mathrm{cm}^2.`$
Adopting a distance of 450 pc (Genzel & Stutzki gen89 (1989)), this column corresponds to a warm H<sub>2</sub> mass of $`(0.06\pm 0.015)\mathrm{M}_{\mathrm{}}`$ within the ISO-SWS aperture.
By summing from $`J=0`$, we extrapolated the observed H<sub>2</sub> $`(v=0,J3)`$ level populations to the unobserved $`(v=0,J=0,1,2)`$ levels. Note that thereby we estimate the total warm $`\mathrm{H}_2`$ column density, but we do not account for the total $`\mathrm{H}_2`$ column along the line of sight, which includes an additional $`10^{22}\mathrm{cm}^2`$ cold gas from the molecular cloud which embeds the outflow. Most of this cold $`\mathrm{H}_2`$ resides in the ground states $`J=0`$ and $`J=1`$, and does not contribute to the emission observed from the shock-excited gas in the outflow.
By changing the order of summation in Eq. 6 we can compute the column densities corresponding to the five excitation temperature components, $`N_{\mathrm{H}_2,i}`$ (Table 4), such that
$$N_{\mathrm{H}_2,\mathrm{tot}}=\underset{i=1}{\overset{5}{}}\underset{v,J}{}g(J)C_ie^{E(v,J)/kT_{\mathrm{ex},i}}=\underset{i=1}{\overset{5}{}}N_{\mathrm{H}_2,i}.$$
(7)
Figure 9 shows the corresponding cummulative column density, $`N_{\mathrm{H}_2}(T_{ex}>T)`$, plotted against $`T`$.
With the interpolated excitation distribution Eq. 5 the column densities of all $`\mathrm{H}_2`$ energy levels can be estimated, even those from which no lines were observed. Then the total H<sub>2</sub> rovibrational emission from the electronic ground state extrapolates to $`(0.28\pm 0.08)`$ erg s<sup>-1</sup>cm<sup>-2</sup>sr<sup>-1</sup>. Over the ISO-SWS aperture this amounts to $`(17\pm 5)\mathrm{L}_{\mathrm{}}`$. Compared with the total observed $`\mathrm{H}_2`$ line emission (after extinction correction) of $`(0.16\pm 0.05)`$ erg s<sup>-1</sup>cm<sup>-2</sup>sr<sup>-1</sup>, we find that our line spectra account for more than half of the total $`\mathrm{H}_2`$ emission.
Our observations target the brightest field in the Orion outflow. The outflow covers an area of about $`2\mathrm{}\times 2\mathrm{}`$. The average H<sub>2</sub> brightness over this area we estimate from the 1-0 S(1) map of Garden (gar86 (1986)) to approximately 20% of that in our observed field, so that the total $`\mathrm{H}_2`$ luminosity of the OMC-1 outflow is estimated to be $`(120\pm 60)\mathrm{L}_{\mathrm{}}`$. This is consistent with the 94 $`\mathrm{L}_{\mathrm{}}`$ estimated by Burton & Puxley (1990b ).
#### 3.4.3 What excites the highest-energy levels?
Table 4 and Fig. 9 illustrate that only a small fraction of the warm molecular gas is at the high excitation temperatures, which reach 3000 K. This is difficult to reconcile with the expected smooth temperature profile of a single planar C-type shock, in which the gas temperature changes smoothly, and where a large fraction of the warm gas is near the maximum temperature (Timmermann 1996b ). Even with a distribution of shock speeds, and a correspondingly wide range in peak temperatures, an excitation temperature distribution similar to that shown in Fig. 9 is difficult to understand. It would require a velocity distribution where only a small fraction, about 1%, of the gas is shocked at the high speed necessary to produce a 3000 K excitation. In bow shocks, e.g., the velocity changes slowly with distance from the apex, and such a distribution of velocities would not be expected.
In dissociative J-type shocks, the molecules are destroyed in the shock, and they reform in a postshock layer where the temperature has dropped much below 3000 K, somewhat dependent on the H<sub>2</sub> formation rate efficiency at higher temperatures, which is essentially unknown (e.g. Bertoldi ber97 (1997)). Dissociative J-shocks can therefore not account for the high excitation H<sub>2</sub> we observe.
Even if temperatures of 3000 K or more can be reached in non-dissociative shocks, the higher H<sub>2</sub> levels would remain subthermally excited unless the gas density is high enough that the collisional excitation and deexcitation rates are comparable to those for radiative decay. A “critical” gas density can be defined for a given level as that for which the total collisional deexcitation rate of this level equals its total radiative decay rate. In Fig. 10 we plot the critical density computed this way for states in the vibrational ground state, $`v=0`$, up to $`J=16`$. We see that even at kinetic temperatures of 3000 K, gas densities above $`10^6\mathrm{cm}^3`$ would be necessary to maintain the high $`v=0`$ levels at populations resembling LTE. Since such high densities may not prevail in the shocked gas of the Orion outflow, we may explore mechanisms other than thermal excitation that could account for the population of the higher energy states (see also Bertoldi et al. ber00 (2000)).
##### Time-dependent C-shocks:
When a high velocity outflow strikes dense molecular gas and thereby a C-type shock is first established, J-type shocks can temporarily form within the C-shock. In such an embedded J-shock, a small column of gas is heated to high temperatures (Chièze et al. chi98 (1998)), and if the density is sufficiently high, this could account for the high-excitation tail of the column density distribution (Flower & Pineau des Forêts flo99 (1999)). The lifetime of the embedded J-shock is small, so that the high-temperature excitation tail would be a transient phenomenon, unless shocks are constantly reforming. Embedded J-shocks may also form when a C-shock encounters dense clumps.
##### Formation pumping:
Another possible pumping mechanism of the high-energy states is the formation of H<sub>2</sub>. Molecular hydrogen is believed to form on the surfaces of dust grains. Some of the 4.5 eV released during the formation of an H<sub>2</sub> molecule is used up to leave the grain, and the remainder is split between translation, rotation, and vibration of the new molecule. The exact level distribution of newly formed H<sub>2</sub> is yet unknown, but it could very well contribute to the observed excitation at intermediate energies, $`E13`$ eV (Black & van Dishoeck bla87 (1987); Le Bourlot et al. leb95 (1995)).
Using Eq. 6 we can sum up the column densities of all levels with energy $`E/k\mathrm{10\hspace{0.17em}000}`$ K, to find a column density $`1.30\times 10^{18}`$ cm<sup>-2</sup>, a fraction $`6.8\times 10^4`$ of the total warm H<sub>2</sub> column. Could H<sub>2</sub> formation in a steady state produce such a fraction of molecules in highly excited states? The pumping rate due to formation pumping is equal to the H<sub>2</sub> formation rate, $`n(\mathrm{H})n_\mathrm{H}R_{\mathrm{gr}}`$, where $`R_{\mathrm{gr}}5\times 10^{17}`$ cm<sup>3</sup> s<sup>-1</sup> is the H<sub>2</sub> formation rate coefficient per hydrogen nucleus. We estimate the radiative decay rate by starting with the characteristic radiative lifetime of $`10^6`$ sec for a molecule in a vibrational level $`v5`$, and note that $`5`$ jumps may be required to reach the ground state, so that the effective $`A`$-coefficient $`A_\mathrm{x}2\times 10^7s^1`$. The population balance for the excited states then writes
$$R_{\mathrm{gr}}n_\mathrm{H}n(\mathrm{H})=n_\mathrm{x}(\mathrm{H}_2)A_\mathrm{x},$$
(8)
which yields an excited H<sub>2</sub> fraction
$`{\displaystyle \frac{n_\mathrm{x}(\mathrm{H}_2)}{n(\mathrm{H}_2)}}`$ $`=`$ $`{\displaystyle \frac{n(\mathrm{H})}{n(\mathrm{H}_2)}}{\displaystyle \frac{n_\mathrm{H}R_{\mathrm{gr}}}{A_\mathrm{x}}}`$ (9)
$`=`$ $`5\times 10^4\left({\displaystyle \frac{n_\mathrm{H}}{10^6\mathrm{cm}^3}}{\displaystyle \frac{n(\mathrm{H})}{2n(\mathrm{H}_2)}}\right),`$
which would be consistent with the observed value, if the term in brackets assumes a value of order unity. This simple estimate thus shows that H<sub>2</sub> formation could account for some of the high excitation level populations if the density is high, the atomic fraction not small, and the formation rate coefficient in the warm shocked gas is somewhat higher than the value implied at $`100`$ K from Copernicus observations, which is $`R_{\mathrm{gr}}3\times 10^{17}`$ cm<sup>3</sup> s<sup>-1</sup> (Jura 1975).
To illustrate the possible importance of H<sub>2</sub> formation for the high-excitation level pumping we show that a simple superposition of two gas layers with hydrogen nuclei density $`n_\mathrm{H}=10^6\mathrm{cm}^3`$, atomic fraction $`n(\mathrm{H})/n_\mathrm{H}=0.5`$, and respective temperatures of 200 K and 800 K, with column densities $`N_{\mathrm{H}_2}=1.2\times 10^{22}\mathrm{cm}^2`$ and $`1.2\times 10^{21}\mathrm{cm}^2`$, can in fact reproduce the observed level column distribution better than any shock model currently available. We used the photodissociation front code of Draine & Bertoldi (1996), but without UV illumination, to compute the non-LTE level distributions for gas at a fixed temperature, density, and molecular fraction. We include H<sub>2</sub> formation with a rate coefficient $`R_{\mathrm{gr}}=5\times 10^{17}\mathrm{cm}^3\mathrm{s}^1`$ and assume a level distribution for the newly formed H<sub>2</sub> following $`N(v,J)(2J+1)e^{E(v,J)/kT_{form}}`$, with a “formation temperature” $`T_{form}=5000`$ K chosen to match the slope of the observed high-excitation level distribution.
Fig. 11 illustrates how the newly-formed H<sub>2</sub> molecules give rise to a high-excitation tail in the levels’ column density distribution. The remaining gas displays a thermal distribution at least up to the levels which are mainly populated by H<sub>2</sub> formation.
##### Non-thermal collisions:
An even more important pumping mechanism for the high-excitation levels may be non-thermal collisions between molecules and ions in a magnetic shock. In magnetic C-type shocks, which are believed to be responsible for most of the emission in Peak 1, the gas is accelerated through fast inelastic collisions. In a magnetic precursor the ions, which are tied to the magnetic field, collide with the undisturbed pre-shock gas at relative velocities comparable to the shock speed. Such non-thermal ion–molecule collisions lead to the acceleration of the molecules and to their internal excitation. High-velocity molecules subsequently collide with other molecules, leading to a cascade of collisions during which the relative kinetic energy is in part converted to internal excitation of the molecules (O’Brien & Drury bri96 (1996)). In sufficiently fast C-shocks, the ion–H<sub>2</sub> and H<sub>2</sub>–H<sub>2</sub> collisions can even lead to a significant collisional dissociation rate. The molecules dissociated in a steady-state, partially dissociative shock reform further downstream, so that across such a shock the H<sub>2</sub> dissociation rate equals the H<sub>2</sub> reformation rate. For every collisionally dissociated molecule there will be a larger number of inelastic collisions which did not lead to dissociation, but to the excitation of the molecule into high ro-vibrational states, up to the dissociation limit. The high-excitation H<sub>2</sub> level column densities thereby created should therefore be larger than those caused by H<sub>2</sub> formation alone.
Note that such energetic collision between ions and H<sub>2</sub> in C-shocks are relatively infrequent because ions are rare – thus the excited H<sub>2</sub> has time enough to cascade to lower levels between collisions, giving rise to line emission from the highly excited levels. In C-type shocks, however, dissociations take place too quickly for highly excited H<sub>2</sub> to radiatively decay.
We conclude that non-thermal collisions in partially dissociative C-shocks could pump the high-excitation states in the H<sub>2</sub> electronic ground state to the levels observed. However, no detailed shock models are available yet which account for this process.
##### 0 – 0 S(25):
The $`J=27`$ level observed through the $`00`$ S(25) line appears overpopulated by a factor of seven over what would be expected from the least-squares fit of the data shown in Fig. 8. The $`J=27`$ level is 3.6 eV above ground and only 0.9 eV from the dissociation limit. H<sub>2</sub> molecules which are newly formed on grains are unlikely to be able to populate states so high, because some fraction of the formation energy is lost to overcome the grain surface potential, and some goes to kinetic and vibrational excitation. Unless we misidentified the $`00`$ S(25) line, it appears that a different mechanism may be populating this level and possibly other high levels. The gas-phase formation of H<sub>2</sub> via H<sup>-</sup>, e.g., might be able to leave the new molecule in such a high rotational state (Bieniek & Dalgarno bie79 (1979); Black et al. bla81 (1981); Launay et al. lau91 (1991)).
#### 3.4.4 Comparison with shock models
For over 20 years, evidence accumulated that the $`\mathrm{H}_2`$ emission from OMC-1 may arise from shocks (Gautier et al. gau76 (1976); Kwan & Scoville kwa76 (1976)). However, the physical nature of these shocks remains unclear. Models for planar J-type (Hollenbach & Shull hol77 (1977); Kwan kwa77 (1977); London et al. lon77 (1977)) or C-type (Draine dra80 (1980); Draine & Roberge dra82 (1982); Chernoff et al. che82 (1982); Draine et al. dra83 (1983)) shocks were unable to reproduce the observed wide velocity profiles (Nadeau & Geballe nad79 (1979); Brand et al. 1989b ; Moorhouse et al. moo90 (1990); Chrysostomou et al. chr97 (1997)), or the wide range of excitation conditions observed. Bow shocks were suggested to account for the observed range of excitation conditions and the wide velocity profiles (Hartigan et al. har87 (1987); e. g. Smith et al. 1991a ; 1991b ), but it remains unclear whether these are predominantly C-type, J-type, or a combination of those.
##### Planar shock models:
In Sect. 3.3 we suggested that some fraction of the fine structure line emission may arise from dissociative J-shocks with velocities of about 85 km s<sup>-1</sup>, pre-shock densities $`n_\mathrm{H}10^510^6\mathrm{cm}^3`$, and a beam filling factor of 3–4 (Hollenbach & McKee hol89 (1989)). In such dissociative shocks most of the excitation of the $`v=0`$, $`J5`$ levels is collisional, and the emission arises in the H<sub>2</sub> reformation region where the temperature levels at 400 to 500 K, which is somewhat below the observed excitation temperature of the lowest levels. The higher levels would then be predominantly pumped by newly formed molecules. Such a model however neither fits the low excitation nor the higher excitation level populations very well. The deficits of the J-shock model could be compensated if we combined it with a C-shock model, e.g., one of Kaufman & Neufeld (kau96 (1996)) with $`v_\mathrm{s}=25`$ $`\mathrm{km}\mathrm{s}^1`$, $`n_{\mathrm{H}_2}=10^510^6\mathrm{cm}^3`$, and a beam filling factor 0.3. Such a combined model provides a good fit to the $`v=0`$, $`J=3`$ to 9 level populations, although higher rotational level populations are predicted too large (see Fig. 12: HK).
A combination of J-type and C-type shocks would be consistent with the picture proposed by Chernoff et al. (che82 (1982)), who suggested that a high velocity, $`110\mathrm{km}\mathrm{s}^1`$, wind emanating from an object near IRc2 drives a $`30\mathrm{km}\mathrm{s}^1`$ expanding shell of swept-up material. The low beam filling factor of the C-shock emission could be due to the clumping of the ambient medium.
The $`\mathrm{H}_2`$ level populations implied by previous ground-based observations (e.g. Brand et al. bra88 (1988); Parmar et al. par94 (1994); Burton & Haas bur97 (1997)) were attempted to match with an empirical planar J-shock “cooling flow” model (Brand et al. bra88 (1988); Chang & Martin cha91 (1991); Burton & Haas bur97 (1997)), which assumes that the cooling is dominated by $`\mathrm{H}_2`$, and cooling by other molecules such as H<sub>2</sub>O and CO may be neglected. Such a model can match the medium and high-excitation level populations, although it somewhat overestimates the population of lower rotational levels. These models assume LTE level distributions, which as we argued above, may not be a valid assumption for the high-excitation levels if the gas density is below $`10^6\mathrm{cm}^3`$. Furthermore, theoretical chemical studies show that most oxygen not locked in CO is converted to $`\mathrm{H}_2\mathrm{O}`$ (Draine et al. dra83 (1983); Kaufman & Neufeld kau96 (1996)), which is detected by ISO-LWS observations toward OMC-1 (Harwit et al. har98 (1998); Cernicharo et al. cer99 (1999)). Both H<sub>2</sub>O and CO should therefore be significant coolants, and the neglect of this in these models is worrisome.
Currently available more realistic single shock models do not seem to fit the observed H<sub>2</sub> level distribution. It appears necessary to combine at least two shock models, one to account for the high-excitation level populations, one for the low-excitation levels. For example, combining two models from Kaufman & Neufeld (kau96 (1996)), with shock velocities of 20 and 40 km s<sup>-1</sup>, and beam filling factors of 1 and 0.026, respectively, can well match the level population up to $`E/k\mathrm{20\hspace{0.17em}000}`$ K (Fig. 8, KN96); a pre-shock H<sub>2</sub> number density of $`3\times 10^5\mathrm{cm}^3`$ was adopted. The Kaufman & Neufeld models however do not account for time-dependency, formation pumping, or non-thermal excitation.
##### H<sub>2</sub> velocity dispersion:
Optical and near-IR observations with high spectral resolution toward the OMC-1 outflow show that typical FWHM widths of the H<sub>2</sub> lines are 50–60 km s<sup>-1</sup> (Nadeau & Geballe nad79 (1979); Moorhouse et al. moo90 (1990); Geballe & Garden geb87 (1987); Chrysostomou et al. chr97 (1997)), and that the line wings can extend to several hundred km s<sup>-1</sup> (Ramsey-Howat et al., in prep.). Molecular hydrogen is expected to be destroyed in shocks with velocities larger than 30 to 50 km s<sup>-1</sup>, depending on the magnetic field strength. It is therefore puzzeling how the H<sub>2</sub> emission can show such large velocity dispersions, even in filaments which are only several arcseconds in size.
##### Bow-shocks:
It has been suggested that the H<sub>2</sub> emission arises in bow-shocks, in which the effective shock velocity decreases from the tip to the wake. The shock speed at the apex may be high enough to produce a dissociative J-shock here. But further down the wake, non-dissociative C-type shocks can prevail, with peak temperatures in the shocked molecular layers that decrease steadily down the wake. Thereby a large range of temperatures for the molecular gas exists in a single bow shock. This could account for the observed level excitation, and may also explain the observed constancy of $`\mathrm{H}_2`$ excitation over the entire OMC-1 outflow (Brand et al. 1989a ).
The existence of bow-shocks is also supported by the observation of double-peaked velocity profiles for isolated regions in the outflow (Chrysostomou et al. chr97 (1997)), and of knots of \[Fe ii\] emission which coincide with “fingers” of $`\mathrm{H}_2`$ emission. Allen & Burton (all93 (1993)) suggest that the \[Fe ii\] and $`\mathrm{H}_2`$ emission trace tips and wakes of bow shocks formed in a stellar outflow.
Recent observations of \[Fe ii\] and $`\mathrm{H}_2`$ 1-0 S(1) velocity profiles (Tedds et al. ted99 (1999)) however question the bow shock picture. Alternatively, Stone et al. (sto95b (1995)) proposed that the bows result from Rayleigh-Taylor instabilities when a poorly collimated outflow accelerates in an ambient medium of decreasing density, or when catching up with a slower shock.
It is difficult to understand how the high velocity excited H<sub>2</sub> can be produced in a bow shock which is produced, e.g., by a dense bullet which is moving through a medium initially at rest. Ambient gas which has passed through parts of the bow shock which are not strong enough to dissociate the H<sub>2</sub> will not be accelerated to velocities much larger than 30 km s<sup>-1</sup>, unless the magnetic field is very strong. If alternatively the bow shock arises from a molecular wind impinging on a dense obstacle, then the problem arises how the molecular wind was accelerated to over 100 km s<sup>-1</sup> without destroying the molecules, and why we do not see a lot more mass, traced by CO, e.g., at such high velocities.
Smith et al. (1991a ) are able to reproduce the shape and width of the observed H<sub>2</sub> lines in the Orion outflow with bow shock models, but only by assuming a magnetic field strength of 50 mG, significantly higher than the 10 mG implied by polarization studies (Chrysostomou et al. chr94 (1994)).
In Fig. 12, we compare our data with a bow shock model by Smith (1991b ), adopting a peak shock velocity of $`100\mathrm{km}\mathrm{s}^1`$ and an Alfvén speed of $`2\mathrm{km}\mathrm{s}^1`$. This model is able to match the medium-excitation level populations well, but underestimates the low-excitation, and overestimates the high-excitation levels.
Note that the H<sub>2</sub> excitation in the models of Smith, like for the planar J-shock model of Brand et al. (bra88 (1988)), was calculated under the assumption of LTE, and also it ignores non-thermal excitation mechanisms. These models therefore overestimate the population of the high energy levels by thermal collision, and at the same time they underestimate the level population because they neglect non-thermal excitation mechanisms.
We conclude that current shock models are able to reproduce the overall H<sub>2</sub> level distribution only when combining shocks with a range of velocities. However, most models do not include the physics most likely to account for the highest excitation level populations.
## 4 Summary
We obtained spectra with the ISO short wavelength spectrometer of the 2.4–45 $`\mu `$m emission toward the brightest H<sub>2</sub> emission peak of the Orion OMC-1 outflow. In those spectra we detected a large number of H<sub>2</sub>, H i, and atomic/ionic fine structure lines.
1. Estimating the extinction from relative line intensities we find that the atomic hydrogen emission originates in the foreground H ii region, whereas the H<sub>2</sub> emission comes from the shock-excited gas within the star-forming molecular cloud.
2. Most of the atomic fine structure emission originates in the foreground H ii region and its bordering photodissociation front.
3. The \[S i\]25$`\mu `$m line and some fraction of the \[Si ii\]34.8$`\mu `$m and \[Ne ii\]12.8$`\mu `$m emission could arise in strong J-shocks.
4. The total warm ($`T>`$ a few hundred K) H<sub>2</sub> column density is $`(1.9\pm 0.5)\times 10^{21}\mathrm{cm}^2`$, and the total warm H<sub>2</sub> mass in the ISO-SWS aperture is $`(0.06\pm 0.015)\mathrm{M}_{\mathrm{}}`$. The total H<sub>2</sub> luminosity within the ISO-SWS aperture is $`(17\pm 5)\mathrm{L}_{\mathrm{}}`$, and when extrapolated to the entire outflow, $`(120\pm 60)\mathrm{L}_{\mathrm{}}`$.
5. The H<sub>2</sub> excitation reveals no signs of fluorescence or a deviation from an ortho-to-para ratio of three.
6. The H<sub>2</sub> level column density distribution shows an excitation temperature which increases from about 600 K for the lowest rotational and vibrational levels to about 3200 K at level energies $`E(v,J)/k>\mathrm{14\hspace{0.17em}000}`$ K.
7. No single steady-state shock model can reproduce the observed H<sub>2</sub> level populations. To match both the low- and high-excitation level populations, a combination of slow and fast shocks is required, or time-dependent magnetic shocks which include transient J-shocks. Most shock models lack the processes that are likely to populate the high energy H<sub>2</sub> levels.
8. The higher energy H<sub>2</sub> levels may be excited either thermally in non-dissociative J-shocks, through non-thermal collisions between fast ions and molecules with H<sub>2</sub> in C-shocks, or they are due to newly formed H<sub>2</sub> molecules.
9. In a most simple model, the overall level distribution of the observed H<sub>2</sub> is well reproduced by two columns of warm, partially dissociated gas at H nuclei density $`10^6\mathrm{cm}^3`$, with respective temperatures of 200 K and 800 K and a column density ratio of 10 to 1. In this model the high-excitation tail in the H<sub>2</sub> level populations is due to H<sub>2</sub> formation.
10. The highest-excitation line we detected, 0–0 S(25), implies a column density of the $`v=0,J=27`$ level way above what we would expect from the extrapolated excitation of the lower energy levels. A different pumping mechanism, such as the gas phase formation of H<sub>2</sub> via H<sup>-</sup>, might be responsible for the population of the highest energy levels.
## ACKNOWLEDGMENTS
We are very thankful to C. Wright, A. Poglitsch, D. Lutz, L. Looney, M. Lehnert, and M. Walmsley for valuable comments, to A. Schultz for providing the NICMOS image, to B. Rubin for providing unpublished results from his H ii region model, to B.T. Draine for his contributions to the non-LTE H<sub>2</sub> models and his careful reading of the manuscript, to M. Smith for providing the H<sub>2</sub> line strengths for his bow shock model, and to the SWS Data Center at MPE, especially to H. Feuchtgruber and E. Wieprecht.
|
warning/0002/cond-mat0002341.html
|
ar5iv
|
text
|
# Phase-sensitive Evidence for 𝑑-wave Pairing Symmetry in Electron-doped Cuprate Superconductors
## Abstract
We present phase-sensitive evidence that the electron-doped cuprates Nd<sub>1.85</sub>Ce<sub>0.15</sub>CuO<sub>4-y</sub> (NCCO) and Pr<sub>1.85</sub>Ce<sub>0.15</sub>CuO<sub>4-y</sub> (PCCO) have $`d`$-wave pairing symmetry. This evidence was obtained by observing the half-flux quantum effect, using a scanning SQUID microscope, in $`c`$-axis oriented films of NCCO or PCCO epitaxially grown on tricrystal SrTiO<sub>3</sub> substrates designed to be frustrated for a $`d_{x^2y^2}`$ order parameter. Samples with two other configurations, designed to be unfrustrated for a $`d`$-wave superconductor, do not show the half-flux quantum effect.
The intense debate over pairing symmetry in the hole-doped cuprates has been resolved, largely through the development of phase-sensitive symmetry tests, in favor of predominantly $`d`$-wave orbital order parameter symmetry for a number of optimally hole-doped high-T<sub>c</sub> superconductors. However, the symmetry of the superconducting pair state in the electron-doped cuprates remains controversial. In this Letter, we present a series of phase-sensitive tricrystal experiments as evidence for $`d`$-wave pairing in two electron-doped cuprate superconductors NCCO and PCCO.
The electron-doped cuprate superconductors (Ln<sub>2-x</sub>Ce<sub>x</sub>CuO<sub>4-y</sub>, Ln=Nd, Pr, Eu, or Sm; y$``$0.04) are significantly different from their hole-doped counterparts: The hole-doped cuprates such as La<sub>2-x</sub>Sr<sub>x</sub>CuO<sub>4</sub> (LSCO) and YBa<sub>2</sub>Cu<sub>3</sub>O<sub>7</sub> (YBCO) have apical oxygen atoms; the electron-doped cuprates do not. Superconductivity in electron-doped cuprate systems occurs in a very narrow doping range (0.14$`x`$ 0.17 for NCCO and 0.13$`<x<`$ 0.2 for PCCO); in the hole-doped LSCO system the range is broader (0.05$`<x<`$0.3). The highest T<sub>c</sub> values in the hole-doped cuprates are over five times those in the highest T<sub>c</sub> electron-doped cuprate systems. In optimally doped YBCO and LSCO the in-plane resistivity increases linearly with temperature over a wide range , with small or nearly zero extrapolated values at zero temperature; in PCCO and NCCO ($`x`$=0.15) the in-plane resistivity is quadratic in temperature, with a relatively large residual resistivity . Photoemission spectroscopy studies show CuO<sub>2</sub> derived flat energy bands near the Fermi surface (FS) of the high-T<sub>c</sub> hole-doped cuprates such as YBCO and BSCCO; but not within 300meV of the FS of NCCO. Other physical properties such as the Hall coefficient, thermopower, and the pressure dependence of T<sub>c</sub>, are also different.
Therefore, the question naturally arises: Are the pairing symmetries of the electron- and hole-doped cuprates also different? It is widely believed that the electron-doped cuprates are $`s`$-wave superconductors. For example, the in-plane penetration depth $`\lambda _{ab}(T)`$ in NCCO can be fit with an exponential temperature dependence , rather than the power-law expected for unconventional superconductors with a line of nodes in the gap function. However, the distinction between power-law and exponential temperature dependences of $`\lambda _{ab}`$ can be subtle. Further, the expected temperature dependence of $`\lambda _{ab}`$ can change from $`T`$ to $`T^2`$, depending on the amount of disorder scattering, which may be large in the electron-doped cuprates. The paramagnetism arising from Nd<sup>3+</sup> ions in NCCO is a further complication. After correction for this effect, a linear temperature dependence is found for NCCO by some, but not by others . Although there is apparently no need for a similar correction in PCCO, it also shows a quadratic, non-$`s`$-wave behavior, or a BCS $`s`$-wave exponential dependence. Although the quasiparticle tunneling conductance spectra for NCCO closely resemble those of the $`d`$-wave hole-doped cuprates, the absence of a zero bias conductance peak (ZBCP) in NCCO has been taken as evidence for $`s`$-wave pairing symmetry. However, a ZBCP can be suppressed by disorder. Pair tunneling measurements result in the product of the critical current times the junction normal state resistance (I<sub>c</sub>R<sub>N</sub>) between 0.5 and 6$`\mu `$V in Pb/NCCO $`c`$-axis oriented films and single crystals, almost three orders of magnitude smaller than the 3mV Ambegaokar-Baratoff limit expected for an $`s`$-wave superconductor.
In short, contradictory results for NCCO and PCCO underscore the need for a phase-sensitive experiment. Tricrystal phase-sensitive experiments have established $`d`$-wave pairing symmetry for many of the optimally hole-doped cuprate superconductors. However, such experiments are difficult in the electron-doped superconductors because of the difficulty in making grain boundary Josephson junctions with sufficiently large critical currents. As reported in the literature, the critical current density J<sub>c</sub> decreases exponentially with increasing grain boundary misorientation angle $`\mathrm{\Theta }`$ for all cuprates, and can be described by the generic formula:
$$J_c(\mathrm{\Theta })=C_ie^{\mathrm{\Theta }/\mathrm{\Theta }_i}$$
(1)
where C<sub>i</sub> $``$ 4.4$`\times `$10<sup>7</sup>A/cm<sup>2</sup> and $`\mathrm{\Theta }_i`$5<sup>o</sup> for hole-doped superconductors, and C<sub>i</sub> $``$ 1.8$`\times `$10<sup>6</sup>A/cm<sup>2</sup> and $`\mathrm{\Theta }_i`$2<sup>o</sup> for electron-doped superconductors. Despite a similar angular dependence, the J<sub>c</sub> values can differ widely between electron- and hole-doped cuprates, especially for high-$`\mathrm{\Theta }`$ grain boundary junctions. The original tricrystal configuration(Fig. 1a), requires tilt grain boundary junctions with 30<sup>o</sup> misorientation. According to Eq. (1) J<sub>c</sub> for $`\mathrm{\Theta }`$=30<sup>o</sup> should be about 0.5A/cm<sup>2</sup> for NCCO, five orders of magnitude smaller than for YBCO. Although this disadvantage can be partially offset by making thicker films, care must be taken to avoid film inhomogeneity and retain film epitaxy. With these constraints in mind, we have deposited $`c`$-axis oriented epitaxial films (thickness $``$6000-10000$`\AA `$) of NCCO and PCCO on various tricrystal SrTiO<sub>3</sub> substrates. The electron-doped cuprate films were grown epitaxially on the substrate at T=750<sup>o</sup>C using pulsed laser deposition from a stoichiometric target of NCCO or PCCO in 300mTorr of nitrous oxide. The film deposition is followed by a vaccum annealing at 750<sup>o</sup>C for 5 minutes followed by a slow cooling to room temperature. This achieves a full oxygenation during film growth, followed by a controlled reduction to remove the excess oxygen at the apical site (i.e. in Nd<sub>1.85</sub>Ce<sub>0.15</sub>CuO<sub>4-y</sub>, $`y`$ 0.04). A judicious control of the oxygen content is crucial for maximizing T<sub>c</sub>, and to control other superconducting and normal-state properties in the bulk and at the junction interface .
Our films have a T<sub>c</sub> of 22-25K for NCCO, and 22-23K for PCCO. The 10% to 90% resistive transition width is 0.6K or less. The in-plane normal-state resistivity is $``$ 300$`\mu `$Ohm-cm at room temperature, with a quadratic temperature variation. The room temperature to T<sub>c</sub> resistivity ratio is 5 to 6. The critical current of an NCCO bicrystal grain boundary junction ($`\mathrm{\Theta }`$ = 30<sup>o</sup>) at 4.2K and ambient magnetic field is J<sub>c</sub> = 6$`\pm `$2A/cm<sup>2</sup>, about a factor of 10 larger than predicted by Eq. 1, indicating better sample quality.
As in the previous tricrystal experiments, we use a scanning SQUID microscope (SSM) to measure the magnetic fields near the tricrystal point of a $`c`$-axis oriented epitaxial cuprate film deposited on a tricrystal (100) STO substrate (Fig. 1(a)). The crystallographic orientations of the tricrystal were chosen to form an energetically frustrated state at the tricrystal point for a superconductor with $`d_{x^2y^2}`$ pairing symmetry, regardless of whether the grain boundary junction interface is in the clean or dirty limit. This frustration is relaxed by the spontaneous generation of a magnetic vortex with total flux $`\mathrm{\Phi }`$ half of the conventional superconductor flux quantum ($`\mathrm{\Phi }=\mathrm{\Phi }_0/2=hc/4e`$). Direct observation of this half-flux quantum effect serves as conclusive evidence for $`d`$-wave symmetry. Shown in Fig.s 1(b) and 1(c) are SSM images for an NCCO film deposited on an STO substrate with the geometry of Fig. 1(a). These images are of Josephson vortices with fields either pointing out of (b), or into (c) the sample, centered at the tricrystal point and extending along the grain boundaries. The images are complicated by smooth variations in the background signal, inductive interactions between the SQUID and the sample, and dipolar features (often observed in SSM images of cuprate superconductors), presumably due to roughness and/or magnetic inhomogeneities at the surface. Figure 1(d) shows a 3-dimensional rendering of an image obtained by subtracting (c) from (b) (and dividing by 2). This largely removes the extraneous features. There are several evidences that the magnetic signals at the tricrystal point are due to half-flux quantum Josephson vortices. First, although the vortices can switch signs, the field magnitude is the same after these changes, and there is always a vortex with this general appearance at the tricrystal point, under any conditions of cooldown and externally applied magnetic field. Further, the observed magnetic fields agree with those expected for the half-flux quantum Josephson vortex. The normal component of the magnetic field from such a vortex at the superconducting surface is given by
$$B_z(r_i,r_i)=\frac{\mathrm{\Phi }_0a_i}{\pi \lambda _L\lambda _{Ji}}\frac{e^{r_i/\lambda _{Ji}}}{1+a_i^2e^{2r_i/\lambda _{ji}}}e^{r_i/\lambda _L},$$
(2)
where $`r_i`$ is the distance along the i<sup>th</sup> grain boundary from the tricrystal point, $`r_i`$ is the perpendicular distance from the i<sup>th</sup> (closest) grain boundary, $`\lambda _L`$ is the London penetration depth, $`\lambda _{Ji}`$ is the Josephson penetration depth of the i<sup>th</sup> grain boundary, and the $`a_i`$’s are normalization constants chosen such that the magnetic field is continuous at the tricrystal point and the total flux in the vortex is equal to $`\mathrm{\Phi }_0/2=hc/4e`$. The pickup loop is a distance $`z`$ above, and nearly parallel to, the surface of the sample. The distance $`z`$ is determined by fitting data for an isolated Abrikosov vortex to the fields B<sub>z</sub>=$`(\mathrm{\Phi }_0/2\pi )z/r^3`$, integrating over the known area of the pickup loop, with $`z`$ as the sole fitting parameter (see Fig. 2(a)). The fields calculated from Eq. 2 are propagated above the surface by this height $`z`$ and integrated over the SQUID pickup loop for comparison with experiments. An example is shown in Fig 2(b). In this figure the solid points are cross-sections through the SSM image of Figure 1(d) through the tricrystal point and parallel to the horizontal and diagonal grain boundaries. The lines are fits to this data, using the Josephson penetration depth as the sole fitting parameter. The solid line assumes $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$, the dashed line is the best fit for $`\mathrm{\Phi }=\mathrm{\Phi }_0`$. If we vary $`\mathrm{\Phi }`$ while optimizing the Josephson penetration depths, we find $`\mathrm{\Phi }`$=0.47+0.18-0.14$`\mathrm{\Phi }_0`$, using a doubling of the best fit $`\chi ^2`$ values as the criterion for assigning error bars. The value for $`\lambda _J`$ found from this fit assuming $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$ is 48$`\mu `$m, corresponding to J$`{}_{c}{}^{}=\mathrm{}c^2/8\pi ed\lambda _J^2`$20A/cm<sup>2</sup> (taking d=2$`\lambda _{ab}`$=0.5$`\mu `$m). Comparison of the value for J<sub>c</sub> obtained in this way for another tricrystal sample was in good agreement with a 4 point probe measurement on a bicrystal grain boundary junction fabricated at the same time. We obtained qualitatively similar results on repeated measurements of 5 NCCO samples with the same (frustrated) geometry.
We repeated these experiments with tricrystal NCCO films in two different unfrustrated configurations (Fig. 3(b),(c)) designed not to show the half-flux quantum effect if NCCO is a $`d`$-wave superconductor. SSM images from samples in all three geometries are presented in Fig. 3. Fig. 3(c) shows fringing fields from two Abrikosov vortices just outside the scan area. The samples in the non-frustrated geometries show little, if any, flux at the tricrystal points. There is residual signal at the grain boundaries and the tricrystal point, which may be due to topographic effects, facetting , or small changes in the mutual inductance between the SQUID and sample. Even if we attribute the signal at the tricrystal point in the unfrustrated samples to magnetic fields trapped in the grain boundaries, the total magnetic flux is less than a few percent of the flux seen at the tricrystal point in the frustrated sample. We therefore conclude that the frustrated sample 3(a) shows the half-flux quantum effect, while the other two samples 3(b) and 3(c) do not. This is consistent with NCCO being a $`d_{x^2y^2}`$ superconductor.
Similar results have been obtained in a second electron-doped cuprate, PCCO. Fig. 4 shows scanning SQUID microscope images of a PCCO film epitaxially grown on a STO substrate with the frustrated geometry of Fig. 1(a). A half-flux quantum Josephson vortex (Fig. 4(a))was spontaneously generated at the tricrystal point, and could be inverted (Fig. 4(b)) by varying the external field. The difference image (a-b)/2 could be well fit to Eq. 2 assuming $`\mathrm{\Phi }=\mathrm{\Phi }_0/2`$, indicating that PCCO has predominantly $`d`$-wave pairing symmetry. Further, sum images (a+b)/2 (e.g. Fig. 4(c)) showed almost no feature at the tricrystal point in both NCCO and PCCO. This indicates that time-reversal invariance is obeyed in the pairing in the electron-doped cuprates.
In conclusion, we have used a scanning SQUID microscope in a series of tricrystal experiments to produce definitive phase-sensitive evidence for $`d`$-wave pairing symmetry in the electron-doped cuprate superconductors NCCO and PCCO. Thus predominantly $`d`$-wave superconductivity is established in both optimally electron- and hole-doped high-T<sub>c</sub> superconductors. This is consistent with several previously-thought anomalous experimental observations. For example, the extremely small I<sub>c</sub>R<sub>N</sub> product for $`c`$-axis pair tunneling in Pb/NCCO junctions is in accordance with a predominantly $`d`$-wave order parameter in NCCO. The small and finite I<sub>c</sub>R<sub>N</sub> product may be due to symmetry-broken induced $`s`$-wave pairing at the junction interface or other extrinsic mechanisms. The missing ZBCP in the NCCO quasiparticle conductance spectra may well be due to strong scattering effects. Finally, the power-law temperature dependence observed in some $`\lambda _{ab}(T)`$ data supports our conclusion that both NCCO and PCCO are $`d`$-wave superconductors.
We would like to gratefully thank A. Gupta, R.H. Koch, J. Mannhart, K.A. Moler, D.M. Newns, J.Z. Sun, and S.I. Woods for useful discussions, and G. Trafas for technical assistance.
|
warning/0002/cond-mat0002152.html
|
ar5iv
|
text
|
# Magnetic field of an in-plane vortex inside and outside a layered superconducting film
## Abstract
In the present work we study an anisotropic layered superconducting film of finite thickness. The film surfaces are considered parallel to the $`bc`$ face of the crystal. The vortex lines are oriented perpendicular to the film surfaces and parallel to the superconducting planes. We calculate the local field and the London free energy for this geometry. Our calculation is a generalization of previous works where the sample is taken as a semi-infinite superconductor. As an application of this theory we investigate the flux spreading at the superconducting surface.
Scanning superconducting quantum interference device (SQUID) microscope has been used to image interlayer Josephson vortices trapped between the planes of layered superconductors. This technique has been used to measure the out-plane London penetration depth that gives the distance over which the interlayer current $`j_c`$ changes as a function of in-plane coordinates. These measurements have been important to test the interlayer tunneling model as a candidate to explain the mechanism of superconductivity for the high-$`T_c`$ superconductors.
Recently, Kirtley, Kogan, Clem, and Moler have found expressions for the local magnetic field emerging from a superconductor with the vortex lines parallel to the planes, and normal to a crystal face. Their geometry consists of a semi-infinite anisotropic superconductor. Furthermore, they have used these expressions to fit the experimental data at the surface in order to obtain an estimate of the value of the out-plane penetration depth $`\lambda _c`$. They have shown that, neglecting the vortex spreading at the surface may overestimate $`\lambda _c`$ as much as 30%.
In the present paper we extend the work of Ref. to an anisotropic layered superconducting film of finite thickness and of infinite extent in the $`bc`$ face of the crystal. We will show that, if the thickness of the film is of order or smaller than $`\lambda _c`$, the magnetic field distribution is even more affected by flux spreading.
Let us first formulate the problem to be solved. The geometry we consider is illustrated in Fig. 1. We suppose that the vortex line is perpendicular to the film. We will calculate the local field inside the film using the London equation. For this geometry this equation is given by
$$\mathbf{}\times [\stackrel{}{\mathbf{}}\times 𝐡]+𝐡=\widehat{𝐳}\mathrm{\Phi }_0\delta (𝐫),$$
(1)
where $`\stackrel{}{\mathbf{}}`$ is the London (tensor) penetration depth. This tensor is diagonal and its components are given by $`\mathrm{\Lambda }_{xx}=\mathrm{\Lambda }_{yy}=\lambda _c^2`$, $`\mathrm{\Lambda }_{zz}=\lambda _{ab}^2`$; here $`\lambda _{ab}`$ and $`\lambda _c`$ are the in- and out-plane penetration depth respectively; $`\mathrm{\Phi }_0`$ is the quantum flux. The film is anisotropic along the $`c`$ direction.
Outside the sample, the local field satisfies the equation
$$^2𝐡=0.$$
(2)
Although we will consider the case of a single vortex, the generalization to the case of $`N`$ vortices is straightforward. To proceed is more convenient to Fourier transform Eqs. (1) and (2). For $`|z|<d/2`$, using the Maxwell equation $`\mathbf{}𝐡=0`$, we obtain a set of three coupled differential equations for the two dimensional Fourier transform of the local magnetic field $`𝐡(𝐤,z)=d^2re^{ı𝐤𝐫}𝐡(𝐫,z)`$,
$$\left[1+\lambda _{ab}^2k^2\lambda _{ab}^2\frac{^2}{z^2}\right]h_x=0,$$
(3)
$$\left[1+\lambda _{ab}^2k^2\lambda _c^2\frac{^2}{z^2}\right]h_y+(\lambda _c^2\lambda _{ab}^2)ik_y\frac{h_z}{z}=0,$$
(4)
$$\left[1+\lambda _{ab}^2k_x^2+\lambda _c^2k_y^2\lambda _{ab}^2\frac{^2}{z^2}\right]h_z+(\lambda _c^2\lambda _{ab}^2)ik_y\frac{h_y}{z}=\mathrm{\Phi }_0.$$
(5)
For $`|z|>d/2`$ one has
$$(\frac{^2}{z^2}k^2)𝐡=0.$$
(6)
At the vacuum-superconductor interfaces $`z=\pm d/2`$ the field components are continuous and the component of the current perpendicular to both film surfaces vanishes. One has
$$𝐡_<(𝐤,d/2)=𝐡_m(𝐤,d/2),$$
(7)
$$𝐡_m(𝐤,d/2)=𝐡_>(𝐤,d/2),$$
(8)
$$\widehat{𝐳}\left[𝐃_z(𝐤)\times 𝐡_m\right]_{z=\pm d/2}=0,$$
(9)
$$𝐃_z(𝐤)𝐡=0,$$
(10)
where the operator $`𝐃_z(𝐤)=i𝐤+\widehat{𝐳}\frac{}{z}`$. The subscripts ($`<,>`$) stand for below the surface $`z=d/2`$ and above the surface $`z=d/2`$, respectively, whereas the subscript $`m`$ is meant for the field inside the sample.
We start by solving first Eq. (6). The solution which satisfies the boundary condition of Eq. (10) takes the form
$`𝐡_>(𝐤,z)`$ $`=`$ $`(i𝐤+\widehat{𝐳}k)\phi (𝐤)e^{k(zd/2)},`$ (11)
$`𝐡_<(𝐤,z)`$ $`=`$ $`(i𝐤+\widehat{𝐳}k)\phi (𝐤)e^{k(z+d/2)},`$ (12)
where $`\phi (𝐤)`$ is a scalar function which will be determined by using the boundary condition either of Eq. (7) or (8).
Eq. (3) can also be easily solved. We have
$$h_{m,x}(𝐤,z)=W_1e^{\alpha z}+W_2e^{\alpha z},$$
(13)
where
$$\alpha =\sqrt{\frac{1+\lambda _{ab}^2k^2}{\lambda _{ab}^2}},$$
(14)
and the $`W`$’s are two constants to be determined by using the boundary conditions.
The other two components of the local field can be determined by decoupling Eqs. (4) and (5). This can be done by calculating the determinant of the matrix formed by the coefficients of Eqs. (4) and Eq. (5). This yields the following equation for $`h_{m,y}`$
$$\left[1+\lambda _{ab}^2k^2\lambda _{ab}^2\frac{^2}{z^2}\right]\left[1+\lambda _{ab}^2k_x^2+\lambda _c^2k_y^2\lambda _c^2\frac{^2}{z^2}\right]h_{m,y}=0.$$
(15)
The solution for this equation is given by
$$h_{y,m}(𝐤,z)=W_3e^{\alpha z}+W_4e^{\alpha z}+W_5e^{\gamma z}+W_6e^{\gamma z},$$
(16)
where the $`W`$’s are constants to be determined by using the boundary conditions and
$$\gamma =\sqrt{\frac{1+\lambda _{ab}^2k_x^2+\lambda _c^2k_y^2}{\lambda _c^2}}.$$
(17)
The solution for $`h_{m,z}`$ can be found by inserting Eq. (16) back into Eq. (4) or (5).One has
$$h_{m,z}(𝐤,z)=\frac{\mathrm{\Phi }_0}{\lambda _c^2\gamma ^2}+\frac{\alpha }{ik_y}(W_3e^{\alpha z}W_4e^{\alpha z})\frac{ik_y}{\gamma }(W_5e^{\gamma z}W_6e^{\gamma z}).$$
(18)
The determination of the constants $`W_i`$ is very cumbersome and we omit it here. We just present the main steps of the complete solution. First of all, we use the Maxwell equation (10). This allows us to write $`W_3`$ and $`W_4`$ in terms of $`W_1`$ and $`W_2`$. Secondly, we use the boundary condition of Eq. (9) in both faces of the film. This leads us to the solution of $`W_5`$ and $`W_6`$ in terms of $`W_1`$ and $`W_2`$. Then, we are left only with three constants to determine, namely, $`W_1`$, $`W_2`$, and $`\phi `$. Thirdly, we use the continuity of the local field at the film surfaces \[either Eq. (7) or (8); both of them yields the same solution to these constants\]. One obtains,
$`W_1`$ $`=`$ $`ik_x{\displaystyle \frac{\phi }{2\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}},`$ (19)
$`W_2`$ $`=`$ $`ik_x{\displaystyle \frac{\phi }{2\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}},`$ (20)
$`W_3`$ $`=`$ $`{\displaystyle \frac{\lambda _{ab}^2k_xk_y}{1+\lambda _{ab}^2k_x^2}}W_1,`$ (21)
$`W_4`$ $`=`$ $`{\displaystyle \frac{\lambda _{ab}^2k_xk_y}{1+\lambda _{ab}^2k_x^2}}W_2,`$ (22)
$`W_5`$ $`=`$ $`{\displaystyle \frac{k_y}{k_x(1+\lambda _{ab}^2k_x^2)\mathrm{sinh}(\gamma d)}}\left\{W_1\mathrm{sinh}\left[\left(\gamma +\alpha \right){\displaystyle \frac{d}{2}}\right]+W_2\mathrm{sinh}\left[\left(\gamma \alpha \right){\displaystyle \frac{d}{2}}\right]\right\},`$ (23)
$`W_6`$ $`=`$ $`{\displaystyle \frac{k_y}{k_x(1+\lambda _{ab}^2k_x^2)\mathrm{sinh}(\gamma d)}}\left\{W_1\mathrm{sinh}\left[\left(\gamma \alpha \right){\displaystyle \frac{d}{2}}\right]+W_2\mathrm{sinh}\left[\left(\alpha +\gamma \right){\displaystyle \frac{d}{2}}\right]\right\},`$ (24)
$`\phi (𝐤)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\lambda _c^2\gamma ^2}}\mathrm{\Delta }(𝐤),`$ (25)
where
$$\mathrm{\Delta }(𝐤)=\left[k+\frac{\lambda _{ab}^2k_x^2\alpha \mathrm{coth}\left(\frac{\alpha d}{2}\right)+\frac{k_y^2}{\gamma }\mathrm{coth}\left(\frac{\gamma d}{2}\right)}{1+\lambda _{ab}^2k_x^2}\right]^1.$$
(26)
Finally, upon substituting Eqs. (19-25) into Eqs. (13), (16) and (18), we find for the local magnetic field inside the film
$`h_{m,x}(𝐤,z)`$ $`=`$ $`ik_x\phi (𝐤){\displaystyle \frac{\mathrm{sinh}(\alpha z)}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}},`$ (27)
$`h_{m,y}(𝐤,z)`$ $`=`$ $`ik_y{\displaystyle \frac{\phi (𝐤)}{1+\lambda _{ab}^2k_x^2}}\left[\lambda _{ab}^2k_x^2{\displaystyle \frac{\mathrm{sinh}(\alpha z)}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}}+{\displaystyle \frac{\mathrm{sinh}(\gamma z)}{\mathrm{sinh}\left(\frac{\gamma d}{2}\right)}}\right],`$ (28)
$`h_{m,z}(𝐤,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\lambda _c^2\gamma ^2}}{\displaystyle \frac{\phi (𝐤)}{1+\lambda _{ab}^2k_x^2}}\left[\lambda _{ab}^2k_x^2\alpha {\displaystyle \frac{\mathrm{cosh}(\alpha z)}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}}+{\displaystyle \frac{k_y^2}{\gamma }}{\displaystyle \frac{\mathrm{cosh}(\gamma z)}{\mathrm{sinh}\left(\frac{\gamma d}{2}\right)}}\right].`$ (29)
We would like to point out that these results could not be obtained from those of Ref. without solving the problem. In fact, the solution of the London equation for a superconducting film is different and more difficult than for a semi-infinite superconductor.
Let us turn our discussion to the calculation of the London free energy. The energy of the vortex system is given by $`F=F_V+F_S`$, where $`F_V`$ is the field energy in the vacuum and $`F_S`$ is the energy inside the superconductor. One has
$`F_V`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}{\displaystyle \frac{d^2k}{(2\pi )^2}\left\{_{d/2}^{\mathrm{}}𝑑z|h_>(𝐤,z)|^2+_{\mathrm{}}^{d/2}𝑑z|h_<(𝐤,z)|^2\right\}},`$ (30)
$`F_S`$ $`=`$ $`{\displaystyle \frac{1}{8\pi }}{\displaystyle }{\displaystyle \frac{d^2k}{(2\pi )^2}}{\displaystyle _{d/2}^{d/2}}dz\{|h_m(𝐤,z)|^2`$ (32)
$`+[𝐃_z(𝐤)\times 𝐡_m(𝐤,z)]\stackrel{}{\mathbf{}}[𝐃_z(𝐤)\times 𝐡_m(𝐤,z)]\}.`$
By substituting the appropriate expressions of the local magnetic field inside Eqs. (30) and (32), after a length algebra, we obtain
$$F=\frac{\mathrm{\Phi }_0^2}{8\pi }\frac{d^2k}{(2\pi )^2}\frac{1}{\lambda _c^2\gamma ^2}\left[d+2\frac{\mathrm{\Delta }(𝐤)}{\lambda _c^2\gamma ^2}\right].$$
(33)
The free energy can be generalized to an ensemble of $`N`$ interacting vortex lines upon multiplying the integrand of Eq. (33) by $`|S(𝐤)|^2`$ where the structure factor is given by
$$S(𝐤)=\underset{i}{}e^{i𝐤𝐑_i}.$$
(34)
Here $`𝐑_i`$ is the position of the $`i`$-vortex line. Note that this extended result should be valid for an ensemble of distorted vortices, that is, the positions of the vortices do not necessarily correspond to the equilibrium configuration. The first term inside Eq. (33) represents the interaction energy of the vortex lines as if the surfaces were absent. The second term represents the surface energy associated to the magnetic energy of the stray field at the superconductor-vacuum interface. Notice that for $`k`$ small (large $`r`$), $`\gamma ^21/\lambda _c^2`$, and $`\mathrm{\Delta }(𝐤)1/k`$. Thus, the surface energy goes as $`\mathrm{\Phi }_0^2/8\pi ^2r`$. Consequently, the interaction on the surface depends neither on the film thickness nor on the anisotropy. This is the Pearl result for vortices emerging from a semi-infinite isotropic supercondutor. Another interesting particular case of Eq. (33) is the limit of a very thin film $`d0`$, and $`k`$ small. In this limit, from Eq. (26) it is straightforward to show that $`\mathrm{\Delta }(𝐤)=1/(k+2(\lambda _{ab}^2k_x^2+\lambda _c^2k_y^2)/d)`$. Therefore, from Eq. (33) we obtain
$$F=E_0\frac{dk^2}{(2\pi )^2}\frac{2\pi d}{k\mathrm{\Lambda }^1+(k_x^2+\mathrm{\Gamma }k_y^2)},$$
(35)
where $`E_0=(\mathrm{\Phi }_0/4\pi \lambda _{ab})^2`$, $`\mathrm{\Lambda }=2\lambda _{ab}^2/d`$, and $`\mathrm{\Gamma }=\lambda _c^2/\lambda _{ab}^2`$ is the anisotropy parameter. This is precisely the energy of a single vortex in very thin film first obtained by Pearl.
Now we will turn our attention to the streamlines of the integrated field over $`x`$. The distribution of magnetic field emerging on the surface can be probed with a SQUID pickup loop. If the SQUID probe is oriented in the $`xy`$ plane, the total magnetic flux will be nearly equal to the pickup loop size times
$$_z(y,z)=_{\mathrm{}}^{\mathrm{}}h_z(x,y,z)𝑑x=_{\mathrm{}}^{\mathrm{}}\frac{dk_y}{2\pi }h_z(0,k_y,z)e^{ik_yy},$$
(36)
whereas, if the SQUID probe is oriented along the $`xz`$ plane, the total magnetic flux is measured through the pickup loop size times
$$_y(y,z)=_{\mathrm{}}^{\mathrm{}}h_y(x,y,z)𝑑x=_{\mathrm{}}^{\mathrm{}}\frac{dk_y}{2\pi }h_y(0,k_y,z)e^{ik_yy}.$$
(37)
In order to compare our results with the results of Ref. , we will replace the vacuum-superconductor surfaces at $`z=0`$ and $`z=d`$. This can be done through the translation $`zz+d/2`$. From Eqs. (11), (12), and (27-29) we obtain,
$`𝐡_>(𝐤,z)`$ $`=`$ $`(i𝐤+\widehat{𝐳}k)\phi (𝐤)e^{kz},`$ (38)
$`𝐡_<(𝐤,z)`$ $`=`$ $`(i𝐤+\widehat{𝐳}k)\phi (𝐤)e^{k(z+d)},`$ (39)
$`h_{m,x}(𝐤,z)`$ $`=`$ $`ik_x\phi (𝐤){\displaystyle \frac{\mathrm{sinh}\left[\alpha \left(z+\frac{d}{2}\right)\right]}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}},`$ (40)
$`h_{m,y}(𝐤,z)`$ $`=`$ $`ik_y{\displaystyle \frac{\phi (𝐤)}{1+\lambda _{ab}^2k_x^2}}\left\{\lambda _{ab}^2k_x^2{\displaystyle \frac{\mathrm{sinh}\left[\alpha \left(z+\frac{d}{2}\right)\right]}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}}+{\displaystyle \frac{\mathrm{sinh}\left[\gamma \left(z+\frac{d}{2}\right)\right]}{\mathrm{sinh}\left(\frac{\gamma d}{2}\right)}}\right\},`$ (41)
$`h_{m,z}(𝐤,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\lambda _c^2\gamma ^2}}{\displaystyle \frac{\phi (𝐤)}{1+\lambda _{ab}^2k_x^2}}\left\{\lambda _{ab}^2k_x^2\alpha {\displaystyle \frac{\mathrm{cosh}\left[\alpha \left(z+\frac{d}{2}\right)\right]}{\mathrm{sinh}\left(\frac{\alpha d}{2}\right)}}+{\displaystyle \frac{k_y^2}{\gamma }}{\displaystyle \frac{\mathrm{cosh}\left[\gamma \left(z+\frac{d}{2}\right)\right]}{\mathrm{sinh}\left(\frac{\gamma d}{2}\right)}}\right\}.`$ (42)
The substitution of the appropriate expressions into Eqs. (36) and (37) yields for the $`z`$ component of the $`\stackrel{}{}`$ field,
$`_>^z(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle \frac{\mathrm{cos}(y^{}\mathrm{sinh}u)e^{z^{}\mathrm{sinh}u}}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}},`$ (43)
$`_{m,z}(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}\{{\displaystyle \frac{\pi }{2}}e^{|y^{}|}{\displaystyle _0^{\mathrm{}}}du\mathrm{tanh}u`$ (45)
$`\times {\displaystyle \frac{\mathrm{cos}(y^{}\mathrm{sinh}u)}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}}{\displaystyle \frac{\mathrm{cosh}\left[\left(z^{}+\frac{d}{2\lambda _c}\right)\mathrm{cosh}u\right]}{\mathrm{sinh}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}}\},`$
$`_<^z(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle \frac{\mathrm{cos}(y^{}\mathrm{sinh}u)e^{(z^{}+d/\lambda _c)\mathrm{sinh}u}}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}},`$ (46)
where $`y^{}=y/\lambda _c`$ and $`z^{}=z/\lambda _c`$.
The $`y`$ component takes the form
$`_>^y(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle \frac{\mathrm{sin}(y^{}\mathrm{sinh}u)e^{z^{}\mathrm{sinh}u}}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}},`$ (47)
$`_{m,y}(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle \frac{\mathrm{sin}(y^{}\mathrm{sinh}u)}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}}{\displaystyle \frac{\mathrm{sinh}\left[\left(z^{}+\frac{d}{2\lambda _c}\right)\mathrm{cosh}u\right]}{\mathrm{sinh}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}},`$ (48)
$`_<^y(y,z)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Phi }_0}{\pi \lambda _c}}{\displaystyle _0^{\mathrm{}}}𝑑u{\displaystyle \frac{\mathrm{sin}(y^{}\mathrm{sinh}u)e^{(z^{}+d/\lambda _c)\mathrm{sinh}u}}{\mathrm{cosh}u+\mathrm{sinh}u\mathrm{coth}\left(\frac{d}{2\lambda _c}\mathrm{cosh}u\right)}}.`$ (49)
Note that in the limit of $`d\mathrm{}`$, our results are exactly the same as those of Ref. .
The results for the $`\stackrel{}{}`$ field presented above should be useful to interpret the experimental data obtained by using scanning SQUID micorscopy. Unfortunately, the experiments have been performed in samples of large thickness. This renders the test of the theory impracticable. In fact, vortices have been magnetically imaged in films, but for a different geometry, that is, the superconducing planes are taken parallel to the surfaces of the film and the vortex lines are considered perpendicular to the film surfaces. In this case, we can extract the in-plane penetration depth $`\lambda _{ab}`$ rather than $`\lambda _c`$, from the fitting of the experimental data. So, we will restrict our analysis only to the theoretical expressions.
Fig. 2 shows the streamlines of the $`\stackrel{}{}(y,z)`$ field for a single interlayer vortex centered at $`x=0`$, $`y=0`$. The streamlines were generated as sketched in Ref. . We used various values of the film thickness. Note that as the thickness of the film grows, the flux spreading is important only near the surface, whereas deep inside the thinner film the streamlines are still very distorted, except those close to the center of the vortex.
To see how important the flux spreading inside a superconducting film is, we calculated numerically $`\pi \lambda _c_z(y,z)/\mathrm{\Phi }_0`$ as function of $`y/\lambda _c`$ for three different values of $`d`$ at $`z=0`$. As can be seen from Fig. 3, the full width at half maximum of the flux contour is $`1.87\lambda _c`$ for the case $`d=5\lambda _c`$, while it is $`1.65\lambda _c`$ for $`d=\lambda _c`$. Thus, if the flux spreading inside the film is not taken into account, the value of $`\lambda _c`$ could be underestimated by 10%. This error grows as the film thickness decreases.
Finally, we would like to point out that the present results agree with their isotropic counterpart. If we set $`\lambda _{ab}=\lambda _c=\lambda `$ in Eq. (25) and (26), we obtain the same result as in Ref. . Apparently, our results are different of those found in Ref. , but they show very similar streamlines.
In summary, we have calculated the field distribution of a single vortex inside and outside a layered superconducting film of arbitrary thickness. We also calculated the London free energy of an ensemble of vortices. From the expression for the energy one can recover the interaction potential between vortices for a very thin film and the vortices emerging from a semi-infinite superconductor. In addition, we have shown that flux spreading inside a superconducting film of order or smaller than $`\lambda _c`$ affects substantially the full width at half maximum of the flux contour.
###### Acknowledgements.
The author thanks the Brazilian Agencies FAPESP and CNPq for financial support.
Fig. 1/Sardella
|
warning/0002/astro-ph0002171.html
|
ar5iv
|
text
|
# ”
## 1 Introduction
According to the conventional point of view, gravity does not induce any electric polarization in the interior of celestial bodies and electric forces are never considered in the balance of matter of celestial bodies. Moreover, it is generally assumed that the electric interaction plays practically no role in astrophysics. It is a consequence of the comprehension that the appreciable electric polarization cannot arise in metals and other nonsegneto- and nonpiro-electric materials. It is entirely correct for all substances under action of small pressure. But, thus, one can disregard the fact that ultrahigh pressure transmutes all substances into plasma state and radically changes the properties of substance. In ultradense plasma, there is a different additional mechanism of the gravity-induced electric polarization.
In a large celestial body, consisting of ultradense plasma, this gravity-induced electric polarization (GIEP) can be rather great and can play a determining role in the formation of a number of features of the structure of a celestial body and its properties.
First of all, it concerns the following three problems, the statement and the solution of which change drastically:
\- the distribution of pressure and density of matter inside a celestial body;
\- the generation of a magnetic field by celestial bodies;
\- the formation of a spectrum of steady-state values of masses of celestial bodies.
As a consequence, these features of the structure can influence the evolution of stars.
## 2 The gravity-induced electric polarization in conducting matter
The action of gravity on metals has often been a topic of discussion before -. The basic result of these researches is reduced to the statement that inside a metal gravity induces an electric field with an intensity
$$𝐄\frac{𝐦_𝐢𝐠}{𝐞},$$
(1)
where $`m_i`$ is the mass of an ion,
$`𝐠`$ is gravity acceleration,
$`e`$ is the electron charge.
This field is so small that it is not possible to measure it experimentally. It is a direct consequence of the presence of an ion lattice in a metal. This lattice is deformed by gravity and then the electron gas adapts its density to this deformation. The resulting field becomes very small.
Under superhigh pressure, all substances transform into ultradense matter usually named nuclear-electron plasma . It occurs when external pressure enhances the density of matter several times . Such values of pressure exist inside celestial bodies.
In nuclear-electron plasma the electrons form the degenerated Fermi gas. At the same time, the positively charged ions form inside plasma a dense packing lattice ,. As usually accepted, this lattice may be replaced by a lattice of spherical cells of the same volume. The radius $`r_s`$ of such a spherical cell in plasma of the mass density $`\gamma `$ is given by
$$\frac{4\pi }{3}r_s^3=\left(\frac{\gamma }{m_i}\right)^1=\frac{Z}{n},$$
(2)
where Z is the charge of the nucleus, $`m_i=Am_p`$ is the mass of the nucleus, A is the atomic number of the nucleus, $`m_p`$ is the mass of a proton, and n is the electron number density
$$n=\frac{3Z}{4\pi r_s^3}.$$
(3)
The equilibrium condition in matter is described by the constancy of its electrochemical potential . In plasma, the direct interaction between nuclei is absent, therefore the equilibrium in a nuclear subsystem of plasma (at $`T=0`$) looks like
$$\mu _i=m_i\psi +Ze\phi =const.$$
(4)
Here $`\phi `$ is the potential of an electric field and $`\psi `$ is the potential of a gravitational field.
The direct action of gravitation on electrons can be neglected. Therefore, the equilibrium condition in the electron gas is
$$\mu _e=\frac{p_F^2}{2m_e}(e\delta q)\phi =const,$$
(5)
where $`m_e`$ is the mass of an electron and $`p_F`$ is the Fermi momentum.
By introducing the charge $`\delta q`$, we take into account that the charge of the electron cloud inside a cell can differ from $`Ze`$. A small number of electrons can stay at the surface of a plasma body where the electric potential is absent. It results that the charge in a cell, subjected to the action of the electric potential, is effectively decreased on a small value $`\delta q`$. If the radius of a star $`R_0`$ is approximately $`10^{10}cm`$, one can expect that this mechanism gives on the order of magnitude $`\frac{\delta q}{e}\frac{r_s}{R_0}10^{18}`$.
The electric polarization in plasma is a result of changing in density of both nuclear and electron gas subsystems. The electrostatic potential of the arising field is determined by the Gauss’ law
$$^2\phi =\frac{1}{r^2}\frac{d}{dr}\left[r^2\frac{d}{dr}\phi \right]=4\pi \left[Ze\delta (r)en\right],$$
(6)
where the position of nuclei is described by the function $`\delta (r)`$.
According to the Thomas - Fermi method, $`n`$ is approximated by
$$n=\frac{8\pi }{3h^3}p_F^3.$$
(7)
With this substitution, Eq.(6) is converted into a nonlinear differential equation for $`\phi `$, which for $`r>0`$ is given by
$$\frac{1}{r^2}\frac{d}{dr}\left(r^2\frac{d}{dr}\phi (r)\right)=4\pi \left[\frac{8\pi }{3h^3}\right]\left[2m_e(\mu _e+(e\delta q)\phi )\right]^{3/2}.$$
(8)
It can be simplified by introducing the following variables :
$$\mu _e+(e\delta q)\phi =Ze^2\frac{u}{r}$$
(9)
and $`r=ax`$,
where
$`a=\{\frac{9\pi ^2}{128Z}\}^{1/3}a_0`$
with $`a_0=\frac{\mathrm{}^2}{m_ee^2}=`$ Bohr radius.
With the account of Eq.(4)
$$Ze^2\frac{u}{r}=const\frac{m_i\psi }{Z}\delta q\phi .$$
(10)
Then Eq.(8) gives
$$\frac{d^2u}{dx^2}=\frac{u^{3/2}}{x^{1/2}}.$$
(11)
In terms of u and x, the electron density within a cell is given by
$$n_{TF}=\frac{8\pi }{3h^3}p_F^3=\frac{32Z^2}{9\pi ^3a_0^3}\left(\frac{u}{x}\right)^{3/2}.$$
(12)
Under the influence of gravity, the charge of the electron gas in a cell becomes equal to
$$Q_e=4\pi e_0^{r_s}n(r)r^2𝑑r=\frac{8\pi e}{3h^3}\left[2m_e\frac{Ze^2}{a}\right]^{3/2}4\pi a^3_0^{x_s}x^2𝑑x\left[\frac{u}{x}\right]^{3/2}.$$
(13)
Using Eq.(11), we obtain
$$Q_e=Ze_0^{x_s}x𝑑x\frac{d^2u}{dx^2}=Ze_0^{x_s}𝑑x\frac{d}{dx}\left[x\frac{du}{dx}u\right]=Ze\left[x_s\frac{du}{dx}|_{x_s}u(x_s)+u(0)\right].$$
(14)
At $`r0`$ the electric potential is due to the nucleus alone $`\phi (r)\frac{Ze}{r}`$. It means that $`u(0)1`$ and each cell of plasma obtains a small charge
$$\delta q=Ze\left[x_s\frac{du}{dx}|_{x_s}u(x_s)\right]=Zex_{s}^{}{}_{}{}^{2}\left[\frac{d}{dx}\left(\frac{u}{x}\right)\right]_{x_s}.$$
(15)
For a cell placed in the point $`𝐑`$ inside a star
$$\delta q=Zer_s^2\left[\frac{d}{d𝐑}\left(\frac{u}{r}\right)\right]\left[\frac{d𝐑}{dr_s}\right].$$
(16)
Considering that the gravity acceleration $`𝐠=\frac{𝐝\psi }{\mathrm{𝐝𝐑}}`$ and the electric field intensity $`𝐄=\frac{𝐝\phi }{\mathrm{𝐝𝐑}}`$
$$\frac{dr_s}{d𝐑}=\frac{r_s^2}{e}\left[\frac{\frac{m_i}{Z}𝐠+\delta \mathrm{𝐪𝐄}}{\delta q}\right].$$
(17)
This equation has the following solution
$$\frac{dr_s}{d𝐑}=0$$
(18)
and
$$\frac{m_i}{Z}𝐠+\delta \mathrm{𝐪𝐄}=\mathrm{𝟎}.$$
(19)
In plasma, the equilibrium value of the electric field on nuclei according to Eq.(4) is determined by Eq.(1) as well as in a metal. But there is one more additional effect in plasma. Simultaneously with the supporting of nuclei in equilibrium, each cell obtains an extremely small positive electric charge.
As $`div𝐠=4\pi Gnm_i`$ and $`div𝐄=4\pi n\delta q`$, the gravity-induced electric charge in a cell
$$\delta q=\sqrt{G}\frac{m_i}{Z}10^{18}e,$$
(20)
where $`G`$ is the gravity constant.
However, because the sizes of bodies may be very large, the electric field intensity may be very large as well
$$𝐄=\frac{𝐠}{\sqrt{𝐆}}.$$
(21)
In accordance with Eqs.(18,19), the action of gravity on matter is compensated by the electric force and the gradient of pressure is absent.
Thus, a celestial body is electrically neutral as a whole, because the positive volume charge is concentrated inside the charged core and the negative electric charge exists on its surface and so one can infer gravity-induced electric polarization of a body.
## 3 Pressure distribution inside a celestial body.
As at the surface of a celestial body pressure is absent, near this surface there is always a stratum where plasma and polarization are absent. For the large stars, the size of this stratum is insignificant. But for a small planet it can comprise a substantial part of a planet, and thus, only a small relatively internal region will be polarized. At the surface of this core, the electric field intensity falls to zero. The jump in the electric field intensity is accompanied on the surface of the core by the pressure jump $`\mathrm{\Delta }p(R_N)`$(-). The important astrophysical consequence of the GIEP effect is the redistribution of the matter density inside a celestial body. In a celestial body, consisting of matter with an atomic structure, density and pressure grow monotonously with depth. In a celestial body, consisting of electron-nuclear plasma, the GIEP effect results in the fact that the pressure gradient inside the polarized core is absent and the matter density is constant. Pressure affecting the matter inside this body is equal to the pressure jump at the surface of the core
$$p=\mathrm{\Delta }p(R_N)=\frac{E(R_N)^2}{8\pi }=\frac{2\pi }{9}G\gamma ^2R_N^2,$$
(22)
where $`\gamma `$ is the matter density in the core and R<sub>N</sub> is the radius of the core.
One can say that this pressure jump is due to the existence of the polarization jump or, which is the same, the existence of the bonded surface charge, which is formed by electron pushed out from the core and which makes the total charge of the celestial body equal to zero.
## 4 Earth’s structure.
It is important, that the GIEP effect gives the possibility to construct the intrinsically self-consistent theory of the Earth . Although it is rather a solution of a geophysical problem than an astrophysical effect.
Earlier models of the Earth assumed the existence of the monotonous dependence of pressure inside the planet. The division of the Earth into the core and the mantle was explained by the fact that at the creation of the Earth, on its share a certain amount of iron (and other heavy metals) and also a necessary amount of stone were given out. The core consists of metals and the mantle consists of stone. In these models, it was necessary to fit the parameters to get the densities of core and mantle and their sizes. It is not necessary to introduce any free parameters into the Earth theory based on the GIEP effect. Assuming that the Earth consists of homogeneous matter, the division on core and mantle is explained by the existence of the pressure jump on the surface of the core Eq.(22). The basic results of this theory are reduced to the calculation of the following five values:
a) the radius of the Earth’s core;
b) the density of core matter;
c) the density jump on the core-mantle boundary;
d) the mass related to one electron of the Fermi gas in the core;
e) the electric polarization of the core.
To express it in appropriate equations, one should substitute the following four parameters (the gravitational constant $`G`$ is known):
a) the mass of the Earth;
b) the radius of the Earth;
c) the matter density on the surface of the Earth;
d) the bulk module of matter at the surface of the Earth.
Thus, other parameters can be obtained, for example, the pressure distribution inside the Earth. The basic results of this theory are shown in Fig.1.
In addition, from the obtained data it is possible to calculate the angular momentum of the Earth. This calculation gives the value of $`0.339MR^2`$. It is in agreement with the measured value of $`0.331MR^2`$ within several percent of the accuracy.
It is possible to calculate the magnetic moment of the Earth.
Apparently, using the appropriate data of other planets (the mass, the size, and the properties of matter at the surface), it is possible to construct models of these planets. It can be made, if these planets have electrically polarized cores and corresponding magnetic fields.
## 5 The gyromagnetic ratio of a celestial body
Another astrophysical consequence of the GIEP effect is coupled by the rotation of celestial bodies about their axes. A celestial body is electroneutral as a whole. The positive volume charge is concentrated inside the core and the negative charge is located at the surface of the core. When rotating, they move on different radii. As a result, all celestial bodies, when the GIEP effect is present, obtain magnetic moments
$$\mu =\frac{2}{15}\frac{4\pi }{3c}\rho \mathrm{\Omega }R_N^5.$$
(23)
If the size of the body is sufficiently large, the core radius $`R_N`$ does not differ significantly from its external radius R. For this celestial body, the angular momentum of the core coincides by the order of magnitude with the angular momentum of the body as a whole
$$L=\frac{2}{5}M\mathrm{\Omega }R^2$$
(24)
where $`M=\frac{4\pi }{3}\gamma R^3`$ is the mass of a celestial body and $`\mathrm{\Omega }`$ is the velocity of rotation.
Finally, the gyromagnetic ratios for these bodies should be close to the universal value
$$\frac{\mu }{L}=\frac{G^{1/2}}{3c}.$$
(25)
The values of $`\mu `$(L) for all celestial bodies (for which they are known today) are shown in Fig.2. The data for planets are taken from , the data for stars are taken from , and for pulsars - from .
As can be seen from the figure with the logarithmic accuracy, all celestial bodies - stars, planets, and pulsars - really have the gyromagnetic ratio close to the universal value $`\frac{G^{1/2}}{3c}`$. Only the data for the Moon fall out, because its size and inner pressure are too small to create an electrically polarized core. The estimation of the magnetic moment of the Earth within the frame of the theory mentioned above gives $`\mu 410^{25}Gscm^3`$. It is almost precisely one half from the observed value of $`8.0510^{25}Gscm^3`$. For some planets, the values of magnetic moments are in a good agreement with Eq.(25) but they have an opposite sign. Apparently, it means that the hydrodynamic mechanism also plays a certain role.
For the majority of pulsars, there are estimations of magnetic fields obtained using a number of model assumptions . It is impossible to consider these data as the data of measurements, but nevertheless, they also agree in certain way with Eq.(25),(Fig.3)
## 6 The masses of celestial bodies.
The important astrophysical outcome of the GIEP effect is a discrete distribution of masses of celestial bodies. This spectrum is a result of the fact that electron-nuclear plasmas can exist in various states.
The equation of state of matter subjected to high pressure is usually described as a polytrope :
$$p=C\gamma ^{1+\frac{1}{k}},$$
(26)
where C is the dimensional constant,
k is the polytropy.
### 6.1 Nonrelativistic electron-nuclear plasma.
At relatively small pressure, substances are transmuted into nonrelativistic electron-nuclear (or electron-ion) plasma. It is peculiar to conditions existing inside cores of planets. According to , the state equation of the nonrelativistic electron-nuclear plasma (characterized by the polytropy k=3/2) is
$$p_{(3/2)}=\frac{\left(3\pi ^2\right)^{2/3}\mathrm{}^2\gamma ^{5/3}}{5m_e(\beta m_p)^{5/3}},$$
(27)
where $`\beta m_p`$ is the mass of matter related to one electron of the Fermi gas system and m<sub>p</sub> is the proton mass.
If the pressure inside a celestial body is formed by the GIEP effect and is determined by Eq.(22), than from Eq.(27) for the nonrelativistic Fermi gas of electrons, we obtain the steady-state value of mass for a core of planet
$$M_{(3/2)}=C_{(3/2)}\left(\frac{\mathrm{}^2}{Gm_em_p}\right)^{3/2}\frac{\gamma ^{1/2}}{\beta ^{5/2}m_p},$$
(28)
where C$`{}_{(3/2)}{}^{}=\frac{54\pi }{5}\left(\frac{\pi }{10}\right)^{1/2}19.`$
The dependence of Eq.(28) is shown in Fig.4. Therefore, any planet (even consisting from pure hydrogen) should have a mass less than 10$`{}_{}{}^{31}g`$ (if its density is approximately equal to $`1g/cm^3`$).
In Fig.4 the masses of the planets of the Solar system are marked. The mass of the Jupiter is $`1.910^{31}g`$. It is close to the specified limit. For the Jupiter Eq.(28) gives $`\beta 2`$. It is according to the data that the large planets have the deuterium-helium composition. For other planets the mantle is not small in comparison with their sizes. For this reason, Eq.(28) can give an excessive estimation for other planets.
### 6.2 Relativistic electron-nuclear plasma.
When the pressure increases, the substances are transmuted into relativistic electron-nuclear plasma (the polytropy k= 3). Its state equation is
$$p_{(3)}=\frac{\left(3\pi ^2\right)^{1/3}\mathrm{}c\gamma ^{4/3}}{4m_p^{4/3}\beta ^{4/3}}$$
(29)
If this plasma is originated by the GIEP effect, then the steady-state value of mass of a star consisting of it, according to Eqs.(22,29) is
$$M_{(3)}=C_{(3)}A_{}^{3/2}\frac{m_p}{\beta ^2},$$
(30)
where the dimensionless constants are
$$A_{}=\left(\frac{\mathrm{}c}{Gm_p^2}\right)=1.5410^{38}$$
(31)
and C$`{}_{(3)}{}^{}=\left(1.5^5\pi \right)^{1/2}4.88.`$
Because of the electric neutrality, one proton should be related to electron of the Fermi gas of plasma. The existence of one neutron per proton is characteristic for a substance consisting of light nuclei. The quantity of neutrons grows approximately to 1.8 per proton for the heavy nuclei substance. Therefore, it is necessary to expect that inside stars $`2<\beta <2.8`$.
The masses of stars can be measured with a considerable accuracy, if these stars compose a binary system. There are almost 200 double stars which masses are known with the required accuracy . Among these stars there are giants, dwarfs, and stars of the main sequence. Their average masses are described by the equality
$$M_{star}=\left(1.36\pm 0.05\right)M_{},$$
(32)
where $`M_{}`$ is the mass of the Sun.
The center of this distribution (Fig.5) corresponds to Eq.(30) at $`\beta 2.6`$.
### 6.3 Ultrarelativistic electron-nuclear plasma.
Further increase in pressure transmutes substances into ultrarelativistic plasma. Then nuclear reactions of capture of electrons by nuclei become favorable and the neutronization of matter takes place. Equilibrium pressure of ultrarelativistic plasma does not depend on its density. It is formally characterized by the polytropy k=-1 and its state equation is
$$p_{(1)}=\frac{\mathrm{\Delta }^4}{12\pi ^2\left(\mathrm{}c\right)^3}.$$
(33)
Here $`\mathrm{\Delta }`$ is the difference between the energy of the initial nucleus and the energy of the daughter nucleus.
The equilibrium mass of a star, consisting of ultrarelativistic plasma, according to Eqs.(22),Eq.(33) is
$$M_{(1)}=C_{(1)}\left(\frac{\mathrm{\Delta }^6}{\left(\mathrm{}c\right)^{9/2}G^{3/2}\gamma ^2}\right),$$
(34)
where C$`{}_{(1)}{}^{}=\frac{1}{4\pi ^3}\left(\frac{3}{2\pi }\right)^{1/2}610^3.`$
According to the astrophysical data a neutronization of matter takes place at the density $`\gamma 10^7\frac{g}{cm^3}`$. Thus, Eq.(34) gives
$$M_{(1)}810^{32}g0.4M_{}.$$
(35)
Certainly this result is the rough estimation on the order of magnitude only, but it is in a satisfactory agreement with measurements of the astronomers related to masses of white dwarfs from double systems.
### 6.4 Nonrelativistic neutron matter.
At higher pressure, the substance is transmuted into a nonrelativistic neutron matter with a small impurity of protons and electrons . The state equation of the nonrelativistic neutron matter will coincide with Eq.(22) with a replacement of $`m_e`$ with $`m_p`$
$$p_{(3/2)}^{(n)}=\frac{\left(3\pi ^2\right)^{2/3}\mathrm{}^2\gamma ^{5/3}}{5m_p^{8/3}\beta ^{5/3}}.$$
(36)
Together with Eq.(22), it gives the equilibrium mass of the nonrelativistic neutron star
$$M_{(3/2)}^{(n)}=C_{(3/2)}\left(\frac{\mathrm{}^2}{G}\right)^{3/2}\frac{\gamma ^{1/2}}{m_p^4\beta ^{5/2}}.$$
(37)
As the density $`\gamma 410^{13}g/cm^3`$ and $`\beta =2.6`$
$$M_{(3/2)}^{(n)}0.05M_{}.$$
(38)
The astronomers have not found such neutron stars.
### 6.5 Relativistic neutron matter.
With further increase in pressure, the neutron Fermi gas becomes a relativistic one. Its state equation completely coincides with the state equation of the relativistic Fermi gas of electrons Eq.(22)
$$p_{(3)}^{(n)}=\frac{\left(3\pi ^2\right)^{1/3}\mathrm{}c\gamma ^{4/3}}{4m_p^{4/3}\beta ^{4/3}}.$$
(39)
As well as the masses of relativistic stars, the masses of relativistic pulsars do not depend on their density and can be directly expressed through world constants
$$M_{(3)}^{(n)}=C_{(3)}A_{}^{3/2}\frac{m_p}{\beta ^2}$$
(40)
(at $`\beta =1`$ for the pure neutron Fermi gas).
As it is specified in , at this density of matter it is necessary to take into account nuclear forces and the presence inside nuclear matter except neutrons also of protons, $`\pi ^{}`$mesons, and electrons. It can be made using $`\beta `$ as a parameter of the correction.
The mass of the neutron star can be measured with a considerable accuracy if it enters into a binary system. The astronomers have found 16 radio-pulsars and 7 X-pulsars in binary systems. They all are located in a very narrow mass interval (Fig.5)
$$M_{pulsar}=(1.38\pm 0.03)M_{}.$$
(41)
The center of this distribution corresponds to Eq.(40) with the correction parameter $`\beta 2.6`$ just as for relativistic stars. Thus, we come to the conclusion that Eq.(40) on the order of magnitude correctly describes the results of astronomical observations.
## 7 Conclusion.
It is expedient to underline the basic obtained results in summary.
1. The developed theory defines a concept of the steady-state values of masses of celestial bodies related to their equations of state and gives the possibility to calculate these values.
2. It gives the new way for the determination of the substance density distribution inside celestial bodies. According to early models, it was supposed that density of a substance inside celestial bodies grows more or less monotonically with depth and at the centre of a star, the density has the greatest value and even a black hole may exist there. According to the developed theory, the density of a substance inside a star is constant.
3. It is interesting to emphasize that the ”biography” of such a star appears much poorer than in the Chandrasecar model.
There cannot exist a black hole inside a star, and it should not collapse with a temperature decrease. All the considered effects are based on an equilibrium of the Fermi system. Temperature does not influence the parameters of relativistic plasma. Therefore, a star with a mass close to the steady-state value (Eq.(30)) is in a stable equilibrium not depending on temperature. The existing stars should retain the stability at any (even at zero) temperature. Therefore, a collapse of the already existing stars apparently is impossible. The instability of a star can arise with burning out of light nuclei - deuterium and helium - and with a related increasing of $`\beta `$. This growth leads to the reduction of a steady-state value of mass (Eq.(30)) and, probably, to the distraction of the stars wiht masses more than the steady-state value.
4. The developed theory determines the simple and essential mechanism of generation of the magnetic field by celestial bodies. All early models tried to solve the basic problem - to calculate the magnetic field of a celestial body. Such a statement of the basic problem of planetary and stellar magnetism is unacceptable at present. Space flights and a development of astronomy discovered a remarkable and earlier unknown fact: the magnetic moments of all celestial bodies are proportional to their angular momenta and the proportionality coefficient is determined by the ratio of world constants. The explanation of this phenomenon is the basic problem of planetary and stars magnetism nowadays. Early models cannot explain this phenomenon. The developed theory used for this explanation a standard mechanism.
5. It is possible to consider that now the predicted steady-state values of masses of celestial bodies and the predicted values of their magnetic moments are in a satisfactory agreement with the data of observations. But it is tempting to obtain these data to closer limit of accuracy. Two arrows in the upper part of Fig.5 mark masses of stars consisting of extremely light and heavy nuclei. These values are obtained from Eq.(30) without the use of any fitting parameters. In agreement wiht the developed theory, if stars have a ”usual” chemical composition, there must be no stars outside of this interval (or these stars should be unstable). The histogram on Fig.5 is somewhat wider. It is interesting to understand, whether there is a principal deviation from the developed theory or it is a result of measuring errors. First of all, it requires a more careful and precise measurement of masses of binary stars.
|
warning/0002/hep-ph0002149.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
In a modern theoretical context, one generally expects nonzero neutrino masses and associated lepton mixing. Experimentally, there has been accumulating evidence for such masses and mixing. All solar neutrino experiments (Homestake, Kamiokande, SuperKamiokande, SAGE, and GALLEX) show a significant deficit in the neutrino fluxes coming from the Sun . This deficit can be explained by oscillations of the $`\nu _e`$’s into other weak eigenstate(s), with $`\mathrm{\Delta }m_{sol}^2`$ of the order $`10^5`$ eV<sup>2</sup> for solutions involving the Mikheev-Smirnov-Wolfenstein (MSW) resonant matter oscillations or of the order of $`10^{10}`$ eV<sup>2</sup> for vacuum oscillations. Accounting for the data with vacuum oscillations (VO) requires almost maximal mixing. The MSW solutions include one for small mixing angle (SMA) and one with essentially maximal mixing (LMA).
Another piece of evidence for neutrino oscillations is the atmospheric neutrino anomaly, observed by Kamiokande , IMB , SuperKamiokande with the highest statistics, and by Soudan and MACRO . This data can be fit by the inference of $`\nu _\mu \nu _x`$ oscillations with $`\mathrm{\Delta }m_{atm}^23.5\times 10^3`$ eV<sup>2</sup> and maximal mixing $`\mathrm{sin}^22\theta _{atm}=1`$ . The identification $`\nu _x=\nu _\tau `$ is preferred over $`\nu _x=\nu _{sterile}`$ at about the $`2.5\sigma `$ level , and the identification $`\nu _x=\nu _e`$ is excluded by both the Superkamiokande data and the Chooz experiment .
In addition, the LSND experiment has reported observing $`\overline{\nu }_\mu \overline{\nu }_e`$ and $`\nu _\mu \nu _e`$ oscillations with $`\mathrm{\Delta }m_{LSND}^20.11`$ eV<sup>2</sup> and a range of possible mixing angles, depending on $`\mathrm{\Delta }m_{LSND}^2`$ . This result is not confirmed, but also not completely ruled out, by a similar experiment, KARMEN .
There are currently intense efforts to confirm and extend the evidence for neutrino oscillations in all of the various sectors – solar, atmospheric, and accelerator. Some of these experiments are running; these include the Sudbury Neutrino Observatory, SNO, and the K2K long baseline experiment between KEK and Kamioka. Others are in development and testing phases, such as BOONE, MINOS, the CERN - Gran Sasso program, KAMLAND, and Borexino . Among the long baseline neutrino oscillation experiments, the approximate distances are $`L250`$ km for K2K, 730 km for both MINOS, from Fermilab to Soudan and the proposed CERN-Gran Sasso experiments. The sensitivity of these experiments is projected to reach down roughly to the level $`\mathrm{\Delta }m^210^3`$eV<sup>2</sup>. There is strong motivation for another generation of experiments with even higher sensitivity that can confirm the $`\nu _\mu \nu _\tau `$ transition with the values of $`\mathrm{\Delta }m_{atm}^2`$ and $`\mathrm{sin}^22\theta _{atm}`$ reported so far and carry out further measurements of various neutrino oscillation channels.
Recently, there has been considerable interest in the idea of a muon storage ring that would serve as a “neutrino factory”, i.e., a source of quite high intensity, flavor-pure neutrino and antineutrino beams: $`\nu _\mu +\overline{\nu }_e`$ ($`\overline{\nu }_\mu +\nu _e`$) from stored $`\mu ^{}`$’s ($`\mu ^+`$’s) respectively -. Given the very high intensities anticipated to be of order $`10^{20}`$ and perhaps even $`10^{21}`$ muon decays per year in various preliminary studies, one can envision neutrino oscillation experiments with quite long baselines of order several thousand km, with commensurate sensitivity to various neutrino oscillation channels.
One of the appeals of the muon storage ring/neutrino factory is that one can measure several different neutrino oscillation transitions, using both the $`\nu _\mu `$ ($`\overline{\nu }_\mu `$) and $`\overline{\nu }_e`$ ($`\nu _e`$) from a $`\mu ^{}`$ ($`\mu ^+`$) beam. In this paragraph we assume a $`\mu ^{}`$ beam for definiteness (figures below are shown for neutrinos from both stored $`\mu ^+`$ and $`\mu ^{}`$ beams). In addition to a high-statistics measurement of $`\nu _\mu \nu _\mu `$, as a disappearance test for the $`\nu _\mu \nu _\tau `$ oscillation, one has also various other channels. Among these are $`\overline{\nu }_e\overline{\nu }_\mu `$, for which the signal is a “wrong-sign muon”, $`\mu ^+`$, and $`\overline{\nu }_e\overline{\nu }_\tau `$, which, in about 18 $`\%`$ of its decays, would also yield a wrong-sign muon. The measurement of the muon charge would be possible with either a magnetized iron detector or a combination of a massive water Čerenkov detector followed by a muon spectrometer. With sufficient detector capabilities, one could also search for $`\tau `$ appearance, as is envisioned by the ICANOE and OPERA experiments at Gran Sasso , although this requires neutrino energies $`E_\nu >20`$ GeV to avoid kinematic suppression of $`\tau `$ production.
An important effect that must be taken into account in such experiments concerns the matter-induced oscillations which neutrinos undergo along their flight path through the Earth from the source to the detector. Given the typical density of the earth, matter effects are important for the neutrino energy range $`EO(10)`$ GeV and $`\mathrm{\Delta }m_{atm}^23\times 10^3`$ eV<sup>2</sup> values relevant for the long baseline experiments, in particular, for the oscillation channels involving $`\nu _e`$, as we shall show below. Matter effects can also be important for the neutrino energy range $`EO(10)`$ MeV and $`\mathrm{\Delta }m^210^5`$ eV<sup>2</sup> involved in MSW solutions to the solar neutrino problem. After the initial discussion of matter-induced resonant neutrino oscillations in , an early study of these effects including three generations was carried out in . The sensitivity of an atmospheric neutrino experiment to small $`\mathrm{\Delta }m^2`$ due to the long baselines and the necessity of taking into account matter effects was discussed e.g., in . After Ref. , many analyses were performed in the 1980’s of the effects of resonant neutrino oscillations on the solar neutrino flux, and matter effects in the Earth were studied, e.g., and , which also discussed the effect on atmospheric neutrinos (see also the review ). Recent papers on matter effects relevant to atmospheric neutrinos include . Early studies of matter effects on long baseline neutrino oscillation experiments were carried out in . More recent analyses relevant to neutrino factories include , -.
In this paper we shall present calculations of the matter effect for parameters relevant to possible long baseline neutrino experiments envisioned for the muon storage ring/neutrino factory. In particular, we compare the results obtained with constant density along the neutrino path versus results obtained by incorporating the actual density profiles. We study the dependence of the oscillation signal on both $`E/\mathrm{\Delta }m_{atm}^2`$ and on the angles in the leptonic mixing matrix. We also comment on the influence of $`\mathrm{\Delta }m_{sol}^2`$ and CP violation on the oscillations. Some of our results were presented in Ref. . Additional recent studies include -.
In a hypothetical world in which there were only two neutrinos, $`\nu _\mu `$ and $`\nu _\tau `$, the $`\nu _\mu \nu _\tau `$ oscillations in matter would be the same as in vacuum, since both have the same forward scattering amplitude, via $`Z`$ exchange, with matter. However, in the realistic case of three generations, because of the indirect involvement of $`\nu _e`$ due to a nonzero $`U_{13}`$, and because of the fact that $`\nu _e`$ has a different forward scattering amplitude off of electrons, involving both $`Z`$ and $`W`$ exchange, there will be a matter-induced oscillation effect on $`\nu _\mu \nu _\tau `$ (as well as other channels).
We consider the usual three flavors of active neutrinos, with no light sterile (=electroweak-singlet) neutrinos. This is sufficient to describe the solar and atmospheric neutrino deficit. If one were also to include the LSND experiment, then, to obtain a reasonable fit, one would be led to include light electroweak-singlet neutrinos. Since the LSND experiment has not so far been confirmed, we shall, while not prejudging the outcome of the BOONE experiment, not include this in our fit. We calculate oscillation probabilities in the full $`3\times 3`$ mixing case and we study when $`\mathrm{\Delta }m_{sol}^2`$ can be relevant. In most cases there is only one mass scale relevant for long baseline neutrino oscillations, $`\mathrm{\Delta }m_{atm}^2\mathrm{few}\times 10^3`$ eV<sup>2</sup> and we work with the hierarchy
$$\mathrm{\Delta }m_{21}^2=\mathrm{\Delta }m_{sol}^2\mathrm{\Delta }m_{31}^2\mathrm{\Delta }m_{32}^2=\mathrm{\Delta }m_{atm}^2$$
(1)
In our work we take into account the actual profile of the Earth, as given by geophysical seismic data and compare the results with those calculated using the approximation of average density along the path of the neutrino. Further, when only one mass squared difference is relevant, we present the oscillation probabilities as functions of $`E/\mathrm{\Delta }m^2`$, where here and below, $`E=E_\nu `$. This way of presenting the results is useful since, for a given $`L`$ value, it shows the matter effect for a wide range of $`E`$ and $`\mathrm{\Delta }m^2`$ and hence can serve as an input in the choice of optimal beam energy (along with other considerations such as the cross section dependence $`\sigma E`$ and the beam divergence $`(LE)^2`$, which, together, favor higher values of $`E`$ to achieve a high event rate). We study how these oscillation probabilities vary with the different input parameters and discuss the influence of the matter effects on the sensitivity to each of these parameters.
## II Theoretical Framework
We first recall the form of the lepton mixing matrix. Let us denote the flavor vectors of SU(2) $`\times `$ U(1) nonsinglet neutrinos as $`\nu =(\nu _e,\nu _\mu ,\nu _\tau )`$ and the vector of electroweak-singlet neutrinos as $`\chi =(\chi _1,..,\chi _{n_s})`$. The Dirac and Majorana neutrino mass terms can then be written compactly as
$$_m=\frac{1}{2}(\overline{\nu }_L\overline{\chi ^c}_L)\left(\begin{array}{cc}M_L& M_D\\ (M_D)^T& M_R\end{array}\right)\left(\begin{array}{c}\nu _R^c\\ \chi _R\end{array}\right)+h.c.$$
(2)
where $`M_L`$ is the $`3\times 3`$ left-handed Majorana mass matrix, $`M_R`$ is a $`n_s\times n_s`$ right-handed Majorana mass matrix, and $`M_D`$ is the 3-row by $`n_s`$-column Dirac mass matrix. In general, all of these are complex, and $`(M_L)^T=M_L,(M_R)^T=M_R`$. Without further theoretical input, the number $`n_s`$ of electroweak singlet neutrinos is not determined. For example, in the minimal SU(5) grand unified theory (GUT), $`n_s=0`$, while in SO(10), $`n_s=3`$. Within this theoretical context, since the terms $`\chi _{jR}^TC\chi _{kR}`$ are electroweak singlets, the associated coefficients, which comprise the elements of the matrix $`M_R`$, would not be expected to be related to the electroweak symmetry breaking scale, but instead, would be expected to be much larger, plausibly of order the GUT scale. Furthermore, the left-handed Majorana mass terms can only arise via operators of dimension at least 5, such as
$$𝒪=\frac{1}{M_X}\underset{a,b}{}h_{a,b}(ϵ_{ik}ϵ_{jm}+ϵ_{im}ϵ_{jk})\left[_{aL}^{Ti}C_{bL}^j\right]\varphi ^k\varphi ^m+h.c.$$
(3)
where $`_{La}=(\nu _\mathrm{}_a,\mathrm{}_a)_L^T`$ is the left-handed, $`I=1/2`$, $`Y=1`$ lepton doublet with generation index $`a`$ ($`a=1`$, 2, or 3), where $`\mathrm{}_a=e,\mu ,\tau `$, for $`a=1,2,3`$, $`M_X`$ denotes a generic mass scale characterizing the origin of this term, and $`\varphi `$ is the standard model Higgs or $`H_u`$ in the supersymmetric standard model. Because (3) is a nonrenormalizable operator, the success of the standard model as a renormalizable field theory then implies that $`M_X`$ is much larger than the scale of electroweak symmetry breaking, and, within a GUT context, $`M_X`$ would be of order the GUT scale, as with $`M_R`$. The terms arising from the vacuum expectation values of the Higgs doublets then make up the submatrix $`M_L`$. The resultant diagonalization of the matrix in eq. (2) then naturally leads to a set of 3 light masses for the three known neutrinos, generically of order $`m_\nu m_D^2/M_R`$, and $`n_s`$ very large masses generically of order $`M_R`$, for the electroweak singlet neutrinos. This seesaw mechanism is very appealing, since it can provide a plausible explanation for why the known neutrinos are so light . Although the full leptonic mixing matrix is $`(3+n_s)\times (3+n_s)`$ dimensional, the light and heavy neutrinos largely decouple from each other so that, to a high degree of accuracy, one can describe the linear combinations of the $`(3+n_s)`$ mass eigenstates that form $`(3+n_s)`$ weak eigenstates in a decoupled manner, using a simple $`3\times 3`$ matrix $`U`$, which, to high accuracy, is unitary, for the known neutrinos. This is determined by the diagonalization of the effective $`3\times 3`$ light neutrino mass matrix
$$M_\nu =M_DM_R^1M_D^T$$
(4)
and an $`n_s\times n_s`$ matrix for the heavy neutrino sector, which matrix will not be used here. The lepton mixing matrix can then be written as the unitary matrix
$`U=R_{23}KR_{13}K^{}R_{12}K^{}=\left(\begin{array}{ccc}c_{12}c_{13}& s_{12}c_{13}& s_{13}e^{i\delta }\\ s_{12}c_{23}c_{12}s_{23}s_{13}e^{i\delta }& c_{12}c_{23}s_{12}s_{23}s_{13}e^{i\delta }& s_{23}c_{13}\\ s_{12}s_{23}c_{12}c_{23}s_{13}e^{i\delta }& c_{12}s_{23}s_{12}c_{23}s_{13}e^{i\delta }& c_{23}c_{13}\end{array}\right)K^{}`$ (5)
where $`R_{ij}`$ is the rotation matrix in the $`ij`$ subspace, $`c_{ij}=\mathrm{cos}\theta _{ij}`$, $`s_{ij}=\mathrm{sin}\theta _{ij}`$, $`K=diag(e^{i\delta },1,1)`$ and $`K^{}`$ involves further possible phases due to Majorana mass terms that will contribute here.
In passing, we note that although this theoretical context is appealing, various modifications are possible. For example, string theory generically involves certain moduli fields which are singlets under the standard model gauge group, have flat superpotentials, and hence are massless in perturbation theory down to the energy scale where supersymmetry is broken. The spinor component fields, modulinos, can act as electroweak-singlet neutrinos, and may well have masses much less than the GUT scale (e.g. ). Moreover, models with a low string scale $`<<M_{Planck}`$ and large compact dimensions (e.g., ) also have implications for neutrino phenomenology. Here we shall work within the conventional seesaw-type scenario because of its simplicity and success in accounting for the most striking known feature of neutrinos, namely the fact that they are so light compared with the other known fermions.
For our later discussion it will be useful to record the formulas for the various relevant neutrino oscillation transitions. In the absence of any matter effect, the probability that a (relativistic) weak neutrino eigenstate $`\nu _a`$ becomes $`\nu _b`$ after propagating a distance $`L`$ is
$`P(\nu _a\nu _b)`$ $`=`$ $`\delta _{ab}4{\displaystyle \underset{i>j=1}{\overset{3}{}}}Re(K_{ab,ij})\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{ij}^2L}{4E}}\right)+`$ (6)
$`+`$ $`4{\displaystyle \underset{i>j=1}{\overset{3}{}}}Im(K_{ab,ij})\mathrm{sin}\left({\displaystyle \frac{\mathrm{\Delta }m_{ij}^2L}{4E}}\right)\mathrm{cos}\left({\displaystyle \frac{\mathrm{\Delta }m_{ij}^2L}{4E}}\right)`$ (7)
where
$$K_{ab,ij}=U_{ai}U_{bi}^{}U_{aj}^{}U_{bj}$$
(8)
and
$$\mathrm{\Delta }m_{ij}^2=m(\nu _i)^2m(\nu _j)^2$$
(9)
Recall that in vacuum, CPT invariance implies $`P(\overline{\nu }_b\overline{\nu }_a)=P(\nu _a\nu _b)`$ and hence, for $`b=a`$, $`P(\overline{\nu }_a\overline{\nu }_a)=P(\nu _a\nu _a)`$. For the CP-transformed reaction $`\overline{\nu }_a\overline{\nu }_b`$ and the T-reversed reaction $`\nu _b\nu _a`$, the transition probabilities are given by the right-hand side of (7) with the sign of the imaginary term reversed. (Below we shall assume CPT invariance, so that CP violation is equivalent to T violation.) For most sets of parameters, only one mass scale is relevant for the neutrino oscillations of interest here, namely
$$\mathrm{\Delta }m_{atm}^2=\mathrm{\Delta }m_{32}^2$$
(10)
In this case, CP (T) violation effects are negligibly small, so that in vacuum
$$P(\overline{\nu }_a\overline{\nu }_b)=P(\nu _a\nu _b)$$
(11)
$$P(\nu _b\nu _a)=P(\nu _a\nu _b)$$
(12)
In the absence of T violation, the second equality (12) would still hold in matter, but even in the absence of CP violation, the first equality (11) would not hold. With the hierarchy (1), the expressions for the specific oscillation transitions are
$`P(\nu _\mu \nu _\tau )`$ $`=`$ $`4|U_{33}|^2|U_{23}|^2\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (13)
$`=`$ $`\mathrm{sin}^2(2\theta _{23})\mathrm{cos}^4(\theta _{13})\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (15)
$`P(\nu _e\nu _\mu )`$ $`=`$ $`4|U_{13}|^2|U_{23}|^2\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (16)
$`=`$ $`\mathrm{sin}^2(2\theta _{13})\mathrm{sin}^2(\theta _{23})\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (18)
$`P(\nu _e\nu _\tau )`$ $`=`$ $`4|U_{33}|^2|U_{13}|^2\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (19)
$`=`$ $`\mathrm{sin}^2(2\theta _{13})\mathrm{cos}^2(\theta _{23})\mathrm{sin}^2\left({\displaystyle \frac{\mathrm{\Delta }m_{atm}^2L}{4E}}\right)`$ (21)
In neutrino oscillation searches using reactor antineutrinos, i.e. tests of $`\overline{\nu }_e\overline{\nu }_e`$, the two-species mixing hypothesis used to fit the data is
$$P(\nu _e\nu _e)=1\mathrm{sin}^2(2\theta _{reactor})\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{reactor}^2L}{4E}\right)$$
(22)
where $`\mathrm{\Delta }m_{reactor}^2`$ is the squared mass difference relevant for $`\overline{\nu }_e\overline{\nu }_x`$. In particular, in the upper range of values of $`\mathrm{\Delta }m_{atm}^2`$, since the transitions $`\overline{\nu }_e\overline{\nu }_\mu `$ and $`\overline{\nu }_e\overline{\nu }_\tau `$ contribute to $`\overline{\nu }_e`$ disappearance, one has
$$P(\nu _e\nu _e)=1\mathrm{sin}^2(2\theta _{13})\mathrm{sin}^2\left(\frac{\mathrm{\Delta }m_{atm}^2L}{4E}\right)$$
(23)
i.e., $`\theta _{reactor}=\theta _{13}`$, and the Chooz reactor experiment yields the bound
$$\mathrm{sin}^2(2\theta _{13})<0.10$$
(24)
which is also consistent with conclusions from the SuperK data analysis .
Further, the quantity “$`\mathrm{sin}^2(2\theta _{atm})`$” often used to fit the data on atmospheric neutrinos with a simplified two-species mixing hypothesis, is, in the three-generation case,
$$\mathrm{sin}^2(2\theta _{atm})\mathrm{sin}^2(2\theta _{23})\mathrm{cos}^4(\theta _{13})$$
(25)
Hence for small $`\theta _{13}`$, as implied by (24), it follows that, to good accuracy, $`\theta _{atm}=\theta _{23}`$.
## III Calculation of Matter Effects
The evolution of the flavor eigenstates of neutrinos is given by
$$i\frac{d}{\mathrm{𝑑𝑥}}\nu =\left(\frac{1}{2E}UM^2U^{}+V\right)\nu $$
(26)
where
$$\nu =U\nu _m$$
(27)
$$\nu _m=\left(\begin{array}{c}\nu _1\\ \nu _2\\ \nu _3\end{array}\right)$$
(28)
$$M^2=\left(\begin{array}{ccc}m_1^2& 0& 0\\ 0& m_2^2& 0\\ 0& 0& m_3^2\end{array}\right),V=\left(\begin{array}{ccc}\sqrt{2}G_FN_e& 0& 0\\ 0& 0& 0& \\ 0& 0& 0\end{array}\right)$$
(29)
Here $`N_e`$ is the electron number density and we have $`\sqrt{2}G_FN_e`$ \[eV\]$`=7.6\times 10^{14}Y_e\rho `$ \[g/cm<sup>3</sup>\].
The atmospheric neutrino data suggests almost maximal mixing in the $`23`$ sector. However, a small but non-zero $`s_{13}`$ is still allowed, and this produces the matter effect in the traversal of neutrinos through the Earth. We use the bound (24) on $`\mathrm{sin}^2(2\theta _{13})`$ here, consistent with both the Chooz experiment and the atmospheric neutrino data .
If we assume that the solar neutrino deficiency is explained by the small mixing angle (SMA) MSW solution or by vacuum oscillations, with the hierarchy of eq. (1), it follows that, for the relevant energies $`E>1`$ GeV and path-lengths $`L10^310^4`$ km, only one squared mass scale, $`\mathrm{\Delta }m_{atm}^2`$, is important for the oscillations and the three-species neutrino oscillations can be described in terms of this quantity, $`\mathrm{\Delta }m_{atm}^2`$, and the mixing parameters $`\mathrm{sin}^2(2\theta _{23})`$, and $`\mathrm{sin}^2(2\theta _{13})`$, with negligible dependence on $`\mathrm{sin}^2(2\theta _{12})`$ and $`\delta `$.
In order to write down the probabilities of oscillation for long-baseline and atmospheric neutrinos, it is convenient to transform to a new basis defined by (e.g. )
$$\nu =R_{23}\stackrel{~}{\nu }$$
(30)
The evolution of $`\stackrel{~}{\nu }`$ is given by
$$\stackrel{~}{H}=\frac{1}{2E}KR_{13}K^{}R_{12}M^2R_{12}^{}KR_{13}^{}K^{}+V$$
(31)
In the one mass-scale approximation, this can be reduced to
$$\stackrel{~}{H}\left(\begin{array}{ccc}\frac{1}{2E}s_{13}^2\mathrm{\Delta }m_{32}^2+\sqrt{2}G_FN_e& 0& \frac{1}{2E}s_{13}c_{13}\mathrm{\Delta }m_{32}^2e^{i\delta }\\ 0& \frac{1}{2E}c_{12}^2\mathrm{\Delta }m_{21}^2& 0\\ \frac{1}{2E}s_{13}c_{13}\mathrm{\Delta }m_{32}^2e^{i\delta }& 0& \frac{1}{2E}c_{13}^2\mathrm{\Delta }m_{32}^2\end{array}\right)$$
(32)
It can be seen now that in the basis $`(\nu _e,\stackrel{~}{\nu }_\mu ,\stackrel{~}{\nu }_\tau )`$ the three-flavor evolution equation decouples and it is enough to treat the two-flavor case. We define $`S`$ and $`P`$ by
$$\left(\begin{array}{c}\nu _e\\ \stackrel{~}{\nu }_\mu \\ \stackrel{~}{\nu }_\tau \end{array}\right)(x)=S\left(\begin{array}{c}\nu _e\\ \stackrel{~}{\nu }_\mu \\ \stackrel{~}{\nu }_\tau \end{array}\right)(0)$$
(33)
and
$$P|S_{13}|^2=1|S_{33}|^2$$
(34)
Transforming back to the flavor basis $`(\nu _e,\nu _\mu ,\nu _\tau )`$, the probabilities of oscillation become
$`P(\nu _e\nu _e)`$ $`=`$ $`1P`$ (35)
$`P(\nu _e\nu _\mu )`$ $`=`$ $`P(\nu _\mu \nu _e)=s_{23}^2P`$ (36)
$`P(\nu _e\nu _\tau )`$ $`=`$ $`c_{23}^2P`$ (37)
$`P(\nu _\mu \nu _\mu )`$ $`=`$ $`1s_{23}^4P+2s_{23}^2c_{23}^2[Re(S_{22}S_{33})1]`$ (38)
$`P(\nu _\mu \nu _\tau )`$ $`=`$ $`s_{23}^2c_{23}^2[2P2Re(S_{22}S_{33})]`$ (39)
Note that for the mass hierarchy (1), the CP-violating phases disappear from the oscillation probabilities. In this case what we need to solve is the evolution equation for a two-flavor neutrino system. By subtracting from the diagonal the quantity $`\frac{1}{4E}\mathrm{\Delta }m_{32}^2+\frac{1}{\sqrt{2}}G_FN_e`$, this can be written in the form
$$i\frac{d}{\mathrm{𝑑𝑥}}\left(\begin{array}{c}\nu _a\\ \nu _b\end{array}\right)=\left(\begin{array}{cc}A(x)& B\\ B& A(x)\end{array}\right)\left(\begin{array}{c}\nu _a\\ \nu _b\end{array}\right)$$
(40)
with
$$A(x)=\frac{\mathrm{\Delta }m_{32}^2}{4E}\mathrm{cos}(2\theta _{13})\frac{G_F}{\sqrt{2}}N_e(x)$$
(41)
$$B=\frac{\mathrm{\Delta }m_{32}^2}{4E}\mathrm{sin}(2\theta _{13})$$
(42)
For our purposes, we recall that the Earth is composed of crust, mantle, liquid outer core, and solid inner core, together with additional sublayers in the mantle, with particularly strong changes in density between the lower mantle and outer core. The density profile of the Earth is shown in Fig.1. The densities of the different layers are given in Table 1 as function of the normalized radius $`x=R/R_E`$, $`R_E=6371`$ km being the radius of the Earth. The core has average density $`\rho _{core}=11.83`$ g/cm<sup>3</sup> and electron fraction $`Y_{e,core}=0.466`$, while the mantle has average density $`\rho _{mantle}=4.66`$ g/cm<sup>3</sup> and $`Y_{e,mantle}=0.494`$.
| Radius \[Km\] | Density \[g/cm<sup>3</sup>\] |
| --- | --- |
| $`01221.5`$ | $`13.08858.8381x^2`$ |
| $`1221.53480.0`$ | $`12.58151.2638x3.6426x^25.5281x^3`$ |
| $`3480.05701.0`$ | $`7.95656.4761x+5.5283x^23.0807x^3`$ |
| $`5701.05771.0`$ | $`5.31971.4836x`$ |
| $`5771.05971.0`$ | $`11.24948.0298x`$ |
| $`5971.06151.0`$ | $`7.10893.8045x`$ |
| $`6151.06346.6`$ | $`2.6910+0.6924x`$ |
| $`6346.66356.0`$ | $`2.900`$ |
| $`6356.06371.0`$ | $`2.600`$ |
Table 1 Density Profile of the Earth
Since, to very good accuracy, the Earth is spherically symmetric (Fig.1), the neutrino flight path is described only by the zenith angle $`\theta _z`$ (or $`\eta =\pi \theta _z`$). For
$$\frac{R_{i+1}}{R}<\mathrm{sin}\eta <\frac{R_i}{R}$$
(43)
the neutrinos pass through $`2i+1`$ layers in the Earth. The distances traveled by the neutrinos in each of these layers are
$`L_1=R\mathrm{cos}\eta \sqrt{R_1^2R^2\mathrm{sin}^2\eta }`$ (44)
$`L_k=L_{2i+2k}=\sqrt{R_{k1}^2R^2\mathrm{sin}^2\eta }\sqrt{R_k^2R^2\mathrm{sin}^2\eta },\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}2}ki`$ (45)
$`L_{i+1}=2\sqrt{R_i^2R^2\mathrm{sin}^2\eta }`$ (46)
Studies have been done using the average density of the Earth along the neutrino path. In this case the evolution equation can be easily solved and the probability of oscillation is given by
$$P(\nu _a\nu _b)=\mathrm{sin}^2(2\theta _m)\mathrm{sin}^2(\omega L)$$
(47)
where $`\omega =\sqrt{A^2+B^2}`$ and $`\theta _m`$ is the effective mixing in matter given by
$$\mathrm{sin}^2(2\theta _m)=\frac{\mathrm{sin}^2(2\theta )}{\mathrm{sin}^2(2\theta )+\left(\mathrm{cos}(2\theta )\frac{2\sqrt{2}G_FN_eE}{\mathrm{\Delta }m^2}\right)^2}$$
(48)
Just as was the case with the application of the MSW analysis to solar neutrinos, the key observation is that although the angle $`\theta `$, which is essentially $`\theta _{13}`$ here, is small, the vanishing of the term in parentheses in the denominator of (48) renders the effective mixing angle $`\theta _m=\pi /4`$, thereby producing maximal mixing in matter. The important point is that, given the range of densities in the layers of the Earth, and the value of $`\mathrm{\Delta }m_{atm}^23.5\times 10^3`$ eV<sup>2</sup>, this matter resonance occurs for neutrino energies of order $`O(10)`$ GeV, in the range planned for long baseline neutrino oscillation experiments. Since this effect clearly depends on the sign of $`\mathrm{\Delta }m^2`$, the measurement of matter effects can give information on this sign. When one takes account of the actual variable-density situation in the Earth, it is necessary to perform a numerical integration of the evolution equation, which we have done. We also go beyond the one mass-scale approximation and study the effect of $`\mathrm{\Delta }m_{sol}^2`$ and $`\theta _{12}`$ on the oscillations. In this case we calculate the oscillation probabilities for nonzero values of all six oscillation parameters (three angles, one phase, and two mass square differences) and discuss when the simpler cases are very good approximations.
## IV Results and Discussion
For long baseline experiments like K2K, MINOS, and CERN to Gran-Sasso, the neutrino flight path only goes through the upper mantle. The density in this region is practically constant, and the oscillation probabilities can easily be calculated. The matter effects are small, but possibly detectable for the longer baselines. We show in Fig.2 an example for $`P(\nu _\mu \nu _e)`$ relevant for MINOS or the CERN to Gran-Sasso experiments (if SMA or VO solve the solar problem).
However, there are several motivations for very long baseline experiments, since, with sufficiently high-intensity sources, these can be sensitive to quite small values of $`\mathrm{\Delta }m^2`$ and since the matter effects, being larger, can amplify certain oscillations and can, in principle, be used to get information on the sign of $`\mathrm{\Delta }m_{atm}^2`$. Hence we concentrate here on these very long baseline experiments; for these, the neutrino flight path goes through several layers of the Earth with different densities, including the lower mantle for some. We show results for the Fermilab to SLAC path length $`L2900`$ km and for $`L7330`$ km, the distance from Fermilab to Gran Sasso. Path lengths corresponding to the distance BNL to SLAC, $`L4500`$ km, and BNL to Gran Sasso, $`L6560`$ km, are also considered. We have also performed calculations for $`L`$ 9200 km, the Fermilab to SuperKamiokande path length.
In Fig.3 we compare the probabilities calculated with constant density along the neutrino path versus the results obtained by numerically integrating the evolution equation with the actual density profile of the Earth, as given by . The results are almost the same for most of the parameter range. However, at given energies, as for example for the second maximum in $`P(\nu _\mu \nu _e)`$, the correction to the probability is of the order of 20%. The results in Fig.3 are obtained for the $`L=7330`$ km distance, for which the beam goes through all layers of the mantle. In we gave a series of similar comparisons of oscillation probabilities calculated with the constant density approximation and with the actual density function. In the following, we only present results obtained with the actual density function in the Earth.
If one assumes the LMA solution to the solar neutrino problem, then for the $`L=2900`$ km baseline, both $`\mathrm{\Delta }m_{atm}^2`$ and $`\mathrm{\Delta }m_{sol}^2`$ have to be considered. In Fig.4 we show $`P(\nu _\mu \nu _e)`$ and the $`\nu _\mu `$ survival probability $`P(\nu _\mu \nu _\mu )`$ as functions of energy for $`\mathrm{\Delta }m_{atm}^2=3.5\times 10^3`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta _{23}=1`$, as suggested by the atmospheric data, and $`\mathrm{sin}^22\theta _{13}=0.1`$, the maximum value allowed, and two different choices of $`\mathrm{\Delta }m_{sol}^2`$ and $`\mathrm{sin}^22\theta _{12}`$. One choice corresponds to the LMA solution, with $`\mathrm{\Delta }m_{sol}^2=5\times 10^5`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta _{12}=0.8`$. For this LMA case, the choice of the CP violating phase $`\delta `$ is relevant; here we take $`\delta =0`$ and compare with nonzero $`\delta `$ below. The other choice is for the VO solution, with $`\mathrm{\Delta }m_{sol}^2=10^{10}`$ eV<sup>2</sup> and $`\mathrm{sin}^22\theta _{12}=1`$. The SMA solution gives the same results as the VO solution. One sees that the terms involving $`\mathrm{\Delta }m_{sol}^2`$ can have non-negligible effects on the $`\nu _\mu \nu _e`$ oscillation probability for this path-length, especially at lower energies.
As noted earlier, in the one-mass scale approximation, there are no CP violation effects in these oscillations; however, when we take into account $`\mathrm{\Delta }m_{sol}^2`$, we also have to consider CP-violating effects. We present in Fig.5 a comparison showing the results for the probabilities $`P(\nu _\mu \nu _e)`$ and $`P(\nu _e\nu _\mu )`$ for $`\delta =0`$ and $`\delta =\pi /2`$. We consider $`L=2900`$ km, $`\mathrm{sin}^22\theta _{23}=1`$, $`\mathrm{sin}^22\theta _{13}=0.1`$ and the LMA solution for the solar neutrino problem. We can see that the effects of the CP-violating phase are small. Note however that for non-zero CP-violation, $`P(\nu _\mu \nu _e)P(\nu _e\nu _\mu )`$. For no CP-violation, even with matter effects, there is no difference between these two probabilities. Since with a muon storage ring, by switching between $`\mu ^{}`$ and $`\mu ^+`$ beams, one could obtain both $`\nu _\mu `$ and $`\nu _e`$ beams, there is the possibility of searching for the CP (actually T) violating difference $`P(\nu _\mu \nu _e)P(\nu _e\nu _\mu )`$. In practice, however, it would be difficult to identify the $`e^{}`$ from $`\nu _e`$, given that the $`\mu ^{}`$ stored beam that yields the initial $`\nu _\mu `$ also yields $`\overline{\nu }_e`$, which produce $`e^+`$ in the detector, and given that it would be quite difficult to measure the sign of the $`e^\pm `$ in planned detectors. An alternate method, to measure the asymmetry
$$D=\frac{P(\nu _e\nu _\mu )P(\overline{\nu }_e\overline{\nu }_\mu )}{P(\nu _e\nu _\mu )+P(\overline{\nu }_e\overline{\nu }_\mu )}$$
(49)
is, in principle, possible, although it is complicated by the fact, as noted above, that $`D`$ is rendered nonzero by matter effects even in the absence of CP violation (see also ). If the solar neutrino problem is solved by the SMA or vacuum oscillations, CP-violation effects are not observable in the experiments of interest here. Indeed, even for the LMA solution, the CP violation would be very hard to detect for path lengths larger than $`3000`$ km because of matter effects.
For the Fermilab to Gran Sasso distance $`L7330`$ km (or the BNL to Gran Sasso distance $`L6560`$ km), the $`\mathrm{\Delta }m_{sol}^2`$ corrections are negligible, so we can analyze the problem using fewer relevant parameters: $`\mathrm{\Delta }m_{atm}^2`$, $`\theta _{13}`$ and $`\theta _{23}`$. We calculate the oscillation probabilities in long baseline experiments as a function of $`E/\mathrm{\Delta }m^2`$, rather than using a particular value for $`\mathrm{\Delta }m^2`$ or the energy. The relevant ranges are $`\mathrm{\Delta }m^2\mathrm{few}\times 10^3`$ eV<sup>2</sup> and energies $`E`$ of the order of tens of GeV. This way of presenting the results can be useful in studying the optimization of the beam energy. We calculate the oscillation probabilities for different values of the mixing angles $`\theta _{13}`$ and $`\theta _{23}`$ allowed by the atmospheric neutrino data and the CHOOZ experiment.
We consider both neutrinos and antineutrinos. The matter effects reverse sign in these two cases; for antineutrinos, $`V`$ in (29) is replaced with $`(V)`$. This implies that if $`\mathrm{\Delta }m^2`$ is positive (as considered here), one can get a resonant enhancement of the oscillations for neutrinos, while for antineutrinos the matter effects would suppress the oscillations. The situation would be reversed if $`\mathrm{\Delta }m^2`$ were negative. The fact that the matter effects are opposite in sign for neutrinos and antineutrinos is well illustrated in Fig.2, where both results are presented, together with the vacuum case.
In order to study the effects at different distances, we show the same type of graphs for both $`L=7330`$ km and $`L=2900`$ km. For $`L=2900`$ km, the probabilities can be expressed as functions of $`E/\mathrm{\Delta }m_{atm}^2`$ only for the SMA and VO solutions to the solar neutrino problem. For LMA, small $`\mathrm{\Delta }m_{sol}^2`$ and CP violation corrections are added, as shown in Fig.4 and Fig.5.
We first study the survival probability of $`\nu _\mu `$. If the beam went through vacuum, the probability of oscillation would be given by Fig.6 for a wide range of allowed values of $`\mathrm{sin}^2(2\theta _{13})`$. In matter, this probability becomes sensitive to all oscillation parameters for longer baselines such as 7330 km. In order to illustrate this, we calculate the probability for $`\mathrm{sin}^2(2\theta _{13})=0.1`$, 0.04, and 0.01, and $`\mathrm{sin}^2(2\theta _{23})=0.8`$ and $`\mathrm{sin}^2(2\theta _{23})=1`$. The results are presented in Fig.7 and Fig.8. Evidently, the matter effect increases as $`\theta _{13}`$ increases (and vanishes if $`\theta _{13}=0`$). While the shift in the positions of the maxima and minima, as functions of $`E/\mathrm{\Delta }m^2`$ are small, there is a considerable change in the maximum at $`E/\mathrm{\Delta }m^23\times 10^3`$. This is of great interest, since the use of a typical neutrino energy of $`E10`$ GeV (somewhat less than the stored muon energy) would produce this value of $`E/\mathrm{\Delta }m^2`$, given the central value $`\mathrm{\Delta }m^2=\mathrm{\Delta }m_{atm}^23.5\times 10^3`$ eV<sup>2</sup> reported by SuperKamiokande .
We also want to compare the solution in vacuum (Fig.6), with the solution in matter for neutrinos (Figs. 7,8) and antineutrinos (Fig.9). For antineutrinos the $`\nu _\mu `$ survival probability is not sensitive to the value of $`\theta _{13}`$. One can again see the opposite effects of matter on neutrinos and antineutrinos. The difference in the results for different mixing angles makes it possible, in principle, to use this probability for relatively precise measurements of the oscillation parameters. Measuring separately the probability for $`\nu _\mu `$ and $`\overline{\nu }_\mu `$ can be very useful in detecting the matter effects and using these to constrain the relevant mixings and squared mass difference. Clearly, if one could use two path lengths, as may be possible with a neutrino factory, this would provide more information and constraints.
The relative effects of matter can be especially dramatic in the oscillation probability $`P(\nu _e\nu _\mu )`$, since these directly involve $`\nu _e`$. Since the $`\nu _e`$ beam would arise from a stored $`\mu ^+`$ beam, and the $`\overline{\nu }_\mu `$’s from the decays of the $`\mu ^+`$’s would produce $`\mu ^+`$’s in the detector, the signature for the $`\nu _e\nu _\mu `$ oscillation would be wrong-sign muons. As noted above, planned detectors would be capable of searching for such wrong-sign muons. Since this is a sub-dominant channel, the oscillation effect is small. If the beam went through the vacuum, neither $`P(\nu _e\nu _\mu )`$ nor the charge conjugate, $`P(\overline{\nu }_e\overline{\nu }_\mu )`$, would be enhanced by the matter effect (see Fig. 10, showing $`P(\nu _\mu \nu _e)`$, which is equal to $`P(\nu _e\nu _\mu )`$ for the present situation where CP violation is negligible). Because of the matter effect however, this probability can be strongly enhanced, as is evident in Fig.11 and Fig.12. For $`L=7330`$ km, the enhancement is largest for $`E/\mathrm{\Delta }m^22.5\times 10^3`$ GeV/eV<sup>2</sup>. This is close to the ratio that one would get for a neutrino energy of $`E10`$ GeV, given the indication from the data that $`\mathrm{\Delta }m_{atm}^2=3.5\times 10^3`$ eV<sup>2</sup>. For $`L=2900`$ km, the largest enhancement is obtained for $`E/\mathrm{\Delta }m^2`$ a factor of 3 lower. We also show in this case the results for the baselines corresponding to possible BNL-SLAC and BNL-Gran Sasso distances; see Fig.13. As is evident, the matter effect can amplify $`P(\nu _e\nu _\mu )`$ and enable this transition to be measured with good accuracy, thereby yielding very important information on the oscillation parameters. This probability is quite sensitive to the value of $`\theta _{13}`$, so one should be able to use it for a good determination of this angle. This physics capability motivates careful design studies to optimize the choice of $`L`$ and $`E`$ for this measurement. The sensitivity to $`\mathrm{\Delta }m^2`$ is also quite strong, due to the pronounced peak given by the matter effect in the relevant region. Note that for antineutrinos, the oscillation is suppressed (Fig.14), so an independent measurement of the two channels ($`\nu _\mu \nu _e`$ and $`\overline{\nu }_\mu \overline{\nu }_e`$) would be very valuable.
The atmospheric neutrino data tells us that the dominant oscillation channel is actually $`\nu _\mu \nu _\tau `$. Consequently, it would be very useful to measure $`P(\nu _\mu \nu _\tau )`$; this would provide further confirmation of this oscillation and could also provide further information on $`\mathrm{\Delta }m^2`$ and $`\theta _{23}`$. In addition to the MINOS experiment , the ICANOE and OPERA detectors that will operate in the CERN to Gran Sasso neutrino beam envision $`\tau `$ appearance capabilities . Results for $`P(\nu _\mu \nu _\tau )`$ are presented in Fig.15, and Fig.16 shows $`P(\overline{\nu }_\mu \overline{\nu }_\tau )`$. Next, we present $`P(\nu _e\nu _\tau )`$ in Fig.17 and $`P(\overline{\nu }_e\overline{\nu }_\tau )`$ in Fig.17. These calculations show that matter effects are important and enhance oscillations of the neutrinos and suppress oscillations of antineutrinos in the relevant region of parameters. By combining results from different types of measurements, in different channels of oscillations, the allowed parameter space can be strongly constrained, leading to precise measurements of all mixings and $`\mathrm{\Delta }m^2`$.
For a baseline over 9000 km, as would be the case for an experiment from Fermilab to SuperKamiokande, the main features discussed above remain true. Matter effects are significant for the oscillation of neutrinos for $`\mathrm{\Delta }m^2`$ in the region suggested by the atmospheric data and energies of the order of 10 GeV . Due to the matter effects, oscillations probabilities become very sensitive to $`\theta _{13}`$. Matter effects also improve the sensitivity to $`\mathrm{\Delta }m^2`$. Since matter effects for antineutrinos are opposite to those for neutrinos, independent measurements of neutrino and antineutrino oscillations would give a precise measure of the matter effects and, consequently, of the parameters relevant to the oscillations. Due to the longer path through the Earth with bigger density, the matter effects can become even more dramatic. However, the statistics of the experiment would be limited by the lower neutrino flux at larger distances, and a careful study is necessary in order to choose optimal values of $`L`$ and $`E`$.
## V Summary
To summarize, in planning for very long baseline neutrino oscillation experiments, it is important to take into account matter effects. These effects are significant for the range of neutrino energies $`E`$ of order 10’s of GeV that are planned for these experiments, given the density of the Earth and the value of $`\mathrm{\Delta }m_{atm}^23\times 10^3`$ eV<sup>2</sup> indicated by current atmospheric neutrino data. We have performed a study of these including realistic density profiles in the earth. Matter effects can be useful in amplifying neutrino oscillation signals and helping one to obtain measurements of mixing parameters and the magnitude and sign of $`\mathrm{\Delta }m_{atm}^2`$.
Acknowledgments
We thank Bob Bernstein and Debbie Harris for helpful comments and have benefitted from participation in the ongoing working groups studying the design and physics capabilities of storage rings as neutrino factories . The research of R. S. was supported in part at Stony Brook by the U. S. NSF grant PHY-97-22101 and at Brookhaven by the U.S. DOE contract DE-AC02-98CH10886.Accordingly, the U.S. government retains a non-exclusive royalty-free license to publish or reproduce the published form of this contribution or to allow others to do so for U.S. government purposes.
|
warning/0002/hep-lat0002032.html
|
ar5iv
|
text
|
# Untitled Document
RUHN-99–8
Noncompact chiral $`U(1)`$ gauge theories on the lattice.
Herbert Neuberger
neuberg@physics.rutgers.edu
Department of Physics and Astronomy
Rutgers University
Piscataway, NJ 08855–0849
Abstract
A new, adiabatic phase choice is adopted for the overlap in the case of an infinite volume, noncompact abelian chiral gauge theory. This gauge choice obeys the same symmetries as the Brillouin-Wigner (BW) phase choice, and, in addition, produces a Wess-Zumino functional that is linear in the gauge variables on the lattice. As a result, there are no gauge violations on the trivial orbit in all theories, consistent and covariant anomalies are simply related and Berry’s curvature now appears as a Schwinger term. The adiabatic phase choice can be further improved to produce a perfect phase choice, with a lattice Wess-Zumino functional that is just as simple as the one in continuum. When perturbative anomalies cancel, gauge invariance in the fermionic sector is fully restored. The lattice effective action describing an anomalous abelian gauge theory has an explicit form, close to one analyzed in the past in a perturbative continuum framework.
1. Introduction
The overlap is a general procedure to regulate chiral gauge theories which also naturally fits on the lattice . At the moment this procedure is unique; the so called “Ginsparg-Wilson” approach$`^{f_1}`$ For a recent review see . is essentially identical . The overlap is based on the domain wall approach of Kaplan and on the infinite fermion approach of Frolov and Slavnov . The infinite number of fermions reside along the extra dimension of the $`d+1`$ dimensional space into which the $`d`$ dimensional domain wall is embedded. For this setup to fit into the Frolov Slavnov approach one needs to make the gauge fields purely $`d`$ dimensional, merely Xeroxed into the extra dimension . The domain wall set up of Kaplan inherits an appealing physical picture from the continuum domain wall setup of Callan and Harvey . This physical picture will be exploited later on.
In this paper I focus on abelian chiral gauge theories. As long as one does not think about embedding the abelian group into a nonabelian one, one is free to consider a theory with noncompact gauge group and unquantized charges. That this might be useful on the lattice has been emphasized by ’t Hooft . To avoid any other issues related to charge quantization, this work addresses an infinite lattice, rather than a finite one. It is simply assumed that a reasonable thermodynamic limit exists; I made no attempts to control it rigorously.
The vectorial version of the class of models we consider is analyzed in almost any field theory textbook. Here I shall use a gauged fixed version with finite photon mass (see equation (2.5) for a lattice transcription), and assume familiarity with some basic facts, for example as explained in J. Zinn-Justin’s textbook on Euclidean Field Theory. Mainly, I am relying on the fact that, as a result of the gauge breaking terms being purely quadratic, gauge breaking effects can be controlled throughout the process of renormalization, and gauge invariance continues to play a crucial role in determining (within perturbation theory) the physical content of the continuum theory. Therefore, although we shall be working in a fixed gauge as above, but now in the chiral case, the issue of anomalies and their cancelation still plays a central role.
Since the models strictly speaking do not exist as continuum nontrivial field theories, the appropriate framework is that of effective Lagrangians. The material I need is presented in the abelian section of a paper by J. Preskill . (The abelian section of that paper is joint work of J. Preskill and M. Wise.) An anomalous chiral abelian gauge theory differs from one in which anomalies cancel by the range of energies it is applicable to, and by how small the photon mass can be made in a natural way. When anomalies cancel it is natural to make the photon massless.
A technical simplification I shall employ has been mentioned already in a talk I gave at Lattice’98 . It amounts to replacing throughout the construction the Wilson Dirac Hamiltonian $`H_W(A)`$ by its sign function. The overlap is invariant under any replacement of $`H_W(A)`$ by a monotonic function of $`H_W(A)`$, $`f(H_W(A))`$. In addition to being monotonic $`f`$ must be smooth and vanish at zero. Although the sign function is not smooth at zero, if the spectrum of $`H_W(A)`$ is excluded from a vicinity of zero, one is allowed to take $`f`$ as the sign function. This simplification makes some formulae look simpler, because all energy denominators are now simple integers, on account of the sign function attaining only the values $`\pm 1`$. This simplification is merely technical.
The plan of the paper follows: After setting up notations and conventions (section 2) the main new ingredient of this paper is presented. It consists of a new phase choice for the overlap, an adiabatic phase choice (section 3). The main advantage of the new phase choice is that the Wess-Zumino functional, which measures the residual gauge dependence in the fermion determinant, is linear in the gauge variables (section 4). The linearity is an exact property holding at finite lattice spacing. The previously adopted BW phase choice yielded a Wess-Zumino functional that was nonlinear in the gauge variables. The old and new phase choices are related by a phase redefinition, explained later in the paper (sections 12 and 13). Two important consequence of the linearity of the Wess-Zumino action are highlighted: absence of gauge dependence on the trivial orbit (a property that was true of the BW phase choice, but only in two dimensions ) and a simple relation between the covariant and the consistent anomaly functionals, on the lattice (sections 5 and 6). From a previous paper on the geometrical aspects of the overlap it is known that the difference between the consistent and covariant anomalies is controlled in the overlap by a geometrical object, namely the Berry curvature over the space of gauge fields. This curvature is associated with the fermionic ground state that plays a central role in the overlap. Because with the new phase choice the difference of the anomalies is now simply related to the consistent anomaly, the consistent anomaly can be related to Berry’s curvature too. This is done by improving the adiabatic (section 7) phase choice so as to simplify the lattice Wess-Zumino functional, without loosing the linearity in the gauge transform variable. The new form of the Wess-Zumino functional involves Berry’s curvature under the disguise of the expectation value of a nontrivial commutator of currents, a Schwinger term (section 8). After this first phase redefinition the road is open to take another step of improvement, this time reducing the lattice Wess-Zumino term to a minimal form which is a direct counterpart of the continuum form. The second step is taken in detail in two dimensions (section 9). This part is entirely technical and generalizations to higher dimensions than 2 can be tedious. Some of the tedium has been removed by recent work, as shall be mentioned where appropriate (section 10). After that the lattice effective action for an anomalous theory is briefly discussed in the context of its relation to the Preskill-Wise work (section 11). Towards the end of the paper, the single particle version of the adiabatic phase choice is worked out (section 12); this is used to show that the good symmetry properties of the BW phase choice also hold for the adiabatic phase choice made in this paper (section 13). After some brief comments about the nonabelian case (section 14) the paper ends with a summary (section 15).
2. Notations and Conventions
As explained above, I work at infinite volume. No attempt will be made to establish a thermodynamic limit in any rigorous way. Still, at times it may be necessary to subtract some trivial thermodynamic infinities. At those times I shall use $`V`$ to denote the total number of sites on our $`d`$-dimensional hypercubic lattice. The Dirac type operators will involve infinite matrices that can be viewed as matrices of dimension $`N\times N`$ where $`N=2^{d_2}V`$. Here, $`d_2=\frac{d}{2}`$ and the dimension $`d`$ is always even. Dirac indices will be denoted by $`\alpha ,\beta `$, etc. Sites will be identified by $`x,y`$, etc.
A fundamental set of fermionic creation/annihilation operators are
$$\{\widehat{a}_\alpha ^{}(x),\widehat{a}_\beta (y)\}=\delta _{\alpha \beta }\delta _{xy}$$
$`(2.1)`$
Operators in the associated Fock space will carry hats. Typically, they are bilinear in the $`\widehat{a}`$’s, of the form $`\widehat{X}=\widehat{a}^{}X\widehat{a}`$. Here, suppressed indices are summed over. $`X`$ is the kernel of $`\widehat{X}`$.
I shall make frequent use of forward and backward directional finite difference operators:
$$\begin{array}{cc}\hfill _\mu ^{x+}f(x)=& f(x+\mu )f(x)\hfill \\ \hfill _\mu ^xf(x)=& f(x)f(x\mu )\hfill \end{array}$$
$`(2.2)`$
$`\mu ,\nu ,`$ etc. denote directions on the lattice. The site superscript on $``$ identifies the site index on which the finite difference is taken.
The noncompact vector potential will be denoted by $`A`$ in abbreviated form. $`A_\mu (x)`$ is an unbounded real number associated with the link going from the site $`x`$ into the positive $`\mu `$ direction. A gauge transformation is defined by a set of real numbers $`\alpha (x)`$ associated with sites on the lattice. Under a gauge transformation we have:
$$A_\mu (x)A_\mu (x)+_\mu ^{x+}\alpha (x)A^{(\alpha )}(x)$$
$`2.3)`$
The field strength associated with a plaquette starting at site $`x`$, going one lattice spacing in the positive $`\mu `$ direction followed by another step in the positive $`\nu `$ direction will be denoted by $`F_{\mu \nu }(x)`$.
$$F_{\mu \nu }(x)=_\mu ^{x+}A_\nu (x)_\nu ^{x+}A_\mu (x)$$
$`(2.4)`$
The pure gauge action is gauge fixed, and to avoid unnecessary infrared complications the photon is given a mass:
$$S_{\mathrm{pure}\mathrm{gauge}}=\frac{1}{4e^2}\underset{x}{}F_{\mu \nu }^2(x)+\frac{m_\gamma ^2}{2}\underset{x}{}A_\mu ^2(x)+\frac{1}{2\xi }\underset{x}{}[_\mu ^xA_\mu (x)]^2$$
$`(2.5)`$
This makes the path integral over $`A`$ well defined for each momentum mode. In the above formula I made standard implicit assumptions about contracting $`\mu `$ and/or $`\nu `$ indices. $`e`$ is the (unquantized) coupling constant. The pure gauge theory is Gaussian. When matter is coupled in a gauge invariant way, the gauge dependence of all 1PI correlation functions can be isolated in closed form because the mass and gauge fixing terms are quadratic and $`A`$ is not an angular field variable.
In order to couple the gauge degrees of freedom to a fermion of charge $`q`$ we introduce the unitary link variables $`U_\mu (x)`$, defined by
$$U_\mu (x;q)=e^{iqA_\mu (x)}$$
$`(2.6)`$
Note that the link variables are not fundamental, the fundamental field is $`A`$. If we have a single fermion, its charge $`q`$ can be absorbed into a redefinition of $`e,m_\gamma ,\xi `$. When we have several fermions we shall absorb in this way the charge highest in absolute value, $`|q|`$. Therefore, when we deal with any fermion individually its charge $`q`$ obeys $`|q|1`$. In the following we only write equations for the highest charge fermion, picked to have $`q=1`$; $`U_\mu (x;1)U_\mu (x)`$. To deal with a charge $`q`$ fermion one simply has to replace $`A`$ by $`qA`$ everywhere in the context of that fermion.
Out of the link matrices $`U`$ we construct the directional parallel transporters $`T_\mu `$, which act both on the site index and on the group index of fermions $`\psi `$:
$$T_\mu (A)(\psi )(x)=U_\mu (x)\psi (x+\widehat{\mu })$$
$`(2.7)`$
Euclidean Dirac matrices are denoted by $`\gamma _\mu `$, act only on spinorial indices, and are defined as usual: $`\{\gamma _\mu ,\gamma _\nu \}=2\delta _{\mu \nu }`$. The chirality matrix $`\gamma _{d+1}=i^{d_2}\gamma _1\gamma _2\mathrm{}\gamma _d`$ anticommutes with all $`\gamma _\mu `$, obeys $`\gamma _{d+1}^2=1`$, and is hermitian.
Gauge transformations act by pointwise multiplication by $`e^{i\alpha (x)}`$.
$$(G(\alpha )\psi )(x)=e^{i\alpha (x)}\psi (x)$$
$`(2.8)`$
The $`T_\mu `$ matrices are “gauge covariant”,
$$G(\alpha )T_\mu (A)G^{}(\alpha )=T_\mu (A^{(\alpha )})$$
$`(2.9)`$
The continuum massive Dirac operator has many lattice analogues, all obeying full hypercubic symmetry. Among these lattice matrices, the Wilson Dirac matrix, $`D_W(A)`$ is the sparsest. $`D_W(A)`$ can be written as:
$$D_W(A)=m+\underset{\mu }{}(1V_\mu );V_\mu =\frac{1\gamma _\mu }{2}T_\mu +\frac{1+\gamma _\mu }{2}T_\mu ^{}$$
$`(2.10)`$
One easily checks that $`V_\mu ^{}V_\mu =1`$ which tells us that $`D_W(A)`$ is bounded. The matrix $`H_W(A)=\gamma _{d+1}D_W(A)`$ is easily seen to be hermitian. The parameter $`m`$ is fixed at some value close to $`1`$.
Most of the time the $`A`$ configurations are restricted by requiring
$$|F_{\mu \nu }(x)|\eta $$
$`(2.11)`$
for every plaquette. If this bound is obeyed for the fields felt by the fermion of charge one, it evidently is also obeyed by the fields seen by the charge $`q`$ fermions.
Pick some number $`\eta `$ which obeys
$$0<\eta <\frac{1(1+m)^2}{(1+\frac{\sqrt{2}}{2})d(d1)}$$
$`(2.12)`$
Clearly, one needs $`|1+m|<1`$. Since $`|\mathrm{sin}(\theta )||\theta |`$ for any $`\theta `$, the matrix $`H_W(A)`$ falls in the class analyzed in reference . As a result, the lowest eigenvalue of $`H_W^2(A)`$, $`\lambda _{\mathrm{min}}(A)`$, obeys the following bound:
$$\left[\lambda _{\mathrm{min}}(A)\right]^{\frac{1}{2}}\left[1\eta (1+\frac{\sqrt{2}}{2})d(d1)\right]^{\frac{1}{2}}|1+m|>0$$
$`(2.13)`$
Therefore, on the restricted space of configurations characterized by (2.11) $`H_W(A)`$ never has a zero eigenstate. By a deformation argument this proves that $`H_W(A)`$ has the same number of negative eigenvalues as the free case $`H_W(0)`$. This number, $`N_v`$, is half the total dimension $`N`$. Although both $`N_v`$ and $`N`$ are infinite, for an unrestricted field $`A`$ it would have been conceivable that $`N_v\frac{1}{2}N`$ be some finite integer. This integer is a meaningful quantity by deformation arguments. The restriction on the field strength makes the integer vanish for all restricted $`A`$’s. The space of restricted $`A`$’s is contractible to $`A=0`$ because if $`A`$ is in the restricted set, so is $`tA`$ where $`t`$ is between zero and one.
Over the restricted set of $`A`$’s the sign function of $`H_W(A)`$, $`ϵ(A)`$, is well defined and local$`^{f_2}`$ The Ginsparg-Wilson relation is equivalent to $`ϵ^2(A)=1`$.. Below I define a Hamiltonian acting on the Fock space generated by polynomials in $`\widehat{a}^{}`$ acting on a vacuum annihilated by all $`\widehat{a}`$:
$$\widehat{H}(A)=\widehat{a}^{}ϵ(A)\widehat{a}+N_v,ϵ(A)=\mathrm{sign}(H_W(A))$$
$`(2.14)`$
The role of the additive constant is to assure that the ground state energy is zero. Another important operator in the Fock space is the local charge:
$$\widehat{n}(x)=\frac{1}{2}[a^{}(x)a(x)a(x)a^{}(x)]$$
$`(2.15)`$
An additive constant was chosen so that the vacuum have zero total charge. The total charge operator is
$$\widehat{N}=\underset{x}{}\widehat{n}(x)$$
$`(2.16)`$
The states in the Fock space will be denoted by a Dirac bra-ket notation. The ground state has zero energy and zero total charge:
$$\widehat{H}(A)|v(A)=0,\widehat{N}|v(A)=0$$
$`(2.17)`$
Under a gauge transformation $`ϵ(A)`$ is gauge covariant, inheriting this property from the parallel transporters, $`T_\mu (A)`$. In the Fock space, gauge transformations are represented by
$$\widehat{G}(\alpha )=e^{i_x\alpha (x)\widehat{n}(x)}$$
$`(2.18)`$
The covariance of $`ϵ(A)`$ implies:
$$\widehat{G}(\alpha )\widehat{H}(A)\widehat{G}^{}(\alpha )=\widehat{H}(A^{(\alpha )})$$
$`(2.19)`$
The ground state of $`\widehat{H}(A)`$ is obtained by occupying all negative energy eigenstates of $`ϵ(A)`$. The ground state is nondegenerate. All single particle-hole excitations are degenerate, and have energy equal to $`2`$. All energy eigenvalues are $`A`$-independent.
Replacing $`ϵ(A)`$ above by $`\gamma _{d+1}`$ and denoting the associated Hamiltonian by $`\widehat{H}^{}`$ defines a reference ground state
$$\widehat{H}^{}|v^{}=0$$
$`(2.20)`$
Since $`\gamma _{d+1}`$ is diagonal in site space
$$\widehat{G}(\alpha )|v^{}=|v^{}$$
$`(2.21)`$
if we require $`_x\alpha (x)=0`$. Moreover, for a constant gauge transformation, where $`\alpha (x)`$ is independent of $`x`$, equation (2.17) still applies because exactly half of the total number of states is filled in the reference ground state, just as in $`|v(A)`$. Therefore, equation (2.21) holds for all $`\alpha `$.
The overlap provides expressions for the fermion determinant and for all fermionic correlation functions . Throughout this paper we shall only need an explicit formula for the determinant:
$$v^{}|v(A)$$
$`(2.22)`$
The absolute value of the determinant is gauge invariant. However, the phase of $`|v(A)`$ has still not been defined and whatever definition one chooses it is implausible that the resulting phase of the overlap would turn out gauge invariant. This paper is about how to define the phase so that, on the lattice, before any limits are taken, gauge invariance be violated by an amount not larger than in continuum formulations.
As far as the fermionic correlation functions go, all we need to know about them is that by themselves they are gauge covariant; thus, any gauge breaking effects reside in the fermionic determinant. In short, if a gauge invariant phase choice is found the entire fermionic sector is gauge covariant and the entire quantization procedure of the chiral theory proceeds just as in textbook QED.
3. Adiabatic phase definition
There are many possibilities to make an adiabatic phase choice: one connects the Hamiltonian one is working with to a standard reference Hamiltonian by a slow time evolution and makes a standard choice for the phases of the eigenstates of the standard Hamiltonian.
The earliest suggestion to employ an adiabatic phase choice for the overlap was made in . Any adiabatic phase choice involves an evolution law, so would require some integration. This is not easy to implement numerically and therefore previous work on the overlap almost exclusively employed the so called BW phase choice. This phase choice will be defined later on when its connection to the present phase choice will be worked out. It is easier to work with the BW phase choice numerically. Also, the BW phase choice is amenable to perturbation theory. The adiabatic phase choice though, has nicer properties and is more geometrical. The $`BW`$ phase choice also has a geometrical interpretation, but it seems less useful in the gauge theory context.
An adiabatic phase choice was also suggested in the past by S. Randjbar-Daemi and J. Strathdee . These authors used traditional, exponential in time, adiabatic turn on of the entire interaction piece of the Lagrangian and showed that anomalies are reproduced in perturbation theory with their phase choice. It was left unclear whether this adiabatic phase choice preserved on the lattice as large a set of symmetries as the BW phase choice did. But, their work made it quite evident that an adiabatic phase choice was a possible alternative to the BW phase choice and is one of the main motivations for this paper.
My choice of adiabatic phase is different from that of S. Randjbar-Daemi and J. Strathdee in that I choose linear interpolation of gauge fields: The gauge fields $`A_\mu (x)`$ appear as parameters in the Hamiltonian. They are replaced by time dependent gauge fields $`A_\mu (x,t)=tA_\mu (x)`$. The time variable $`t`$ is taken to vary between zero and one, but the Hamiltonian is assumed multiplied by a large number $`T`$, so that the evolution
$$i|\frac{dv(t;A)}{dt}_T=T\widehat{H}(tA)|v(t;A)_T$$
$`(3.1)`$
is almost adiabatic. At $`t=0`$
$$|v(t=0;A)_T=|v_0$$
$`(3.2)`$
where $`|v_0`$ is the ground state of $`\widehat{H}(0)`$ with a specific phase choice. Multiplying $`|v_0`$ by a pure phase makes all subsequent states in the evolution change by the same phase, which becomes an $`A`$-independent, immaterial constant.
$`\widehat{H}(A)`$ has zero ground state energy at all $`A`$. In the limit of large $`T`$ the adiabatic theorem tells us that for those gauge fields that the ground state of $`\widehat{H}(A)`$ is separated by a gap from the excited state, the following limit exists and is approached with corrections that go as $`\frac{1}{T}`$ :
$$\underset{T\mathrm{}}{lim}|v(t;A)_T=|v(tA),t>0$$
$`(3.3)`$
The states $`|v(A)`$ play an important role in what follows; they are uniquely fixed by two conditions:
$$\begin{array}{cc}\hfill \widehat{H}& (A)|v(A)=0\hfill \\ \hfill & v(tA)|\frac{dv(tA)}{dt}=0\hfill \end{array}$$
$`(3.4)`$
The states $`|v(A)`$ depend smoothly on $`A`$. With their help the overlap
$$v^{}|v(A)$$
$`(3.5)`$
is completely defined. The phase choice is fixed by the second line in equation (3.4).
There is another way to view the adiabatic phase choice: The state $`|v(A)`$ can be viewed as a complex scalar field in a $`CP(𝒩)`$ model with very large $`𝒩`$ and with the gauge fields $`A`$ as base space. Berry’s connection is the $`U(1)`$ gauge field a physicist would naturally associate with a $`CP(𝒩)`$ model. When the $`U(1)`$ in a $`CP(𝒩)`$ model is gauged one writes an action that does not depend on “local” (this means $`A`$-dependent) phase transformations of the states $`|v(A)`$. This $`U(1)`$ gauge field over the space of $`A`$’s changes by a $`U(1)`$ gauge transformation when the state $`|v(A)`$ is multiplied by a phase. Therefore, local phase independence is achieved by writing a Lagrangian in terms of the abelian field strength associated with the gauge field. Since the gauge field is Berry’s connection $`𝒜_{\mu x}(A)`$, the field strength $`_{\mu x,\nu y}(A)`$ is Berry’s curvature.
$$𝒜_{\mu x}(A)=v(A)|\frac{v(A)}{A_\mu (x)}$$
$`(3.6)`$
A phase choice for $`|v(A)`$ amounts to fixing the gauge in this $`U(1)`$ gauge theory over $`A`$-space. The BW phase choice more or less corresponds to a “unitary gauge” because it fixes the phase of the component of $`|v(A)`$ in a given direction in the Hilbert space. The adiabatic phase choice we adopt in this paper corresponds to the Fock-Schwinger gauge choice:
$$\underset{\mu x}{}A_\mu (x)𝒜_{\mu x}(A)=0$$
$`(3.7)`$
In (3.7) I chose to make explicit also the summation over $`\mu `$, just to stress the similarity to the more familiar form of the Fock-Schwinger gauge in ordinary continuum QED, $`_\mu x_\mu A_\mu (x)=0`$. Equation (3.7) is trivially equivalent to the second line in equation (3.4). The Fock-Schwinger gauge is particularly appropriate for abelian gauge theories on contractible spaces, which is the case here.
4. The adiabatic phase choice produces a linear Wess-Zumino action
The goal now is to derive the relation between two states corresponding to backgrounds differing by a gauge transformation $`\alpha `$:
$$A_\mu ^{(\alpha )}(x)=A_\mu (x)+\alpha (x+\mu )\alpha (x)$$
$`(4.1)`$
Recall that the Hamiltonian is gauge covariant and that there is no degeneracy in the ground state. Therefore:
$$|v(A^{(\alpha )})=e^{i\mathrm{\Phi }(\alpha ,A)}G(\alpha )|v(A)$$
$`(4.2)`$
This defines the Wess-Zumino action $`\mathrm{\Phi }(\alpha ,A)`$. To calculate $`\mathrm{\Phi }(\alpha ,A)`$ the adiabatic phase definition of equation (3.4) must be used. Therefore, the “time” parameter $`t`$ has to be reintroduced:
Fix $`\alpha `$ and $`A`$. For all $`t`$,
$$|v((tA)^{(t\alpha )})=e^{i\mathrm{\Phi }(t\alpha ,tA)}G(t\alpha )|v(tA)$$
$`(4.3)`$
with $`\mathrm{\Phi }(t\alpha ,tA)=0`$ at $`t=0`$. The gauge transform acts linearly:
$$(tA)_\mu ^{(t\alpha )}(x)=tA_\mu (x)+t(\alpha (x+\mu )\alpha (x))=tA_\mu ^{(\alpha )}(x)$$
$`(4.4)`$
This relation becomes more complicated in the nonabelian case. That it holds here is the basic reason for the linearity of $`\mathrm{\Phi }(\alpha ,A)`$ in $`\alpha `$.
For brevity, let us temporarily denote $`\mathrm{\Phi }(t\alpha ,tA)=\phi (t)`$. Now, take a time derivative of equation (4.3), and after that multiply the resulting equation from the left by $`v(tA^{(\alpha )})|=v((tA)^{(t\alpha )})|`$. On the left there is one term and on the right there are three terms, one for each time dependent factor on the right hand side of equation (4.3).
$$v(tA^{(\alpha )})|\frac{dv(tA^{(\alpha )})}{dt}=i\frac{d\phi }{dt}+i\underset{x}{}\alpha (x)v(tA)|\widehat{n}(x)|v(tA)+v(tA)|\frac{dv(tA)}{dt}$$
$`(4.5)`$
The adiabatic condition makes the terms containing time derivatives acting on states vanish, leading to
$$\frac{d\phi }{dt}=\underset{x}{}\alpha (x)v(tA)|\widehat{n}(x)|v(tA)$$
$`(4.6)`$
The main result is quite simple:
$$\mathrm{\Phi }(\alpha ,A)=\underset{x}{}\alpha (x)_0^1𝑑tv(tA)|\widehat{n}(x)|v(tA)$$
$`(4.7)`$
The important features of this formula are that it is linear in the gauge degrees of freedom $`\alpha (x)`$ and that the factor multiplying the $`\alpha (x)`$ is a gauge invariant function of $`A`$. Moreover, since $`_x\widehat{n}(x)=\widehat{N}`$ is the total fermion number operator defined so that
$$\widehat{N}|v(A)=0,$$
$`(4.8)`$
we have the identity
$$\underset{x}{}v(A)|\widehat{n}(x)|v(A)=0$$
$`(4.9)`$
for all $`A`$. Although (4.9) has an infinite sum over sites, there is nothing “formal” about it because the matrix element vanishes at large $`x`$. This is so because the action forces $`A_\mu (x)`$ to go to zero as $`x`$ goes to infinity and the state $`|v(A)`$ comes from a massive field theory, ensuring that the matrix element is a local functional of $`A`$.
Since the overlap is the regulated chiral fermion determinant, the functional $`\mathrm{\Phi }(\alpha ,A)`$ indeed deserves to be viewed as the regulated abelian Wess-Zumino action describing the gauge dependence of the lattice fermion determinant.
$$v^{}|v(A^{(\alpha )})=e^{i\mathrm{\Phi }(\alpha ,A)}v^{}|v(A)$$
$`(4.10)`$
Here we used that $`\widehat{n}(x)|v^{}=0`$ for all $`x`$.
The derivation of the main result and its main properties was simple because the underlying physics is simple: Think in terms of the Callan Harvey setup, but adopt the overlap viewpoint of the coordinate perpendicular to the domain wall as a time direction. $`\widehat{n}(x)`$ becomes then the local charge operator and the total charge $`\widehat{N}`$ is conserved. The state $`|v(A)`$ is built up slowly starting from $`|v_0`$ and gradually increasing $`A`$ to its final value. The local charge density, $`v(A)|\widehat{n}(x)|v(A)`$ starts off at $`A=0`$, where it vanishes. The total charge $`_xv(A)|\widehat{n}(x)|v(A)`$ is conserved, so that the single way local charges $`v(A)|\widehat{n}(x)|v(A)`$ build up is by flow of charge carrying currents moving local charge from one place to another. The flow of these currents during the slow buildup causes local charge to be redistributed, making the expectation value of $`\widehat{n}(x)`$ nontrivially dependent on time and its final value on the final field $`A`$. The history of the slow buildup of local charge tells us what phases will be acquired when we change the background by a gauge transformation because the local charge is also the generator of local phase transformations. This picture not only explains the formula for $`\mathrm{\Phi }(\alpha ,A)`$, but also indicates that the charge buildup is described by local currents. This will play a role in what follows.
Before proceeding let me remark that the adiabatic phase choice makes the action induced by integrating out the fermions depend on the entire, single site, noncompact vector potential $`A_\mu (x)`$, and not just on the vector potential modulo some integer valued field, as would be the case if all dependence went through $`e^{iA_\mu (x)}`$. In this sense, the effective action induced by integrating out the fermions is similar to the pure gauge action. However, it is only the phase choice that depends also on the $`2\pi \times `$ integer part of the vector potential. The real part of the fermionic contribution to the action depends only on the $`e^{iA_\mu (x)}`$’s.
It is worthwhile to stress that also the variables $`\alpha (x)`$ do not appear as angular variables in the above Wess-Zumino action. In other words, $`\mathrm{\Phi }(\alpha ,A)`$ does not change by an integral multiple of $`2\pi `$ when we shift $`\alpha (x)\alpha (x)+2\pi z(x)`$ with $`z(x)Z`$. This is intrinsically related to the linearity of $`\mathrm{\Phi }(\alpha ,A)`$ in $`\alpha `$. For example, the $`BW`$ phase convention keeps both $`A_\mu (x)`$ and $`\alpha (x)`$ at the status of angular variables. This is a source of difficulties, see for example . While “unrolling” $`A_\mu (x)`$ may seem a mere technicality, doing the same for the gauge transformation parameter, on the lattice, is a bit more surprising:
A very well known view of anomalies, due to Fujikawa , is that the Wess-Zumino action is the Jacobian associated with the change in fermion integration measure under the gauge transformation $`\psi _R(x)e^{i\alpha (x)}\psi (x)`$. In particular on a lattice this really means, for example, that a transformation with $`\alpha (x)=2\pi z(x)`$ does not do a thing. But, with our adiabatic phase choice, we do get a nontrivial Wess-Zumino action for such a gauge transformation! Hence, a literal realization of Fujikawa’s interpretation is ruled out in the overlap defined with the adiabatic phase choice. Still, much of the essence of Fujikawa’s viewpoint is preserved: The lack of gauge invariance is entirely contained in the Wess-Zumino term and there are no other sources of gauge breaking; all this is just as it would be if the anomaly truly could be viewed as a fermion integration measure effect. But, on the lattice, it seems that insisting on a “measure” terminology is inappropriate. In short, the Wess-Zumino action is the single source of gauge noninvariance because the overlap preserves the slightly amended continuum Fujikawa viewpoint, namely that all gauge violation can be viewed as if coming from a gauge field dependent fermionic integration “measure”. Thus, if the Wess-Zumino action vanishes, full lattice gauge invariance gets restored. Taking the word “measure” to really mean measure however, is useless and confusing.
In the continuum the imaginary part of the logarithm of the chiral determinant is parity odd and purely imaginary. It is also gauge dependent and gives the anomaly in the so called consistent form. In the abelian case the gauge dependence is exactly linear in the gauge transform variables $`\alpha `$. The coefficient of $`\alpha `$ is gauge invariant and homogeneous in the field strength where the degree of homogeneity is defined by the number of field strength factors required to saturate the $`d`$-dimensional antisymmetric epsilon symbol.
With the overlap and the adiabatic phase choice we preserve most of the properties listed in the previous paragraph (symmetries ensuring parity oddness will be dealt with later, in section 13) except that the coefficient of $`\alpha `$ is not that simple. Had we used the BW phase choice instead, we would have also lost the linearity in $`\alpha `$.
Let us now turn to establish some direct consequences of the linearity of the lattice Wess-Zumino action with the adiabatic phase choice.
5. Gauge invariance on the trivial gauge orbit
The trivial orbit is given by gauge transforms of $`A=0`$. The main result of the previous section implies that the dependence on $`\alpha `$ is now
$$\mathrm{\Phi }(\alpha ,0)=\underset{x}{}\alpha (x)v(0)|\widehat{n}(x)|v(0)$$
$`(5.1)`$
Because of translational invariance the matrix element $`v(0)|\widehat{n}(x)|v(0)`$ must be a constant and this constant is zero since the total charge of all $`|(A)`$ is zero. Thus, there is no gauge dependence on the trivial orbit. This property was shown to be true in 2 dimensions with the BW phase choice . With the adiabatic phase choice we now see that it holds in any dimension.
Note that absence of dependence on the gauge degrees of freedom on the trivial orbit holds independently of whether the theory is anomalous or not. This is what one would expect, based on the continuum, at infinite volume. However, in the past, people often focused on models reduced to the trivial orbit and the failure to satisfactorily eliminate the gauge dependence there (for example, recall the Yukawa/gauge fixed approach ) was taken as an indication of the difficulties associated with lattice chiral fermions. It is now apparent that this line of thought was in error, as probably many long suspected when observing that difficulties were appearing even before a single fermion loop was taken into account.
6. Simple relation between consistent and covariant anomalies
Since we are dealing with the abelian case the terminology in the title is misleading: both the consistent and the covariant anomaly are gauge invariant in the continuum. This is also true on the lattice with the overlap in the adiabatic phase choice. Nevertheless, there are two anomalies: they differ by a prefactor that can be understood as coming from imposing full Bose symmetry on the external legs of the anomaly diagram in the consistent case.
The lattice consistent anomaly is trivially read off the function $`\mathrm{\Phi }`$:
$$\mathrm{}_{\mathrm{consistent}}(x;A)=_0^1𝑑tv(tA)|\widehat{n}(x)|v(tA)$$
$`(6.1)`$
The covariant anomaly differs from the consistent anomaly by the divergence of the Berry connection viewed as a current. To understand this statement I need to review the definitions of the consistent and covariant currents in the overlap context:
The nonlocal consistent current, given by definition by
$$𝒥_{\mu \mathrm{consistent}}(x;A)=\frac{\mathrm{log}v^{}|v(A)}{A_\mu (x)}$$
$`(6.2)`$
is naturally decomposed into a gauge invariant part and a local part:
$$𝒥_{\mu \mathrm{consistent}}(x;A)=\frac{1}{v^{}|v(A)}v^{}|\frac{v(A)}{A_\mu (x)}_{}+v(A)|\frac{v(A)}{A_\mu (x)}$$
$`(6.3)`$
The first term is the gauge invariant (covariant in the nonabelian case) $`𝒥_{\mu \mathrm{covariant}}(x;A)`$ current and the second term is Berry’s connection $`𝒜_{\mu x}(A)`$. By definition $`|\frac{v(A)}{A_\mu (x)}_{}=\left[1|v(A)v(A)|\right]|\frac{v(A)}{A_\mu (x)}`$. A more explicit formula for $`𝒥_{\mu \mathrm{covariant}}(x;A)`$ will be derived in section 8. Berry’s connection is defined only in terms of the state $`|v(A)`$ which is the ground state of a massive system; this makes $`𝒜_{\mu x}(A)`$ a local functional of $`A`$. The consistent and covariant currents depend on both states $`|v^{}`$ and $`|v(A)`$ and are nonlocal functionals of the $`A`$; it is only the current difference that is independent of $`|v^{}`$ and therefore local.
The consistent anomaly $`\mathrm{}_{\mathrm{consistent}}(x;A)`$ quoted above is the divergence of the consistent current
$$\mathrm{}_{\mathrm{consistent}}(x;A)=_\mu ^x𝒥_{\mu \mathrm{consistent}}(x;A),$$
$`(6.4)`$
and the covariant anomaly $`\mathrm{}_{\mathrm{covariant}}(x;A)`$ quoted above is the divergence of the covariant current
$$\mathrm{}_{\mathrm{covariant}}(x;A)=_\mu ^x𝒥_{\mu \mathrm{covariant}}(x;A)$$
$`(6.5)`$
Therefore, by definition,
$$\mathrm{}_{\mathrm{consistent}}(x;A)\mathrm{}_{\mathrm{covariant}}(x;A)=_\mu ^xv(A)|\frac{v(A)}{A_\mu (x)}$$
$`(6.6)`$
We can explicitly evaluate the right hand side by making $`\alpha `$ infinitesimal in the gauge transformation rule of the states $`|v(A)`$:
$$\begin{array}{cc}\hfill v(A)|v(A^{(\alpha )})=1+& e^{i\mathrm{\Phi }(\alpha ,A)}v(A)|G(\alpha )|v(A)=\hfill \\ & i\underset{x}{}\alpha (x)_0^1t𝑑t\frac{dv(tA)|\widehat{n}(x)|v(tA)}{dt}+O(\alpha ^2)\hfill \end{array}$$
$`(6.7)`$
This implies:
$$\mathrm{}_{\mathrm{covariant}}(x;A)=v(A)|\widehat{n}(x)|v(A)$$
$`(6.8)`$
One can directly verify quite easily that indeed the above is the divergence of the covariant current as defined below equation (6.3) .
That the relationship between the two anomalies comes out right in the continuum limit is obvious if one accepts that the continuum limit of the covariant anomaly is homogeneous in $`A`$. Then one has:
$$\mathrm{}_{\mathrm{consistent}}(x;A)=c\mathrm{}_{\mathrm{covariant}}(x;A)$$
$`(6.9)`$
The degree of homogeneity is $`d_2`$ because one needs to saturate the antisymmetric $`d`$ dimensional epsilon symbol by field strength factors. The integral over $`t`$ in the consistent anomaly is trivial. We obtain a result better known from arguments based on the symmetry of a one fermion loop diagram:
$$c=\frac{1}{1+d_2}\frac{1}{1+\frac{d}{2}}$$
$`(6.10)`$
Establishing this relation, for example, with the BW phase choice is more difficult.
The main result here is that, even before the continuum limit is taken, at finite lattice spacing, the difference between the two anomalies is quite similar in structure to either anomaly:
$$\mathrm{}_{\mathrm{consistent}}(x;A)\mathrm{}_{\mathrm{covariant}}(x;A)=_0^1t𝑑t\frac{dv(tA)|\widehat{n}(x)|v(tA)}{dt}$$
$`(6.11)`$
From (6.11) we learn that in order to set the difference between the consistent and covariant anomaly to zero, identically for all $`A`$, we need $`v(A)|\widehat{n}(x)|v(A)0`$. But, if this is true all anomalies vanish. In short, making the difference between the two anomalies vanish makes them vanish individually.
On the other hand, the difference between the consistent and covariant anomalies is governed by Berry’s curvature, as explained in . Berry’s curvature is a gauge invariant object at finite lattice spacing. It is a rank two antisymmetric tensor over the space of gauge fields $`A`$. We conclude that it must be possible to improve the adiabatic phase choice so as to make it explicit that anomalies are non-vanishing if and only if Berry’s curvature is nonzero. This is the objective of the next section.
7. Improving the adiabatic phase choice
To simplify further the Wess-Zumino lattice action we need to change the adiabatic phase choice. After this simplification it will become evident that a further phase redefinition exists for which the Wess-Zumino term vanishes altogether if (and only if) perturbative anomalies cancel.
The adiabatic phase choice obeys the symmetries first established for the BW phase choice in (the proof will be presented in section 13). Since the BW phase choice and the adiabatic phase choice both obey the symmetries of , so does their difference, so one can view our final phase choice as coming from the BW phase choice directly. But, it is easier to do get there starting from the adiabatic phase choice.
To redefine the phase so that the Wess-Zumino action $`\mathrm{\Phi }(\alpha ,A)`$ simplifies as much as possible we need a better understanding of the quantity $`v(A)|\widehat{n}(x)|v(A)`$. As discussed before, the basic physics observation is that $`v(A)|\widehat{n}(x)|v(A)`$ is built up adiabatically from $`0`$ by the flow of currents. Let us look for the current by calculating the time evolution of the local charge:
$$\begin{array}{cc}\hfill \frac{dv(tA)|\widehat{n}(x)|v(tA)}{dt}=& 2\mathrm{}[v(A)|\widehat{n}(x)|\frac{dv(tA)}{dt}]=\hfill \\ \hfill 2\mathrm{}& \left[\underset{n}{}v(tA)|\widehat{n}(x)\widehat{H}(tA)|v_n(tA)\frac{1}{E_n(tA)}v_n(tA)|\frac{dv(tA)}{dt}\right]\hfill \end{array}$$
$`(7.1)`$
The sum over $`n`$ extends over all single particle-hole excited states above the ground state. The energies $`E_n(A)`$ are all positive and equal to 2. Since the ground state energy has been chosen to be zero we can replace $`\widehat{n}(x)\widehat{H}(tA)`$ by the commutator $`[\widehat{n}(x),\widehat{H}(tA)]`$. This is a commutator between bilinears in creation-annihilation operators, so it is completely determined by the commutators of the matrix kernels. This is a very familiar exercise and we immediately conclude that
$$[\widehat{n}(x),\widehat{H}(A)]=i_\mu ^x\widehat{J}_\mu (x;A)$$
$`(7.2)`$
where the kernel of the current operator $`\widehat{J}_\mu (x;A)=\widehat{J}_\mu ^{}(x;A)`$ is local in $`A_\mu (x)`$. So, we obtain
$$v(A)|\widehat{n}(x)|v(A)=i_\mu ^x_0^1dt\frac{1}{2}[v(tA)|\widehat{J}_\mu (x;tA)|\frac{dv(tA)}{dt}c.c.]_\mu ^xj_\mu (x;A)$$
$`(7.3)`$
Although the left hand side is gauge invariant, the “current” $`j_\mu (x;A)`$ is not, as we shall see below. Still, under a gauge transformation, $`j_\mu (x;A)`$ transforms linearly. $`j_\mu (x;A)`$ is explicitly constructed in the massive theory and therefore is a local gauge functional of the gauge fields. This means that its value at $`x`$ depends on values of $`A_\mu (y)`$ only exponentially weakly as $`|yx|\mathrm{}`$.
The current is used to redefine the phases of the adiabatic states by
$$|v(A)|v^{\mathrm{new}}(A)=\mathrm{exp}\left[i_0^1𝑑tA_\mu (x)j_\mu (x;tA)\right]|v(A)$$
$`(7.4)`$
By design, the contribution from taking a gauge variation of the factor $`A_\mu (x)`$ multiplying the current in the first term on the right hand side of equation (7.4) exactly cancels the adiabatic $`e^{i\mathrm{\Phi }(\alpha ,A)}`$ pre-factor one gets from gauge transforming the state.
$$\begin{array}{cc}& |v^{\mathrm{new}}(A^{(\alpha )})=\hfill \\ & e^{i_x\alpha (x)_0^1𝑑t_\mu ^xj_\mu (x;tA)}e^{i\mathrm{\Phi }(\alpha ,A)}e^{i_xA_\mu (x)_0^1𝑑t[j_\mu (x;tA^{(\alpha )})j_\mu (x;tA)]}\widehat{G}(\alpha )|v^{\mathrm{new}}(A)\hfill \\ & =e^{i_xA_\mu (x)_0^1𝑑t[j_\mu (x;tA^{(\alpha )})j_\mu (x;tA)]}\widehat{G}(\alpha )|v^{\mathrm{new}}(A)\hfill \end{array}$$
$`(7.5)`$
Taking into account $`\widehat{G}(\alpha )|v^{}=|v^{}`$ for all $`\alpha `$ we obtain the new lattice Wess-Zumino action:
$$v^{}|v^{\mathrm{new}}(A^{(\alpha )})=e^{i_xA_\mu (x)_0^1𝑑t[j_\mu (x;tA^{(\alpha )})j_\mu (x;tA)]}v^{}|v^{\mathrm{new}}(A)$$
$`(7.6)`$
What is happening reflects the well known ambiguity of the Wess-Zumino orbit action: namely it can be altered by gauge transforms of arbitrary local functionals of the gauge field. The new lattice form of the Wess-Zumino action is determined by the gauge transformation properties of the current $`j_\mu (x;A)`$.
Recall the formula for the current $`j_\mu (x;A)`$:
$$j_\mu (x;A)=i_0^1dt\frac{1}{2}[v(tA)|\widehat{J}_\mu (x;tA)|\frac{dv(tA)}{dt}c.c.]$$
$`(7.7)`$
The current is not gauge invariant because of the derivative acting on the state. When we insert the known gauge dependence of the states, and use the gauge covariance of the operator $`\widehat{J}_\mu (x;tA)`$ we get two contributions, one from $`\widehat{G}(t\alpha )`$ and the other from $`e^{i\mathrm{\Phi }(t\alpha ,tA)}`$:
$$\begin{array}{cc}& j_\mu (x;A^{(\alpha )})j_\mu (x;A)=\frac{1}{2}\underset{y}{}\alpha (y)_0^1dt[v(tA)|\widehat{J}(x;tA)\widehat{n}(y)|v(tA)+c.c.]+\hfill \\ & \frac{1}{2}_0^1dt\underset{y}{}\alpha (y)v(tA)|\widehat{n}(y)|v(tA)[v(tA)|\widehat{J}_\mu (x;tA))|v(tA)+c.c]\hfill \end{array}$$
$`(7.8)`$
In the first term above, $`\widehat{n}(y)`$ will either produce the ground state back, whose energy is zero, or create a single particle-hole excitation of energy 2. The term containing the ground state will cancel against the second term above, so we are left only with single pair excited states. Therefore we can insert $`\widehat{H}(tA)`$ into the matrix element of the first term, divide by 2 outside and remove the second term. Further, the product $`\widehat{H}(tA)\widehat{n}(y)`$ can be replaced by a commutator since the ground state energy is zero. Using the definition of the operator $`\widehat{J}_\mu (x;A)`$, we end up with
$$\begin{array}{cc}\hfill j_\mu (x;A^{(\alpha )})& j_\mu (x;A)=\frac{i}{4}\underset{y}{}[_0^1dtv(tA)|[\widehat{J}_\mu (x;tA),\widehat{J}_\nu (y;tA)]|v(tA)c.c.]_\nu ^{y+}\alpha (y)\hfill \\ \hfill =& \frac{i}{2}\underset{y}{}\left[_0^1𝑑tv(tA)|[\widehat{J}_\mu (x;tA),\widehat{J}_\nu (y;tA)]|v(tA)\right]_\nu ^{y+}\alpha (y)\hfill \end{array}$$
$`(7.9)`$
We are thus led to introduce the “Schwinger term”, $`S_{\mu \nu }(x,y;A)`$:
$$S_{\mu \nu }(x,y;A)=v(A)|[\widehat{J}_\mu (x;A),\widehat{J}_\nu (y;A)]|v(A)$$
$`(7.10)`$
All the information about the gauge dependence of the current is contained in the Schwinger term. The Schwinger term is closely related to Berry’s curvature introduced in the overlap context in . Thus our objective from the previous section has been realized. The precise relation between the Schwinger term and Berry’s curvature is interesting in its own right and shall be worked out in the next section.
We thus learn that the new Wess-Zumino action is:
$$i\mathrm{\Phi }^{\mathrm{new}}(\alpha ,A)=\frac{1}{2}\underset{x,y}{}A_\mu (x)[_0^1𝑑t_\nu ^yS_{\mu \nu }(x,y;tA)]\alpha (y)$$
$`(7.11)`$
Note that $`S_{\mu \nu }(x,y;A)`$ is purely imaginary and antisymmetric under simultaneous switch of $`\mu `$ with $`\nu `$ and $`x`$ with $`y`$. $`S_{\mu \nu }(x,y;A)`$ is bilocal, gauge invariant and obeys the rather restrictive identity:
$$_\mu ^x_\nu ^yS_{\mu \nu }(x,y;A)0$$
$`(7.12)`$
This identity is easy to prove:
$$\begin{array}{cc}\hfill _\mu ^x_\nu ^yS_{\mu \nu }(x,y;A)& v(A)|\left[\widehat{n}(x)\widehat{H}^2(A)\widehat{n}(y)\widehat{n}(y)\widehat{H}^2(A)\widehat{n}(x)\right]|v(A)\hfill \\ & v(A)|[\widehat{n}(x),\widehat{n}(y)]|v(A)\hfill \end{array}$$
$`(7.13)`$
The abelian structure of the group is the essential. By “bilocal” I mean that $`S_{\mu \nu }(x,y;A)`$ approaches zero as $`|xy|\mathrm{}`$, exponentially, with a decay that is bounded away from zero uniformly in $`A`$ and $`x`$. Moreover, the dependence on $`A_\mu (z)`$ decreases exponentially with $`z`$ as both $`|xz|`$ and $`|yz|`$ go to infinity.
The identity in equation (7.12) establishes that the coefficient of $`\alpha `$ in equation (7.11) indeed is gauge invariant. To make this explicit we would like to replace the $`A`$ factor by a field strength factor $`F`$. This is indeed possible as we shall see in equation (9.4).
8. Schwinger term, Berry’s curvature and covariant current
Berry’s curvature $``$ is defined as the curl over the space of $`A`$’s of Berry’s connection $`𝒜`$:
$$𝒜_{\mu x}(A)=v(A)|\frac{v(A)}{A_\mu (x)}$$
$`(8.1)`$
$$_{\mu x,\nu y}(A)=\frac{𝒜_{\mu x}(A)}{A_\nu (y)}\frac{𝒜_{\nu y}(A)}{A_\mu (x)}=\frac{v(A)}{A_\mu (x)}|\frac{v(A)}{A_\nu (y)}\frac{v(A)}{A_\nu (y)}|\frac{v(A)}{A_\mu (x)}$$
$`(8.2)`$
A connection is sought between $`_{\mu x,\nu y}(A)`$ and $`S_{\mu \nu }(x,y;A)`$. For this we need a formula for the current operators $`\widehat{J}_\mu (x;A)`$ which were defined by requiring locality and
$$[\widehat{n}(x),\widehat{H}(A)]=i_\mu ^x\widehat{J}_\mu (x;A)$$
$`(8.3)`$
Expanding
$$e^{i_x\alpha (x)\widehat{n}(x)}\widehat{H}(A)e^{i_x\alpha (x)\widehat{n}(x)}=\widehat{H}(A^{(\alpha )})$$
$`(8.4)`$
to linear order in $`\alpha `$ we immediately learn that a possible choice for the current is:
$$\widehat{J}_\mu (x;A)=\frac{\widehat{H}(A)}{A_\mu (x)}$$
$`(8.5)`$
Since $`\widehat{H}(A)|v(A)=0`$ we have:
$$\frac{\widehat{H}(A)}{A_\mu (x)}|v(A)=\widehat{J}_\mu (x;A)|v(A)=\widehat{H}(A)|\frac{v(A)}{A_\mu (x)}$$
$`(8.6)`$
Hence,
$$\begin{array}{cc}\hfill S_{\mu \nu }(x,y;A)=& v(A)|[\widehat{J}_\mu (x;A),\widehat{J}_\nu (y;A)]|v(A)=\hfill \\ & \frac{v(A)}{A_\mu (x)}|\widehat{H}^2(A)|\frac{v(A)}{A_\nu (y)}\frac{v(A)}{A_\nu (y)}|\widehat{H}^2(A)|\frac{v(A)}{A_\mu (x)}\hfill \end{array}$$
$`(8.7)`$
The states $`|\frac{v(A)}{A_\mu (x)}`$ are a linear combination of the ground state and single particle-hole excitations of energy 2. For the excited states we can replace $`\widehat{H}^2(A)`$ by 4. For the ground state we get no contribution. But, the ground state does not contribute as an intermediate state to the inner products defining $``$ either because of antisymmetry and the purely imaginary character of $`𝒜`$. Hence,
$$S_{\mu \nu }(x,y;A)=4_{\mu x,\nu y}(A)$$
$`(8.8)`$
Thus, up to a trivial proportionality factor the Schwinger term is the same as Berry’s curvature.
Let us now derive a more physical formula for the covariant current functional, $`𝒥_{\mu \mathrm{covariant}}(x;A)`$. In the abelian case this gauge covariant current is actually gauge invariant. This was proven in the general case (abelian and nonabelian) in ref . We wish to derive an expression in which the gauge invariance becomes explicit. The state $`|\frac{v(A)}{A_\mu (x)}_{}`$ is the same as the state $`|\frac{v(A)}{A_\mu (x)}`$ only the component in the direction of $`|v(A)`$ is removed. The state $`|\frac{v(A)}{A_\mu (x)}`$ contains one component in the direction of the ground state and all other components are single particle hole excited states of energy $`2`$. Therefore:
$$|\frac{v(A)}{A_\mu (x)}_{}=\frac{1}{2}\widehat{H}(A)|\frac{v(A)}{A_\mu (x)}$$
$`(8.9)`$
Using equation (8.6), we can now write the covariant current as:
$$𝒥_{\mu \mathrm{covariant}}(x;A)=\frac{1}{2}\frac{v^{}|\widehat{J}_\mu (x;A)|v(A)}{v^{}|v(A)}$$
$`(8.10)`$
Gauge invariance is now a consequence of the gauge covariance of the current operator $`\widehat{J}_\mu (x;A)`$ and the cancelation of the Wess-Zumino action between numerator and denominator. This formula says that the covariant current is the expectation value of the charge current at the domain wall boundary. Although $`\widehat{J}_\mu (x;A)`$ is local as an operator, the matrix element between the two states $`v^{}`$ and $`|v(A)`$ induces the expected nonlocality in the covariant current functional reflecting the integration of a massless fermionic degree of freedom.
The formulae would look slightly nicer had we rescaled $`\widehat{H}(A)`$ by 2. The standard definition of the overlap Dirac operator indeed has the extra factor of 2 removed.
9. Further improvement of the phase choice: two dimensions.
At this point, to simplify the analysis, we restrict ourselves to two dimensions. At the conceptual level, nothing is lost by this restriction.
Consider, for fixed $`y`$, the quantity
$$ϵ_{\mu \rho }_\nu ^yS_{\rho \nu }(x,y;A)$$
$`(9.1)`$
This quantity can be viewed as an abelian noncompact vector potential on the hypercubic lattice with varying argument $`x`$, defined so that now the backwards derivatives induce gauge transformations. Equation (7.12) says that the abelian field strength associated with this vector potential vanishes everywhere. Therefore, the vector potential is pure gauge, which means there exists a gauge function $`\chi (x,y;A)`$ such that
$$ϵ_{\mu \rho }_\nu ^yS_{\rho \nu }(x,y;A)=_\mu ^x\chi (x,y;A)$$
$`(9.2)`$
$`\chi `$ can be calculated starting from the site $`x=y`$, where $`\chi `$ is set to zero and getting to any other point by summing terms $`ϵ_{\mu \rho }_\nu ^yS_{\rho \nu }(x,y;A)`$ along a chosen path. The vanishing of the above “abelian field strength” tells us that the result is independent of which particular path we chose. $`\chi `$ is fixed uniquely up to a constant in $`x`$. This constant can still depend on $`y`$ and $`A`$. The free constant can be fixed by requiring $`\chi `$ to vanish at any fixed $`y`$ and $`A`$ when $`x`$ is taken to $`\mathrm{}`$. With this, $`\chi `$ is uniquely determined.
Explicitly, we construct $`\chi `$ as follows: We choose some path of minimal length in the lattice “Manhattan” metric. At large distances from $`y`$ the so constructed $`\chi (x,y;A)`$ will have a vanishing gradient in $`x`$, so will become constant in $`x`$, but possibly $`y`$ and $`A`$ dependent. This constant is approached exponentially as $`x`$ increases. We redefine $`\chi (x,y;A)`$ by subtracting this constant (different for each $`y`$ and $`A`$) from the $`\chi (x,y;A)`$ we have. The above equation is still obeyed but now we see that $`\chi (x,y;A)`$ is bilocal. The entire construction only involved quantities invariant under $`AA^{(\alpha )}`$, so $`\chi (x,y;A)`$ is also gauge invariant. We now rewrite the last equation in the form:
$$_\nu ^yS_{\mu \nu }(x,y;A)=ϵ_{\mu \sigma }_\sigma ^x\chi (x,y;A)$$
$`(9.3)`$
The above construction is unique and therefore must maintain the discrete symmetries of the original lattice Schwinger term. So, we can write the new Wess-Zumino action in the form
$$i\mathrm{\Phi }^{\mathrm{new}}(\alpha ,A)=\frac{1}{2}\underset{x,y}{}F_{12}(x)[_0^1𝑑t\chi (x,y;tA)]\alpha (y)$$
$`(9.4)`$
This realizes the expectation to make it explicit that the coefficient of $`\alpha `$ in the Wess-Zumino action is gauge invariant.
From equation (9.3) we expect $`\chi (x,y;A)`$ to be a total $`y`$-divergence of some other local functional of $`A`$. If this were literally true the finite $`y`$-difference operation could be thrown over to act on the $`\alpha (y)`$ factor in (9.4). But then it would be evident that there exists a functional $`\varphi (A)`$ such that $`\mathrm{\Phi }^{\mathrm{new}}(\alpha ,A)=\varphi (A^{(\alpha )})\varphi (A)`$ and an additional phase redefinition by $`\varphi (A)`$ would restore gauge invariance. So, there must be an obstruction to writing $`\chi (x,y;A)`$ as a total $`y`$-divergence. It is only this obstruction that stands between our present phase choice and full restoration of gauge invariance.
Let us focus therefore on the $`y`$-dependence of $`\chi `$. What can stop $`\chi (x,y;A)`$ from being a total $`y`$-divergence of another local object is that $`_y\chi (x,y;A)0`$. A priori this is possible; there is nothing to prohibit, for example, $`\chi (x,y;A)=\mathrm{Const}.\times \delta _{xy}`$. On the other hand, if the sum over $`y`$ of $`\chi (x,y;A)`$ were identically zero, there would be many ways to write it as a total $`y`$ divergence. A subset of these ways produces a bilocal current; this is the representation we are after.
So, we consider the quantity
$$\underset{y}{}\chi (x,y;A)=b(x;A)$$
$`(9.5)`$
Because of equation (9.2) $`b`$ is $`x`$-independent. On the other hand, $`b(x,A)`$ is also a local functional of $`A`$. Therefore, $`b(x,A)`$ must be just a constant number:
$$b(x;A)=b$$
$`(9.6)`$
Hence, the obstruction has boiled down to the value of one constant. The constant number $`b`$ came from the uniquely defined $`\chi `$, which in turn came from the Schwinger term. Since $`b`$ is independent of $`A`$, we can determine it at $`A=0`$. There we have full translational invariance so that $`S_{\mu \nu }(x,y;0)=S_{\mu \nu }(xy)`$ and similarly for $`\chi `$. We go to Fourier space, and denote the Fourier transforms by tildes. The antisymmetry of the Schwinger term implies that $`\stackrel{~}{S}_{\mu \nu }(0)=ϵ_{\mu \nu }s_0`$. Now, we easily deduce that $`s_0=b`$. Hence:
$$b=\frac{1}{2}ϵ_{\mu \nu }\underset{x}{}S_{\mu \nu }(x,y;0)$$
$`(9.7)`$
Let us now define a bilocal quantity $`\psi (x,y;A)`$ with $`b`$ subtracted:
$$\psi (x,y;A)=\chi (x,y;A)b\delta _{x,y}$$
$`(9.8)`$
Now, by construction, we have
$$\underset{y}{}\psi (x,y;A)0$$
$`(9.9)`$
Also, from the properties of $`\chi `$ we know that the sum over $`y`$, restricted to a large square surrounding $`x`$ fixed in its interior, and far from all edges of the square, converges exponentially to zero as the box expands further. We want to convince ourselves that his implies that there exists a bilocal functional with one index, $`\chi _\rho (x,y;A)`$ such that
$$\psi (x,y;A)=_\rho ^y\chi _\rho (x,y;A)$$
$`(9.10)`$
First, fix $`x`$, and focus on the $`y`$ dependence only. Think about $`\psi (x,y;A)`$ as an abelian noncompact two dimensional field strength which is nonzero in some vicinity of a fixed point ($`x`$). The sum condition on $`y`$ means that there is zero total flux through the system. We simply want to find the vector potential producing this field strength. To find this vector potential we set up a maximal axial gauge tree with origin at $`y=0`$. The axial gauge tree is depicted in figure 1. In this gauge it determines the vector potential from the field strength. The sums one needs to do converge as a result of the locality of $`\psi (x,y;A)`$. The question is whether the resulting vector potential is still bilocal in $`x`$ and $`y`$. For this we need the condition $`_y\psi (x,y;A)0`$. Next, for fixed $`x`$, we average over all possible trees with the above structure to restore symmetries. Up to a factor of $`ϵ_{\sigma \rho }`$ we get our $`\chi _\rho (x,y;A)`$ fields.
| Figure 1 | Axial gauge tree for constructing $`\chi _\rho (x,y;A)`$. |
| --- | --- |
Therefore,
$$\chi (x,y;A)=_\nu ^y\chi _\nu (x,y;A)+b\delta _{xy}$$
$`(9.11)`$
With this, the Wess-Zumino action can be rewritten as:
$$i\mathrm{\Phi }^{\mathrm{new}}(\alpha ,A)=\frac{1}{2}\underset{x,y}{}F_{12}(x)[_0^1𝑑t\chi _\nu (x,y;tA)]_\nu ^{y+}\alpha (y)\frac{b}{2}\underset{x}{}F_{12}(x)\alpha (x)$$
$`(9.12)`$
The first term above can be rewritten as
$$\begin{array}{cc}& \frac{1}{2}\underset{x,y}{}F_{12}(x)[_0^1𝑑t\chi _\nu (x,y;tA)]_\nu ^{y+}\alpha (y)=\hfill \\ & \frac{1}{2}\underset{x,y}{}F_{12}(x)[_0^1𝑑t\chi _\nu (x,y;tA)][A_\nu ^{(\alpha )}(y)A_\nu (y)]\hfill \end{array}$$
$`(9.13)`$
By construction, $`\chi _\nu (x,y;A)`$ is invariant under $`AA^{(\alpha )}`$. We therefore redefine our phase again:
$$|v^{\mathrm{new}}(A)|v^{\mathrm{final}}(A)=e^{\frac{1}{2}_{x,y}F_{12}(x)A_\nu (y)[_0^1𝑑t\chi _\nu (x,y;tA)]}|v^{\mathrm{new}}(A)$$
$`(9.14)`$
The new phase adjustment, just as the previous one, was by a local quantity. The final form of the Wess-Zumino lattice functional has been fine tuned now to its simplest possible structure, exhibiting only the $`b`$ constant, the single irremovable obstruction to full gauge invariance.
$$i\mathrm{\Phi }^{\mathrm{final}}(\alpha ,A)=\frac{b}{2}\underset{x}{}F_{12}(x)\alpha (x)$$
$`(9.15)`$
For each right handed fermion of charge $`e_{Ra}`$ one gets a factor $`+e_{Ra}^2`$ and for each left handed fermion of charge $`e_{Lb}`$ one gets a factor $`e_{Lb}^2`$. One charge factor is associated with $`F`$ and another with $`\alpha `$. The constant $`b`$ can be computed in the free theory. The contributions of all fermions add up and perturbative anomalies cancel if the total $`b`$ vanishes. If perturbative anomalies cancel, the Wess-Zumino action vanishes and gauge invariance is fully restored, as there were no other sources of gauge violation.
On the other hand, if anomalies do not cancel, it is impossible to find an additional phase redefinition, by some local functional $`\varphi (A)`$ such that the gauge dependence be made to disappear by $`\mathrm{\Phi }^{\mathrm{final}}(\alpha ,A)=\varphi (A^{(\alpha )})\varphi (A)`$. One can address this directly on the lattice, but it is easier to just observe that if it were possible, one could take a naive continuum limit and eliminate the anomaly also in the continuum. This is well known to be impossible.
10. Higher dimensions
Essentially the same story goes through in higher (even) dimensions than two. The considerations used in section 9 for constructing the fields $`\chi `$ and $`\chi _\nu `$ were those of ordinary abelian noncompact lattice field theory only accidentally. Going to higher dimensions involves more indices and considerations in auxiliary noncompact lattice abelian gauge theory, generalized to local sub-hypercubes of arbitrary dimension: From sites, links plaquettes one needs to go to three-cubes, four-cubes and so forth. As is well known, this generalization involves antisymmetric tensors of higher rank, the rank being given by the dimension of the sub-hypercube. All variables are non-compact, real numbers. Gauge transformations and the relation between variables and gauge invariant “field strengths” are all linear and involve lattice finite difference operations. The basic field variables are defined on $`p`$ dimensional sub-hypercubes, and have a gauge invariance under variables that live on their boundary, i.e. on $`p1`$ dimensional sub-hypercubes. The gauge invariant variables that would enter an action are defined on $`p+1`$ dimensional sub-hypercubes and are sums with orientation depending signs of the basic variables on the respective boundaries. Ordinary gauge theory has $`p=1`$, where the variables live on links, the gauge transformation functions on sites ($`p=0`$) and the field strength variables on plaquettes ($`p=2`$). The two basic questions one needs to deal with are when is a variable pure gauge and how to recover a variable from a known field strength. An entire hierarchy labeled by $`p`$ is employed. The highest $`p`$ is determined by the dimension $`d`$. The increase in tedium can be substantial, but an attempt to formalize it to all dimensions is described in . The entire auxiliary geometric structure is hidden by developing a finite difference calculus on antisymmetric lattice tensor fields.
Combining the existence of the current $`j_\mu (x;A)`$ established in section 7 with the symmetry properties of the adiabatic phase choice proven in section 13 and with the result of on local abelian lattice BRS cohomology, we learn that most of the gauge dependence of $`j_\mu (x;A)`$ can be eliminated by the following decomposition:
$$\begin{array}{cc}\hfill j_\mu (x& ;A)=k_\mu (x;A)+\hfill \\ & a_0\overline{ϵ_{\mu ,\nu _1,\mathrm{}.\mu _k,\nu _k}A_{\nu _1}(x)F_{\mu _2\nu _2}(x+\widehat{\nu }_1)\mathrm{}F_{\mu _{d_2}\nu _{d_2}}(x+\widehat{\nu }_1+\widehat{\mu }_2+\widehat{\nu }_2+\mathrm{}+\widehat{\mu }_{d_21}+\widehat{\nu }_{d_21})}\hfill \\ \hfill +& j_\mu ^{g.i.}(x;A)\hfill \end{array}$$
$`(10.1)`$
Any $`\mu ,\nu `$ index repeated more than once is summed over in (10.1). The top bar indicates that the quantity needs to be averaged with respect to lattice symmetries to comply with those of the current $`j_\mu (x;A)`$. The current $`j_\mu ^{g.i.}(x;A)`$ is both gauge invariant and local. The current $`k_\mu (x;A)`$ is not gauge invariant but has zero divergence. The $`a_0`$ term is the single term that is simultaneously not gauge invariant and not divergence-less. The entire divergence of the current $`j_\mu (x;A)`$ is contained in the $`a_0`$ term together with the $`j_\mu ^{g.i.}(x;A)`$ term. The constant $`a_0`$ can be determined in perturbation theory, and is proportional to the coefficient of the perturbative anomaly. It again comes from the Schwinger term, but this time one needs to expand to order $`d_21`$ in the field strength.
If we now redefine the phases as above, we see that we are left with a Wess-Zumino functional $`\mathrm{\Phi }(\alpha ,A)`$ of the following form:
$$\begin{array}{cc}& \mathrm{\Phi }(\alpha ,A)=ia_0\underset{x}{}\alpha (x)\hfill \\ & \overline{ϵ_{\mu _1,\nu _1,\mathrm{}.\mu _k,\nu _k}F_{\mu _1\nu _1}(x)F_{\mu _2\nu _2}(x+\widehat{\mu }_1+\widehat{\nu }_1)\mathrm{}F_{\mu _{d_2}\nu _{d_2}}(x+\widehat{\mu }_1+\widehat{\nu }_1+\mathrm{}+\widehat{\mu }_{d_21}+\widehat{\nu }_{d_21})}\hfill \end{array}$$
$`(10.2)`$
When there are several fermions of charges $`q`$ and we choose for each a phase definition that produces this minimal form, we shall restore gauge invariance if the sum of all the $`a_0\times q^{d_2+1}`$ constants vanishes. This is the standard anomaly cancelation condition. When $`d_2`$ is even we can discuss only one handedness and cancelations occur between positive and negative charges. When $`d_2`$ is odd, fermions of both handednesses must participate to get a cancelation.
11. Lattice effective actions for anomalous theories in four dimensions
An anomalous chiral abelian gauge theory with an ultraviolet cutoff $`\mathrm{\Lambda }`$ can be viewed as an effective theory describing continuum physics at momenta low relative to $`\mathrm{\Lambda }`$. Assuming a single Weyl fermion, this effective theory has predictive power only if the physical photon mass, $`m_\gamma ^{\mathrm{ph}}`$, obeys:
$$1>>\frac{m_\gamma ^{\mathrm{ph}}}{\mathrm{\Lambda }}\frac{(e^{\mathrm{ph}})^3}{64\pi ^3}$$
$`(11.1)`$
This bound is derived by perturbative power counting arguments in the continuum without specifying the cutoff .
The work in the present paper presents a way to check (11.1) outside perturbation theory, with a lattice cutoff. We take the overlap with one of the phase definitions presented earlier and including it into the pure gauge action defines the path integral from which the photon mass in lattice units can be extracted. We should find that we cannot make the photon mass smaller than the bound, no matter what we do (within reason) with the bare photon mass parameter, $`m_\gamma `$.
It would be interesting to know if the bound depends on whether we use the simple adiabatic phase choice, or any of the subsequent improvements. The simplest expectation is that the sensitivity of the bound on whether we pick the simplest adiabatic phase choice or any of the improved ones be low: The phase choices differ only by local terms added to the gauge action. For all phase choices the Wess-Zumino functional is linear in the gauge transform field $`\alpha `$ and one can use in perturbation theory Stueckelberg’s trick of making the gauge field $`\alpha `$ dynamical, thus endowing the fermion sector with a new gauge invariance, where $`\alpha `$ transforms by shifts under the new gauge transformation. This trick is employed in the perturbative analysis of Preskill and Wise.
Unfortunately, a practical numerical simulation would have to deal with the complex measure in the path integral. There are no known generic ways to deal with extensive phases, so one would need to invent a specific procedure for this case. Another technical obstacle is the numerical local bound on the abelian field strength, which might force the simulations to impractically large lattices.
If anomalies do cancel, one should find that it is easy to make the photon massless as $`m_\gamma `$ is taken to zero. This seems quite obvious in the case one adopts the perfect phase choice. It would be very interesting to see what happens if one adopts the less perfect adiabatic phase choice we started with. This question, posed here in the noncompact context, is somewhat similar to tests of “gauge averaging” in compact formulations. A more precise formulation of gauge averaging in the noncompact framework seems difficult: Gauge restoration in the infrared by gauge averaging is a mechanism (due to ) that works in a strong coupling expansion. In this strong coupling expansion the compactness of the gauge group plays a central role.
Equation (11.1) can be read in two additional ways: as an upper bound on the ultraviolet cutoff and as an upper bound on the physical coupling constant. The difference between the two effective theories, one with canceled anomalies and the other with uncanceled anomalies can be phrased as follows: Let $`m^{\mathrm{ph}}`$ denote a typical low physical scale where the effective Lagrangian applies. We assume $`\frac{m^{\mathrm{ph}}}{\mathrm{\Lambda }}<<1`$. The physical coupling is bounded from above in either case. The bound depends on the ultraviolet cutoff, just like in the Higgs mass bound case . If anomalies do not cancel we have a bound of the following type:
$$e^{\mathrm{ph}}c_1\left(\frac{m^{\mathrm{ph}}}{\mathrm{\Lambda }}\right)^{\frac{1}{3}}$$
$`(11.2)`$
If anomalies do cancel the limitation is far less severe:
$$e^{\mathrm{ph}}\frac{c_2}{\mathrm{log}\left(\frac{\mathrm{\Lambda }}{m^{\mathrm{ph}}}\right)}$$
$`(11.3)`$
Equations (11.2) and (11.3) probably are the most physical way to express the difference between an anomalous and a non-anomalous abelian chiral gauge theory in four space-time dimensions.
This paper can easily be generalized to include an abelian group which has several $`U(1)`$ factors. The various individual and mixed anomalies would work out just as in the continuum. Let us demand that the“triviality” bound on the couplings be logarithmic and take the four $`U(1)`$ factors one would get from the gauge group of the minimal standard model, gauge them, and introduce the fermion content of one generation. We shall find, as in the continuum, that the ratios between all hypercharges are fixed to their known values by anomaly cancelation conditions.
12. Single particle version of the adiabatic phase choice
Since
$$\widehat{H}(A)=\widehat{a}^{}ϵ(A)\widehat{a}+N_v$$
$`(12.1)`$
the overlap is determined by the matrix $`^{f_3}`$ This matrix is closely related to the overlap Dirac operator $`D_o=\frac{1}{2}(1+\gamma _{d+1}ϵ(A))`$ . $`ϵ(A)`$. The eigenstates of this matrix are $`\stackrel{}{v}_i(A)`$ and $`\stackrel{}{w}_i(A)`$ for eigenvalues $`1`$ respectively. Note that the $`\stackrel{}{v}_i(A)`$ are not necessarily eigenstates of $`H_W(A)`$, but they exactly span the negative energy subspace of $`H_W(A)`$.
The ground state of $`\widehat{H}(A)`$ has all the states $`\stackrel{}{v}_i(A)`$ occupied and the rest empty. Thus the information contained in the ground state $`|v(A)`$ is the same as that contained in the rectangular matrix $`v=(\stackrel{}{v}_1(A),\stackrel{}{v}_2(A),\mathrm{}.,\stackrel{}{v}_{N_v}(A))`$, where $`N_v`$ is the total number of negative energy states. Similarly we define the matrix $`w`$ made out of all positive energy states. Also, for the reference system, we introduce the same quantities, all with a prime superscript.
The overlap corresponding to right handed chiral fermions is
$$v^{}|v(A)=detM_R,M_R=v^{}v(A)$$
$`(12.2)`$
and that to left handed chiral fermions is
$$w^{}|w(A)=detM_L,M_L=w^{}w(A)$$
$`(12.3)`$
The matrix $`v(A)`$ is not fully defined because we can unitarily mix the negative energy states. If we make some arbitrary choice, the corresponding state $`|v(A)`$ will not have an adiabatically defined phase. Note that most of the details of a unitary mix are unimportant, only the collective effect on the phase of the second quantized state $`|v(A)`$ matters.
Let us start from some initial choice of single particle states $`\stackrel{~}{v}(A)`$ and $`\stackrel{~}{w}(A)`$. We are restricting our attention to the set of $`A`$’s for which $`H^2(A)`$ is bounded from below by some small positive number. The lower bound (2.13) on $`H^2(A)`$ means that no state can cross between the negative energy and positive energy groups of states. The states $`v(A)`$ and $`w(A)`$ have adiabatic phases and are given by
$$v(A)=\stackrel{~}{v}(A)𝒪_v^{}(A),w(A)=\stackrel{~}{w}(A)𝒪_w^{}(A)$$
$`(12.4)`$
The overlaps with the two phase choices are related by:
$$v^{}|v(A)=detv^{}v(A)=det\stackrel{~}{v}^{}\stackrel{~}{v}(A)det𝒪_v^{}(A),$$
$`(12.5)`$
where, by convention $`v^{}=\stackrel{~}{v}^{}`$. There always are equal numbers of negative energy and positive energy states. The matrices $`𝒪`$ are defined to be unitary and required to be equal to unit matrices for $`A=0`$. They are completely fixed by the differential equation below:
$$\frac{d𝒪_v(tA)}{dt}=𝒪_v(tA)\stackrel{~}{v}^{}(tA)\frac{d\stackrel{~}{v}(tA)}{dt},\frac{d𝒪_w(tA)}{dt}=𝒪_w(tA)\stackrel{~}{w}^{}(tA)\frac{d\stackrel{~}{w}(tA)}{dt},$$
$`(12.6)`$
In turn, this implies,
$$v^{}(tA)\frac{dv(tA)}{dt}=0,w^{}(tA)\frac{dw(tA)}{dt}=0$$
$`(12.7)`$
which, in particular, makes $`|v(A)`$ have adiabatic phases:
$$v(tA)|\frac{dv(tA)}{dt}=det\left[v^{}(tA)\frac{dv(tA)}{dt}\right]$$
$`(12.8)`$
Our objective is to relate the adiabatic phase choice to the BW one, defined by
$$\stackrel{~}{v}(0)|\stackrel{~}{v}(A)>0,\stackrel{~}{w}(0)|\stackrel{~}{w}(A)>0,|v(0)=\stackrel{~}{v}(0),|w(0)=\stackrel{~}{w}(0)$$
$`(12.9)`$
When found, this relation will allow us to extend to the adiabatic phase choice some symmetry properties proven before for the BW phase choice.
The procedure to find the relation between the two phases relies on the fact that in both cases one uses the $`A=0`$ as a reference point. The reference points for the two phase choices can be made the same. The adiabatic evolution takes one from reference states, unitarily, to the adiabatic states at arbitrary $`A`$. The unitary matrix, $`K(A)`$ , which does this was introduced by Kato and is defined below:
$$K(A)=v(A)\stackrel{~}{v}^{}(0)+w(A)\stackrel{~}{w}^{}(0)=\stackrel{~}{v}(A)𝒪_v^{}(A)\stackrel{~}{v}^{}(0)+\stackrel{~}{w}(A)𝒪_w^{}(A)\stackrel{~}{w}^{}(0)$$
$`(12.10)`$
By definition, $`K(0)=1`$. The rest of $`K`$ is defined by deriving a first order evolution equation. For it one employs the projectors
$$P(A)=v(A)v^{}(A)=\frac{1ϵ(A)}{2},1P(A)=w(A)w^{}(A)=\frac{1+ϵ(A)}{2}$$
$`(12.11)`$
The projectors are defined unambiguously, by the subspaces they project on. Kato observed that
$$\frac{dK(tA)}{dt}=[\frac{dP(tA)}{dt},P(tA)]K(tA)$$
$`(12.12)`$
This formula is easy to prove.
By equation (12.4) the relation between the BW phase choice and the adiabatic phase choice is contained in the matrix $`𝒪_v^{}(A)`$. But there is no independent definition of the matrix $`𝒪_v^{}(A)`$. To trade the matrix $`𝒪_v^{}(A)`$ for the matrix $`K(A)`$ which is independently defined we write:
$$\stackrel{~}{v}^{}(0)\stackrel{~}{v}(A)𝒪_v^{}(A)=\stackrel{~}{v}^{}(0)K(A)\stackrel{~}{v}(0)$$
$`(12.13)`$
Taking determinants we get:
$$\begin{array}{cc}\hfill det& 𝒪_v^{}(A)det[\stackrel{~}{v}^{}(0)\stackrel{~}{v}(A)]=det\left[1\stackrel{~}{v}^{}(0)[1K(A)]\stackrel{~}{v}(0)\right]=\hfill \\ & \mathrm{exp}\left\{\underset{m=1}{\overset{\mathrm{}}{}}\frac{1}{m}Tr\left[\stackrel{~}{v}^{}(0)[1K(A)]\stackrel{~}{v}(0)\right]^m\right\}=\mathrm{exp}\left\{\underset{m=1}{\overset{\mathrm{}}{}}\frac{1}{m}Tr\left[[1K(A)]P(0)\right]^m\right\}\hfill \end{array}$$
$`(12.14)`$
So, we learn that:
$$det𝒪_v^{}(A)\stackrel{~}{v}(0)|\stackrel{~}{v}(A)=det\left[1P(0)+K(A)P(0)\right]=det\left[1P(0)+P(0)K(A)\right]$$
$`(12.15)`$
Since, by definition of the BW phase choice, the factor $`\stackrel{~}{v}(0)|\stackrel{~}{v}(A)`$ is positive, the determinants after the first and second equality have the same phase as $`det𝒪_v^{}(A)`$. Thus we arrive at a formula relating the overlap with BW phase choice to the overlap with an adiabatic phase choice:
$$v^{}|v(A)=\stackrel{~}{v}^{}|\stackrel{~}{v}(A)\frac{det\left[1P(0)+K(A)P(0)\right]}{|det\left[1P(0)+K(A)P(0)\right]|}$$
$`(12.16)`$
By definition, $`v^{}|=\stackrel{~}{v}^{}|`$. So long the matrix $`1P(0)+K(A)P(0)`$ is not singular the two phase choices are smoothly and locally related.
In numerical work one needs more explicit expressions for the various matrix elements in Fock space we encountered. For completeness I include below an overlap “master formula”, provable by purely combinatorial means . This formula tells how to transcribe all Fock space expressions one may need into single particle language.
Assume that we have the rectangular matrix $`v`$ as above and another matrix of similar structure $`u`$. In applications $`u`$ can be taken as $`v^{}`$ or as $`v`$. The main assumption is that the columns of $`u`$ are linearly independent and their number is the same as in $`v`$. Let $`P_v,P_u`$ denote the projectors on the subspaces spanned by the columns:
$$P_v=vv^{},P_u=uu^{},v^{}v=1,u^{}u=1$$
$`(12.17)`$
Let $`\widehat{a}^{}X\widehat{a}=\widehat{X}`$ be a bilinear operator. Then:
$$\begin{array}{cc}& u|v=det(u^{}v)\hfill \\ \hfill u|\widehat{X}|v=& det(u^{}v)Tr\left[XP_v\frac{1}{1P_uP_v}\right]=det(u^{}v)Tr\left[P_uX\frac{1}{1P_uP_v}\right]\hfill \end{array}$$
$`(12.18)`$
For $`u=v`$ we can use $`P_v(12P_v)=P_v`$ to get
$$v|\widehat{X}|v=Tr(XP_v)$$
$`(12.19)`$
The expression $`\frac{1}{1P_uP_v}`$ comes accompanied by $`det(u^{}v)`$ and reflects the nonlocality in a theory with massless fermions when $`u=v^{}`$. However, for $`u=v`$ the nonlocality disappears. This is how the Wess-Zumino action becomes local in the first place. Equation (12.18) with $`u=v^{}`$ makes it explicit that all gauge breaking comes from the fermion determinant factor $`u|v`$. Setting $`u=v`$ typically produces expressions with naive gauge transformation properties.
Another combinatorial fact, proven in is that even for $`uv`$ Wick’s theorem still holds. By this I mean that the $`uv`$ matrix element of any string of fermionic creation and annihilation operators is given by sums of products of $`u|a_\mu ^{}(x)a_\nu (y)|v`$, where each matrix element is divided by the overlap $`u|v`$. Therefore, equation (12.18) provides enough information to evaluate $`u|\widehat{X}|v`$ for any $`\widehat{X}`$, not just bilinear ones.
13. Symmetry properties of adiabatic phase choice
Relation (12.16) is now used to show that left and right handed chiral fermions of the same charge have determinants of opposite phase. First relation (12.16) has to be generalized to the opposite handedness. This is easy: replace all $`v`$’s by $`w`$’s, which induces also replacing the projector $`P(A)`$ by $`1P(A)`$. The matrix $`K(A)`$ remains unchanged, because the defining first order differential equation and the initial condition have not changed. Also, $`detK(A)=1`$ because it is so at $`A=0`$ and the evolution does not change the determinant. Taking a complex conjugate of (12.16) we get:
$$\begin{array}{cc}\hfill v^{}|v(A)^{}=& \stackrel{~}{v}^{}|\stackrel{~}{v}(A)^{}\frac{det\left[1P(0)+P(0)K^{}(A)\right]}{|det\left[1P(0)+P(0)K^{}(A)\right]|}=\hfill \\ & \stackrel{~}{v}^{}|\stackrel{~}{v}(A)^{}\frac{det\left[[1P(0)]K(A)+P(0)\right]}{|det\left[[1P(0)]K(A)+P(0)\right]|}=\hfill \\ & \stackrel{~}{v}^{}|\stackrel{~}{v}(A)^{}\frac{det\left[P(0)+K(A)[1P(0)]\right]}{|det\left[P(0)+K(A)[1P(0)]\right]|}\hfill \end{array}$$
$`(13.1)`$
We know from that
$$\stackrel{~}{v}^{}|\stackrel{~}{v}(A)^{}=\stackrel{~}{w}^{}|\stackrel{~}{w}(A)$$
$`(13.2)`$
With this substitution we get
$$v^{}|v(A)^{}=w^{}|w(A)$$
$`(13.3)`$
Therefore, one of the key properties of the BW phase choice, namely that the imaginary part of the induced action switches sign when handedness is switched also holds with the adiabatic phase choice. Also, the real parts of the induced action are identical for the left and right handedness.
The above property is true in any dimensions. In some special dimensions, like four, one can deal with fermions of only one handedness and use conjugate representations for the other handedness and thus avoid defining separately lattice fermions of left and right handedness. But, his would not work in two dimensions for example.
It is now not hard to see how all other symmetry properties of the overlap with a BW phase choice hold also with the adiabatic phase choice. One always uses the relation (12.16) and symmetry properties of $`H(A)`$. The symmetry properties of the matrix $`H(A)`$ are directly inherited by $`K(A)`$, so long the free fermion ground state is invariant, making $`P(0)`$ invariant.
14. Nonabelian case
It is quite clear from the above that the abelian nature of the gauge group enters in many places. The most technical component in the above construction is the decomposition of the current into a divergenceless piece, a topological piece and a gauge invariant piece. For our abelian case this technical part could probably be done more efficiently in Fourier space. But, when one contemplates a generalization to the nonabelian case, coordinate space might be better. In coordinate space the abelian current decomposition formula comes from a study of local abelian lattice BRS cohomology. The continuum BRS cohomology structure of nonabelian gauge theories has been transcribed to the lattice in references . There one also dealt with cohomology issues, but not the local version, which is relevant here. It is conceivable that on these lines some generalizations to the nonabelian case can be found. Even if this is done, one still has more work to do if one wants to treat nonabelian lattice chiral gauge theories on lines that closely generalize the abelian treatment of this paper.
15. Summary
In the particular case of noncompact chiral abelian gauge theories at infinite volume it seems possible to formulate a lattice version which for all purposes of principle has all the simplicity of a continuum formulation, only now holding nonperturbatively. It is possible to make what was termed in the last section of a “perfect” phase choice in this abelian noncompact case: restoration of gauge invariance is possible if and only if perturbative anomalies cancel. In the anomalous case one has an effective theory that looks very much what one would write down based on perturbative arguments. One could make concrete the meaning of the nonrenormalizability bounds introduced by Preskill and Wise in the abelian case, this time outside perturbation theory. Generalization of the perfect phase choice to the nonabelian case seems difficult. On the other hand, the construction in the abelian case is quite physical and the concept of adiabatic evolution does generalize.
Technically, the physical aspects of the construction are much more transparent in Fock space (second quantized language). This is not unexpected in the context of field theory. But, the field theory is quadratic, because we only quantize some auxiliary fermions in a fixed gauge background. Therefore, all expressions can be easily transcribed into single particle language. This language is needed in making the expressions useful for numerical work. The translation between the two languages is just a matter of some combinatorics.
This paper does not shed new light on the question whether a perfect phase choice should be viewed as fine tuning or not. Since we work in the noncompact case the issue of gauge averaging a slightly imperfect phase choice cannot be directly addressed. An indirect approach was sketched, but seems difficult to implement computationally at present.
If we think about the standard model, we now can put on the lattice a theory with the fermion content of one or more generations, in which we gauged the largest continuous abelian subgroup of $`U(1)\times SU(2)\times SU(3)`$ and maintained exact (noncompact) gauge invariance. The entire intricate mechanism of anomaly cancelation comes then into play. For example, the well known restrictions on physical charges apply now in a nonperturbative setting. So, we probably are one step closer to putting the entire minimal standard model on the lattice.
Acknowledgments
My research at Rutgers is partially supported by the DOE under grant \# DE-FG05-96ER40559.
References
1 R. Narayanan, H. Neuberger, Phys. Lett. B 302, 62 (1993); Phys. Rev. Lett. 71, 3251 (1993); Nucl. Phys. B 412, 574 (1994).
2 M. Lüscher, hep-lat/9909150.
3 R. Narayanan, H. Neuberger, Nucl. Phys. B 443, 305 (1995).
4 H. Neuberger, hep-lat/9912013.
5 D. B. Kaplan, Phys. Lett. B 288, 342 (1992).
6 S. A. Frolov, A. A. Slavnov, Phys. Lett. B 309, 344 (1993).
7 C. Callan, J. Harvey, Nucl. Phys. B 250, 427 (1984).
8 G. ’t Hooft, Phys. Lett. B 349, 491 (1995).
9 J. Zinn-Justin, Quantum Field Theory and Critical Phenomena (Oxford University Press, 1993).
10 J. Preskill, Annals of Physics 210, 323 (1991).
11 H. Neuberger, Nucl. Phys. Proc. Suppl. 73, 697 (1999).
12 B. Simon, Phys. Rev. Lett. 51, 2167 (1983).
13 H. Neuberger, Phys. Rev. D 59, 085006 (1999).
14 M. Berry, Phys. Rev. Lett. 392, 45 (1984).
15 H. Neuberger, hep-lat/9911004.
16 P. Ginsparg, K. Wilson, Phys. Rev. D 25, 2649 (1982).
17 S. Randjbar-Daemi, J. Strathdee, Nucl. Phys. B 466, 335 (1996).
18 T. Kato, J. Phys. Soc. Jpn. 5, 435 (1950).
19 R. Narayanan, H. Neuberger, Nucl. Phys. B 477, 521 (1996).
20 K. Fujikawa, Phys. Rev. Lett. 42, 1195 (1979).
21 Y. Shamir, PRD 57, 132 (1998).
22 H. Neuberger, Phys. Lett. B 417, 141 (1998).
23 T. Fujiwara et. al., hep-lat/9906015.
24 D. Förster, H. B. Nielsen, M. Ninomiya, Phys. Lett. B 94, 135 (1980).
25 U. M. Heller et. al., Nucl. Phys. B 405, 555 (1993).
26 H. Neuberger, Phys. Lett. B 175, 69 (1986); Phys. Lett. B 183, 337 (1987).
|
warning/0002/hep-th0002157.html
|
ar5iv
|
text
|
# Contents
## 1 Introduction
We find a new stable D2-D0 brane bound state of the non-commutative fuzzy sphere. This is realized as an ellipsoid with three independent angular momenta corresponding to rotation in three different planes. The idea is to cancel the attractive force due to tension with the repulsive centrifugal force. As a result, the three angular velocities are uniquely determined through the principal radii of the ellipsoid. We compare this to the spherical membrane solution where the sphere oscillates in a manner that inverts the orientation of the sphere at the point where the geometry is singular. In contrast, our solution is non-singular at all times.
Before presenting our new solution in Section 3 we give in Section 2.1 the necessary background on the matrix mechanics of D0-branes, and in particular the new approaches to the coupling of D0-branes to higher form RR fields. It is sufficient to consider the 3-form RR field which corresponds to D2-brane charge. In Section 2.2, we review the spherical membrane solution, the essential elements of which we use in our construction of the rotating ellipsoidal membrane in Section 3.
In Section 4 we analyze the classical stability of our solution against small perturbations in the initial conditions. It is known that the diagonal oscillating sphere solution is unstable , resulting in the system spending most of the time in long, string-like configurations. In contrast, our rotating solution is stable against small perturbations of initial conditions, as evidenced by the absence of non-unitary Poincare-Lyapunov exponents in the S-matrix. We use the full machinery of the Hamiltonian theory of dynamical systems to isolate the zero modes of the variational equations of motions. All the remaining eigenvalues were measured numerically and found to be unitary.
The system exhibits interesting dynamical properties due to the interaction with various supergravity fields. This leads to semi-classical energy loss, and we compute in Section 5 the functional form and total power emitted by the system in three different sectors.
Section 5.1 is devoted to discussion of radiation carried by the RR 1-form field $`C^{(1)}`$. The following phenomenon appears: even though D0-branes naturally couple to this field and create a static potential, the dipole moment vanishes because only one kind of charge is present. Therefore we look at quadrupole order in search of $`C^{(1)}`$ radiation, and find that for the case of a sphere of radius $`R`$ rotating at the frequency $`\omega `$, the power is proportional to $`\omega ^{12}`$,
$$P=\frac{2^8}{3^35^27^211}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}\omega ^{12}R^4.$$
(1)
In Section 5.2 we use the machinery developed in Section 2.1, and consider the 3-form RR field radiation which corresponds to D2-brane charge. A spherical membrane would not carry a net charge due to cancellation of opposite oriented pieces, but a dipole moment is non-vanishing in accordance with naive expectation. This rotating dipole produces radiation which in most respects behaves like ordinary dipole radiation, with only difference being that the gauge field has three indices instead of just one, in the case of a vector gauge field. This broadly fits with the understanding of RR gauge fields as corresponding to photon-like “spin-one” particle, even though there is no invariant meaning to the notion of spin in 10 dimensions. The power is proportional to $`\omega ^{10}`$, but owing to the specific relation between $`R`$ and $`\omega `$ for our system turns out to differ only in the numerical coefficient
$$P=\frac{3^7}{2^25^27^2}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}\omega ^{10}\alpha ^2R^6.$$
(2)
For gravitational radiation we find in Section 5.3 radiation at the lowest possible order, namely quadrupole. The final result is
$$P=\frac{2^6}{35711}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}\omega ^{12}R^4.$$
(3)
We note that the remaining massless field, the dilaton, at lowest order couples to $`F^2`$ and this quantity turns out to be time-independent. However, at higher order the dilaton may couple to time-dependent combinations of $`F`$. The calculation would require knowledge of higher order terms in the non-abelian DBI action, and we leave this to future work.
Following , we discuss the fundamental particle interpretation of our system in Section 6. It is possible that there exists a completely stable ground state, protected by certain conservation laws from further decay. This particle would appear to the external observer to have N units of D0-brane charge, and (presumably quantized) internal angular momentum.
## 2 Preliminaries
### 2.1 The dynamics of $`N`$ D0-branes
The system of $`N`$ interacting D0-branes in type IIA string theory has been studied extensively in recent years. One of the reasons for this is the conjecture that it should describe the discrete light-cone quantized (DLCQ) M-theory.
Before presenting the action for $`N`$ D0-branes we first shortly review the dynamics of the background fields. The background fields for the $`N`$ D0-branes are governed by the type IIA low energy effective action
$`S_{\mathrm{IIA}}`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle }d^{10}x\sqrt{g}[e^{2\varphi }(R+4(\varphi )^2{\displaystyle \frac{1}{12}}H^2)`$ (4)
$`{\displaystyle \frac{1}{4}}(F^{(2)})^2{\displaystyle \frac{1}{48}}(F^{(4)})^2]{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle \frac{1}{2}}{\displaystyle }BdC^{(3)}dC^{(3)}`$
where $`\varphi `$ is the dilaton field, $`g_{\mu \nu }`$ is the metric, $`H=dB`$ is the NSNS three-form field strength, $`F^{(2)}=dC^{(1)}`$ is the RR two-form field strength and $`F^{(4)}=dC^{(3)}+HC^{(1)}`$ is the RR four-form field strength. The system of $`N`$ D0-branes couples to all of these fields.
The effective action of $`N`$ D0-branes for weak and slowly varying fields is the non-abelian $`U(N)`$ Yang-Mills action plus the Chern-Simons action (for the bosonic part). For weak fields the action is gotten by dimensionally reducing the action of 9+1 dimensional $`U(N)`$ Super Yang-Mills theory to 0+1 dimensions . Up to a constant term it is
$$S=T_0(2\pi l_s^2)^2𝑑tTr\left(\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\right),$$
(5)
where $`F_{\mu \nu }`$ is the non-abelian $`U(N)`$ field strength in the adjoint representation and $`T_0=(g_sl_s)^1`$ is the D0-brane mass. To write this action in terms of coordinate matrices $`X^i`$, one has to use the dictionary
$$A_i=\frac{1}{2\pi l_s^2}X^i,F_{0i}=\frac{1}{2\pi l_s^2}\dot{X}^i,F_{ij}=\frac{1}{(2\pi l_s^2)^2}i[X^i,X^j]$$
(6)
with $`i,j=1,2,\mathrm{},9`$, giving
$$S=T_0𝑑tTr\left(\frac{1}{2}\dot{X}^i\dot{X}^i+\frac{1}{4}\frac{1}{(2\pi l_s^2)^2}[X^i,X^j][X^i,X^j]\right)$$
(7)
To derive this it is necessary to gauge the $`A_0`$ potential away, which is possible for a non-compact time. The Gauss constraint
$$[\dot{X}_i,X^i]=0$$
(8)
persists, and the above action should be taken together with it .
The assumed approximation that the fields $`F_{0i}`$ and $`F_{ij}`$ in (6) should be weak and slowly-varying corresponds for the coordinate matrices $`X^i`$ to
$$\dot{X}^i1,[X^i,X^j]l_s^2,\ddot{X}^il_s^1,[\dot{X}^i,X^j]l_s.$$
(9)
The condition $`\dot{X}^i1`$ means that the D0-branes are in the non-relativistic limit.
Recently it has become clear that even though the D0-brane world-volume is only one-dimensional, a multiple D0-brane system can also couple to the branes of higher dimension via the Chern-Simons action. The Chern-Simons action not only tells us how $`N`$ D0-branes move in the weak background fields of Type IIA supergravity, but also what higher fields the D0-branes produce. Using this, one can see that it is possible to build the higher $`p`$-branes of Type IIA string theory out of D0-branes . The Chern-Simons action derived in for the coupling of $`N`$ D0-branes to bulk RR $`C^{(1)}`$ and $`C^{(3)}`$ fields is
$$S_{CS}=T_0𝑑tTr\left(C_0+C_i\dot{X}^i+\frac{1}{2\pi l_s^2}\left[C_{0ij}[X^i,X^j]+C_{ijk}[X^i,X^j]\dot{X}^k\right]\right)$$
(10)
The idea that a lower-dimensional object under the influence of higher-form RR fields may nucleate into spherically or cylindrically wrapped D-brane was proposed by Emparan for the case of fundamental string. Before that Callan and Maldacena constructed a D- or F-string as a BI soliton solution on the D-brane where the attached string appears as a spike on the brane.
We first clarify the notion of the coupling of the D-brane world-volume RR current to the external form-field, in particular for wrapped D-branes. This will give us a better understanding of the similarity between the D2-brane current and the corresponding expression for a system of D0-branes.
In general the D2-brane couples to the bulk RR field through the well-known CS coupling
$$S_{CS}=C^{(3)}=C_{\mu \nu \rho }J^{\mu \nu \rho }d^3\sigma $$
(11)
where $`J`$ is a three form RR current
$$J^{\mu \nu \rho }=ϵ^{\alpha \beta \gamma }_\alpha X^\mu _\beta X^\nu _\gamma X^\rho ,$$
(12)
or in form notation
$${}_{}{}^{}J_{(3)}^{\mu \nu \rho }=dX^\mu dX^\nu dX^\rho .$$
(13)
One can introduce a charge corresponding to this current, such that it is a world-volume 2-form even though it has three space-time indices, same as the current
$`Q_{\beta \gamma }^{\mu \nu \rho }`$ $`=`$ $`X^{[\mu }_\beta X^\nu _\gamma X^{\rho ]},\mathrm{so}\mathrm{that}`$
$`J_{\alpha \beta \gamma }^{\mu \nu \rho }`$ $`=`$ $`3_{[\alpha }Q_{\beta \gamma ]}^{\mu \nu \rho },`$ (14)
or in form notation
$`Q_{(2)}^{\mu \nu \rho }`$ $`=`$ $`\left(dX^{[\mu }dX^\nu \right)X^{\rho ]},\mathrm{such}\mathrm{that}`$
$`dQ_{(2)}^{\mu \nu \rho }`$ $`=`$ $`J_{(3)}^{\mu \nu \rho },`$ (15)
where the exterior derivative is taken with respect to world-volume indices. This charge first appeared in the theory of bosonic relativistic membrane . There it was interpreted as a topological charge of the membrane. The connection between RR charges and D-branes was discovered by Polchinski .
Instead of (12) one can represent the current as a Poisson bracket with respect to the spatial world-volume coordinates. For a static membrane, the non-zero components of $`J^{\mu \nu \rho }`$ are
$$J^{0ij}=\{X^i,X^j\},$$
(16)
For a moving membrane the completely spatial components also appear. A convenient generalization is, in the static gauge
$$J^{ijk}=\dot{X}^i\{X^j,X^k\}.$$
(17)
The above discussion is for the ordinary D2-brane, and the coupling of the D0-brane matrix-mechanical system to the $`C^{(3)}`$ field is completely analogous
$$S_{CS}=\frac{T_0}{2\pi l_s^2}𝑑tTr\left(C_{0ij}[X^i,X^j]+C_{ijk}[X^i,X^j]\dot{X}^k\right)=CJ𝑑t$$
(18)
The trace of the first term is zero identically, reflecting the fact that the total bare RR charge of the object that we are considering is zero, due to cancellation of the pieces with opposite orientation on the 2-sphere. One needs to expand the expression in powers of $`X`$, to effectively obtain the multipole expansion of $`C^{(3)}`$. Then, it is possible to integrate by parts, to get instead the coupling of the field strength to the charge
$$S_{CS}=\frac{T_0}{2\pi l_s^2}𝑑tF_{0ijk}Tr[X^i,X^j]X^k=FQ𝑑t$$
(19)
In Section 5.2 we will use this expression for the D2-brane dipole moment $`Q`$ to compute $`C^{(3)}`$ dipole radiation. This is convenient because $`J`$ can be easily obtained by time differentiation.
### 2.2 The spherical membrane
In this section we briefly review the spherical D2-D0 brane configuration of type IIA string theory since our new solution of the system of $`N`$ D0-branes presented in Section 3 uses the essential elements of this construction. The solution is equivalent to the spherical membrane solution of M(atrix) theory .
We aim to construct a membrane with an $`S^2`$ geometry. We embed the $`S^2`$ in a three dimensional space spanned by the 123 directions. We take the ansatz
$$X_i(t)=\frac{2}{\sqrt{N^21}}𝐓_ir_i(t),i=1,2,3$$
(20)
where the $`N\times N`$ matrices $`𝐓_1,𝐓_2,𝐓_3`$ are the generators of the $`N`$ dimensional irreducible representation of $`SU(2)`$, with algebra
$$[𝐓_i,𝐓_j]=iϵ_{ijk}𝐓_k$$
(21)
and with the quadratic Casimir
$$\underset{i=1}{\overset{3}{}}𝐓_i^2=\frac{N^21}{4},\mathrm{so}\mathrm{that}Tr(𝐓_i^2)=\frac{N(N^21)}{12}.$$
(22)
For vanishing background fields the Hamiltonian is<sup>3</sup><sup>3</sup>3This system is otherwise known as $`0+1`$ dimensional classical SU(2) YM mechanics .
$$H=\frac{NT_0}{3}\left[\frac{1}{2}\underset{i=1}{\overset{3}{}}\dot{r}_i^2+\frac{\alpha ^2}{2}\left(r_1^2r_2^2+r_1^2r_3^2+r_2^2r_3^2\right)\right],$$
(23)
where we have introduced the convenient parameter $`\alpha =\frac{2}{\sqrt{N^21}}\frac{1}{2\pi l_s^2}`$. This gives the equations of motion
$`\ddot{r}_1=\alpha ^2(r_2^2+r_3^2)r_1`$
$`\ddot{r}_2=\alpha ^2(r_1^2+r_3^2)r_2`$
$`\ddot{r}_3=\alpha ^2(r_1^2+r_2^2)r_3`$ (24)
Let us estimate the physical size of the resulting object. For simplicity we take all radii to be equal to each other: $`r_1=r_2=r_3=r`$. With this we have from (20) and (22) the physical radius of the membrane
$$R^2=X_1^2+X_2^2+X_3^2=𝐈r^2$$
(25)
where $`𝐈`$ is the $`N\times N`$ identity matrix. The formula (25) shows that the $`N`$ D0-branes are constrained to lie on an $`S^2`$ sphere of radius $`r`$. This condition is what determines the normalization of the ansatz (20).
Considering the Chern-Simons action (19) we clearly see that the coupling of this system to $`F_{0123}`$ is non-vanishing. This means that the spherical membrane solution has a D2-brane dipole moment, and it is thus appropriate to recognize the spherical membrane solution as a bound state of a spherical D2-brane and $`N`$ D0-branes .
The equations of motion (2.2) reduce to the equation $`\ddot{r}=2\alpha ^2r^3`$. Thus, it is clearly not possible to have a static solution<sup>4</sup><sup>4</sup>4Myers considers the system of N D0-branes in a constant external 4-form RR field strength, as in (19). Then indeed there is a static solution, where the D0-branes polarize and arrange into a static spherical configuration.. Instead, the solution is
$$r(t)=R_0sn(R_0\alpha t+\varphi )$$
(26)
where $`sn`$ is the sinus amplitudinis of Jacobi defined by
$$y=sn(x)x=_0^y\frac{1}{\sqrt{1z^4}}𝑑z$$
(27)
The solution (26) is thus an $`S^2`$ sphere with the radius $`r(t)`$ oscillating between $`R_0`$ and $`R_0`$ with a period of $`\frac{\mathrm{\Gamma }(1/4)^2}{\sqrt{2\pi }}\frac{1}{\alpha R_0}`$. Clearly the spherical membrane will be classically point-like at the nodes of the sinus. Thus the membrane solution will break down after a finite amount of time, since the classical solution may not be valid at substringy distances, and possibly decay into a Schwarzschild black hole .
The conditions (9) for the spherical membrane solution translate into
$$|r(t)|l_s\sqrt{N},|\dot{r}(t)|1,|\ddot{r}(t)|l_s^1$$
(28)
where we have used the fact that the matrices $`𝐓`$ are of order $`N`$. Since we also require $`|r(t)|l_s`$ we must have $`N1`$. Thus it is necessary to have a large amount of D0-branes to build a macroscopic spherical membrane. In fact, the maximal allowed size with the use of N D0-branes is proportional to $`\sqrt{N}`$.
## 3 The rotating ellipsoidal membrane
The spherical membrane solution reviewed in Section 2.2 is not a stable object of string theory. It is classically unstable under small perturbations, and also reaches a point-like singularity in finite time. This makes it possible for it to collapse into a near-extremal Schwarzschild black hole. In this section we present instead another kind of solution to the system of $`N`$ D0-branes which is free of singularities. We also show in Section 4 that it is a stable object, in the sense of stability under small perturbations of initial conditions. Moreover it has interesting physical properties due to its dynamical nature, like the radiation of various SUGRA fields (see Section 5). The basic idea in the construction is that the attractive force of tension should be cancelled by the centrifugal repulsion force. The motion is at all times transverse, and cannot be gauged away by coordinate reparametrization invariance on the membrane.
We now construct a rotating ellipsoidal membrane, viewed as a collection of D0-branes, in such a manner as to cancel the attractive force of tension with the centrifugal repulsion force. For that we need to take the basic configuration (20) of the non-commutative fuzzy sphere in the 135 directions, and set it to rotate in the transverse space along three different axis, i.e. in the 12, 34 and 56 planes. We thus use a total of 6 space dimensions to embed our D-brane system. The corresponding ansatz is<sup>5</sup><sup>5</sup>5A less general ansatz was considered in to model a spherical rotating membrane, see also . Moreover, a generic rotation-ansatz was proposed in .
$`X_1(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_1r_1(t),X_2(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_1r_2(t),`$
$`X_3(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_2r_3(t),X_4(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_2r_4(t),`$
$`X_5(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_3r_5(t),X_6(t)={\displaystyle \frac{2}{\sqrt{N^21}}}𝐓_3r_6(t).`$ (29)
In this new ansatz we are still using the $`SU(2)`$ matrix structure from (20) such that the coordinate matrices are proportional to the $`SU(2)`$ generators in pairs. We interpret this as a rotation because one could make a rotation in e.g. the 12 plane which makes one of the components vanish, say $`X_2`$, while the other one gets a radius $`r_1^{}=\sqrt{r_1^2+r_2^2}`$. This is possible exactly because both are proportional to the same matrix $`𝐓_1`$. The end result is that at any point in time one can choose a coordinate system in which the object spans only three space dimensions.
Substituting the ansatz into (7) gives the Hamiltonian
$`H={\displaystyle \frac{NT_0}{3}}({\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{6}{}}}\dot{r}_i^2+{\displaystyle \frac{\alpha ^2}{2}}[(r_1^2+r_2^2)(r_3^2+r_4^2)`$
$`+(r_1^2+r_2^2)(r_5^2+r_6^2)+(r_3^2+r_4^2)(r_5^2+r_6^2)])`$ (30)
The corresponding equations of motion are
$`\ddot{r}_1=\alpha ^2(r_3^2+r_4^2+r_5^2+r_6^2)r_1,\ddot{r}_2=\alpha ^2(r_3^2+r_4^2+r_5^2+r_6^2)r_2,`$
$`\ddot{r}_3=\alpha ^2(r_1^2+r_2^2+r_5^2+r_6^2)r_3,\ddot{r}_4=\alpha ^2(r_1^2+r_2^2+r_5^2+r_6^2)r_4,`$
$`\ddot{r}_5=\alpha ^2(r_1^2+r_2^2+r_3^2+r_4^2)r_5,\ddot{r}_6=\alpha ^2(r_1^2+r_2^2+r_3^2+r_4^2)r_6.`$ (31)
We have found the special solution to these equations describing a rotating ellipsoidal membrane with three distinct principle radii $`R_1`$, $`R_2`$ and $`R_3`$
$`r_1(t)=R_1\mathrm{cos}(\omega _1t+\varphi _1),r_2(t)=R_1\mathrm{sin}(\omega _1t+\varphi _1),`$
$`r_3(t)=R_2\mathrm{cos}(\omega _2t+\varphi _2),r_4(t)=R_2\mathrm{sin}(\omega _2t+\varphi _2),`$
$`r_5(t)=R_3\mathrm{cos}(\omega _3t+\varphi _3),r_6(t)=R_3\mathrm{sin}(\omega _3t+\varphi _3).`$ (32)
This particular functional form of the solution ensures that the highly non-linear equations for any of the components $`r_i`$ are reduced to a harmonic oscillator. The solution (3) keeps $`r_1^2+r_2^2=R_1^2`$ , $`r_3^2+r_4^2=R_2^2`$ and $`r_5^2+r_6^2=R_3^2`$ fixed which allows us to say that the object described by (3) rotates in six spatial dimensions as a whole without changing its basic shape.
Using the equations of motion (3), the three angular velocities are determined by the radii, and do not necessarily have to coincide:
$$\omega _1=\alpha \sqrt{R_2^2+R_3^2},\omega _2=\alpha \sqrt{R_1^2+R_3^2},\omega _3=\alpha \sqrt{R_1^2+R_2^2}.$$
(33)
This dependence of the angular frequency on the radii we interpret as if the repulsive force of rotation has to be balanced with the attractive force of tension in order for (3) to be a solution. Thus the radii $`R_1`$, $`R_2`$ and $`R_3`$ parameterize (3) along with the three phases $`\varphi _i`$, to produce altogether a six parameter family of solutions. Unless explicitly stated otherwise, we set $`\varphi _i=0`$ in what follows.
In order to exhibit the properties of the solution (3) we compute the components of the angular momentum
$$M_{ij}=Tr\left[X^i\mathrm{\Pi }^jX^j\mathrm{\Pi }^i\right],$$
(34)
where $`\mathrm{\Pi }^i=T_0\dot{X}^i`$. As expected, the only non-zero components are
$$M_{12}=\frac{1}{3}NT_0\omega _1R_1^2,M_{34}=\frac{1}{3}NT_0\omega _2R_2^2,M_{56}=\frac{1}{3}NT_0\omega _3R_3^2.$$
(35)
The angular momenta $`M_{12}`$, $`M_{34}`$ and $`M_{56}`$ correspond to rotations in the 12, 34 and 56 planes respectively. Their values (35) fit with the interpretation of the solution being $`N`$ D0-branes rotating as an ellipsoidal membrane in that they are time-independent due to conservation law and proportional to $`NT_0\omega _iR_i^2`$.
The conditions (9) on the solution (3) give
$$\omega _iR_i1,R_iR_jNl_s^2,\omega _i^2R_il_s^1,\omega _iR_iR_jNl_s,ij$$
(36)
Using (33) this is seen to be equivalent to
$$R_iR_jNl_s^2,R_iR_j^2N^2l_s^3,ij$$
(37)
Since we also require the membrane to have a size larger than the string length $`l_s`$, we must have $`N1`$, just as for the spherical membrane reviewed in Section 2.2. In the special case $`R_1=R_2=R_3=R`$ the conditions (37) reduce to $`R\sqrt{N}l_s`$.
## 4 Stability analysis
In this section we consider the problem of stability of our solution (3) with respect to an arbitrary change of initial conditions. In principle, there may be other possible sources of instability, classical or quantum. The quantum, or rather quasi-classical, loss of energy associated with radiation of various supergravity fields is computed in Section 5.
In Hamiltonian systems there is no naturally defined positive-definite metric on the phase space, therefore so-called Lyapunov exponents are only meaningful when measured on time-slices of a periodic trajectory, as used by Poincare. Thus in what follows we should take $`\omega _{1,2,3}`$ to be multiples of a lowest common frequency $`\omega _0`$, such that the trajectory is periodic with $`T_0=\frac{2\pi }{\omega _0}`$. Our formulas are applicable to just such a situation, but when it comes to measuring the exponents we take the simplest case $`\omega _0=\omega _1=\omega _2=\omega _3`$, in which the ellipsoid becomes a sphere.
Let us consider linear perturbations to the equations of motion (3), which are essential in all kinds of stability analysis. These explicitly time-dependent, or else called non-autonomous, equations should be understood to contain the original solutions (3) as time-dependent external fields:
$`\alpha ^2\delta \ddot{r}_1=\delta r_1(R_2^2+R_3^2)2r_1\left(r_3\delta r_3+r_4\delta r_4+r_5\delta r_5+r_6\delta r_6\right)`$
$`\alpha ^2\delta \ddot{r}_2=\delta r_2(R_2^2+R_3^2)2r_2\left(r_3\delta r_3+r_4\delta r_4+r_5\delta r_5+r_6\delta r_6\right)`$
$`\alpha ^2\delta \ddot{r}_3=\delta r_3(R_1^2+R_3^2)2r_3\left(r_1\delta r_1+r_2\delta r_2+r_5\delta r_5+r_6\delta r_6\right)`$
$`\alpha ^2\delta \ddot{r}_4=\delta r_4(R_1^2+R_3^2)2r_4\left(r_1\delta r_1+r_2\delta r_2+r_5\delta r_5+r_6\delta r_6\right)`$
$`\alpha ^2\delta \ddot{r}_5=\delta r_5(R_1^2+R_2^2)2r_5\left(r_1\delta r_1+r_2\delta r_2+r_3\delta r_3+r_4\delta r_4\right)`$
$`\alpha ^2\delta \ddot{r}_6=\delta r_6(R_1^2+R_2^2)2r_6\left(r_1\delta r_1+r_2\delta r_2+r_3\delta r_3+r_4\delta r_4\right)`$ (38)
These describe the time evolution of $`\delta 𝐫(t)`$, the linear deviation from a given solution $`𝐫(t)`$. We could rewrite these equations in first-order form, and introduce the 12-component perturbation vector
$$𝐚(t)=(\delta r_1(t),\delta \dot{r}_1(t),\delta r_2(t),\delta \dot{r}_2(t),\mathrm{},\delta r_6(t),\delta \dot{r}_6(t))$$
(39)
We have then the first-order linear system of differential equations
$$\dot{𝐚}(t)=𝐇(t)𝐚(t)$$
(40)
where $`𝐇(t)`$ is a $`12\times 12`$ real matrix, put together from a $`6\times 6`$ matrix to be read off RHS of (38) and a $`6\times 6`$ unit matrix in the appropriate order of components. This matrix is periodic with the period equal to that of the original solution. Time ordered exponentiation gives us the formal expression for the complete evolution S-matrix
$$𝐒(t)=:\left(\mathrm{exp}_0^t𝐇(\tau )d\tau \right):,$$
(41)
so that $`\dot{𝐒}(t)=𝐇(t)`$. We can now find the solution vector by applying the S-matrix to the initial vector:
$$𝐚(t)=𝐒(t)𝐚(0).$$
(42)
In order to investigate the stability of the original periodic trajectory we need to find out whether small perturbations grow. This information is encoded in a single marix $`𝐒(T_0)`$ which gives the evolution of the perturbation vector over one period. This matrix is symplectic, and has unit determinant in order to preserve the Liouville measure on the phase space.
The original solution is parameterized by $`R_1,R_2,R_3`$ and $`\varphi _1,\varphi _2,\varphi _3`$. A small variation of either of these parameters produces a new solution. Together, they give 6 different zero modes of the system. Technically, zero modes are produced by formally differentiating the solutions (3) with respect to a parameter. Physically, this is very familiar phenomenon, e.g. the free scale parameter in the instanton of YM gives a zero mode in exactly the same fashion as here. Zero modes correspond to unit eigenvalues of the S-matrix. The zero modes for $`/\varphi _i`$, $`i=1,2,3`$, are periodic since for example
$$\delta r_1(t)=\frac{}{\varphi _1}r_1(t)=R_1\mathrm{sin}(\omega _1t+\varphi _1)$$
(43)
The initial conditions for $`/\varphi _i`$, $`i=1,2,3`$ are
$`𝐚_{\varphi _1}(0)`$ $`=`$ $`(0,\omega _1R_1,R_1,0,0,0,0,0,0,0,0,0)`$
$`𝐚_{\varphi _2}(0)`$ $`=`$ $`(0,0,0,0,0,\omega _2R_2,R_2,0,0,0,0,0)`$
$`𝐚_{\varphi _3}(0)`$ $`=`$ $`(0,0,0,0,0,0,0,0,0,\omega _3R_3,R_3,0)`$ (44)
On the other hand, the $`/R_i`$ initial vectors are
$`𝐚_{R_1}(0)`$ $`=`$ $`(1,0,0,\omega _1,0,0,0,\alpha ^2{\displaystyle \frac{R_1R_2}{\omega _2}},0,0,0,\alpha ^2{\displaystyle \frac{R_1R_3}{\omega _3}})`$
$`𝐚_{R_2}(0)`$ $`=`$ $`(0,0,0,\alpha ^2{\displaystyle \frac{R_2R_1}{\omega _1}},1,0,0,\omega _2,0,0,0,\alpha ^2{\displaystyle \frac{R_2R_3}{\omega _3}})`$
$`𝐚_{R_3}(0)`$ $`=`$ $`(0,0,0,\alpha ^2{\displaystyle \frac{R_3R_1}{\omega _1}},0,0,0,\alpha ^2{\displaystyle \frac{R_3R_2}{\omega _2}},1,0,0,\omega _3).`$ (45)
These three modes correspond to nearby trajectories of the same form as (3) and have a slightly different period. Consequently the trajectory remains in the vicinity, but gets either ahead or behind of the original one linearly in time due to the difference in period. The direction of this linearly growing deviation is, not surprisingly a combination of $`𝐚_{\varphi _i}`$. For example after $`k`$ periods
$$𝐚_{R_1}(kT_0)=𝐒(kT_0)𝐚(0)=𝐒(T_0)^k𝐚(0)=𝐚_{R_1}(0)+kT_0\left[𝐚_{\varphi _2}(0)+𝐚_{\varphi _3}(0)\right]$$
(46)
Now that we have singled out the zero modes, we can proceed to find the other 6 eigenvalues by integrating the equations numerically and discarding the unit eigenvalues. The remaining 6 eigenvalues were found, together with their eigenvectors, and are exhibited in Table 1.
It is a general fact that the characteristic polynomial of a symplectic matrix is reflexive, and therefore the eigenvalues come in one of the three possibilities :
* $`\lambda _{1,2}`$ are Real $`\lambda _1\lambda _2=1`$
* $`\lambda _{1,2}`$ are Unitary $`\lambda _1=e^{i\theta }\lambda _2=e^{i\theta }\lambda _1\lambda _2=1`$
* when two pairs of the second kind collide on the unit circle, they can go off and form a quartic arrangement:
$`\lambda _1=\lambda e^{i\theta },\lambda _2=\lambda e^{i\theta },\lambda _1\lambda _2\lambda _3\lambda _4=1.`$
$`\lambda _3={\displaystyle \frac{1}{\lambda }}e^{i\theta },\lambda _4={\displaystyle \frac{1}{\lambda }}e^{i\theta },`$
The system is stable if and only if it has pairs of eigenvalues of the second kind ($`\beta `$) only. Eigenvalues outside of the unit circle ($`\alpha `$), ($`\gamma `$) correspond to eigenvectors that grow exponentially with time (Figure 1). Conversely, eigenvalues on the unit circle correspond to a pair of real vectors which are rotated into each other without deformation. The angle of rotation per period $`T_0`$ is equal to arg$`\lambda `$ and quoted in the table as well. The coinciding pairs of eigenvalues in the table are accidental due to symmetry, and we dont expect it to happen for general $`R_i`$.
The conclusion is that using a combination of analytical and numerical methods we were able to demonstrate stability of the specific trajectories of the equations of motion (3). We stress that in general, classical YM mechanics exhibits strong instability of periodic trajectories. In particular, the pure SU(2) system, eqns (2.2) was conjectured to have no stable periodic trajectories, and may even be an ergodic, k-mixing system.
## 5 Radiation
### 5.1 Ramond-Ramond one-form radiation
In this section we compute the $`C^{(1)}`$-field radiation from the rotating ellipsoidal membrane. D0-branes are electrically charged under this field, but as we will see, the dipole moment of the system vanishes and we have to look at quadrupole order to find the radiated wave.
The dynamics of the $`C^{(1)}`$-field is given by the action
$$S_{C^{(1)}}=\frac{1}{2\kappa ^2}d^{10}x\sqrt{g}\frac{1}{4}(F^{(2)})^2+T_0𝑑tTr\left(C_0+C_i\dot{X}^i\right)$$
(47)
Choosing the Lorenz gauge
$$_\mu C^\mu =0$$
(48)
we have from (47) the wave equation
$$_\lambda ^\lambda C_i=2\kappa ^2T_0Tr(\dot{X}^i)\delta (\stackrel{}{x})$$
(49)
In the region far from the source we have the plane wave solution
$$C_\mu (\stackrel{}{x},t)=\widehat{C}_\mu \mathrm{exp}(ik_\lambda x^\lambda )+\text{c.c.}$$
(50)
From (48) and (49) we get
$$k_\mu k^\mu =0,k^\mu \widehat{C}_\mu =0$$
(51)
The energy-momentum tensor is
$$2\kappa ^2T_{\mu \nu }=\frac{1}{2}F_\mu ^\lambda F_{\nu \lambda }\frac{1}{8}\eta _{\mu \nu }F^2$$
(52)
Averaging over a region much larger than $`\omega ^1`$ we get the energy-momentum tensor of the plane wave
$$T_{\mu \nu }=\frac{1}{2\kappa ^2}k_\mu k_\nu \widehat{C}^\lambda \widehat{C}_\lambda ^{}$$
(53)
We set $`k^0=\omega `$ and $`\stackrel{}{k}=\omega \stackrel{}{x}/r`$ in what follows in order to make the local approximation of the spherical wave as a plane wave. Thus, we are only considering the radiation contribution at the particular angular frequency $`\omega `$.
The source for $`C_i`$ in (49) is clearly zero, so there is not any dipole radiation for the rotating ellipsoidal membrane. Instead, we must analyze the differences in propagation times due to the fact that the object is extended in space, since this gives the quadrupole radiation. In order to derive the quadrupole source, we use electrodynamical analogy and consider instead $`N`$ particles of equal charge at positions $`\stackrel{}{x}_a`$ , $`a=1\mathrm{}N`$. For these, the wave equation is
$$_\lambda ^\lambda C_i=2\kappa ^2T_0\underset{a=1}{\overset{N}{}}\dot{x}_a^i\delta (\stackrel{}{x}\stackrel{}{x}_a)e^{i\omega t}$$
(54)
where the angular frequency $`\omega `$ is the considered oscillation frequency. Using the propagator (104), we get
$`C_i(\stackrel{}{x},t)`$ $`=`$ $`i{\displaystyle \frac{2\kappa ^2T_0}{105\mathrm{\Omega }_8}}{\displaystyle \frac{\omega ^3}{r^4}}{\displaystyle d^9\stackrel{}{x}^{}\underset{a=1}{\overset{N}{}}\dot{x}_a^i\delta (\stackrel{}{x}^{}\stackrel{}{x}_a)e^{i\omega |\stackrel{}{x}\stackrel{}{x_a}|}e^{i\omega t}}+\text{c.c.}`$ (55)
$`=`$ $`i{\displaystyle \frac{2\kappa ^2T_0}{105\mathrm{\Omega }_8}}{\displaystyle \frac{\omega ^3}{r^4}}{\displaystyle \underset{a=1}{\overset{N}{}}}\dot{x}_a^ie^{i\omega r}e^{i\omega t}(1ik_jx_a^j)+\text{c.c.}`$
Making the identification $`_{a=1}^N\dot{x}_a^ix_a^jTr(\dot{X}^iX^j)`$ we get
$$C_i(\stackrel{}{x},t)=i\frac{2\kappa ^2T_0}{105\mathrm{\Omega }_8}\frac{\omega ^3}{r^4}e^{i\omega r}e^{i\omega t}\left(Tr(\dot{X}^i)ik_jTr(\dot{X}^iX^j)\right)+\text{c.c.}$$
(56)
In general $`Tr(\dot{X}^iX^j)`$ has both a symmetric part $`Q^{ij}=\frac{1}{2}(Tr(\dot{X}^iX^j)+Tr(\dot{X}^jX^i))`$ and an antisymmetric part $`\frac{1}{2}(Tr(\dot{X}^iX^j)Tr(\dot{X}^jX^i))`$. The symmetric part corresponds to electric quadrupole radiation and the antisymmetric part to magnetic dipole radiation. The antisymmetric part is proportional to the angular momentum tensor $`M^{ij}`$ which is a constant of motion for our particular solution. Thus, for our solution the antisymmetric part does not contribute to the radiation, and we shall therefore set the antisymmetric part to zero in the following. Using (50) and setting $`Tr(\dot{X}^i)=0`$ we finally have
$$\widehat{C}_i=i\frac{2\kappa ^2T_0}{105\mathrm{\Omega }_8}\frac{\omega ^4}{r^4}\widehat{k}_jQ^{ij}$$
(57)
where $`\widehat{k}=\stackrel{}{k}/\omega `$. From (51) we have that $`\omega \widehat{C}_0+k^i\widehat{C}_i=0`$, from which we get
$$\widehat{C}_0=\widehat{k}^i\widehat{C}_i=i\frac{2\kappa ^2T_0}{105\mathrm{\Omega }_8}\frac{\omega ^4}{r^4}\widehat{k}_i\widehat{k}_jQ^{ij}$$
(58)
Therefore, from (57), (58) and (53) we get that the power radiated per solid angle is
$`{\displaystyle \frac{dP}{d\mathrm{\Omega }_8}}`$ $`=`$ $`r^8\widehat{k}_iT^{0i}={\displaystyle \frac{1}{2\kappa ^2}}\omega ^2r^8\widehat{C}^\lambda \widehat{C}_\lambda ^{}`$ (59)
$`=`$ $`{\displaystyle \frac{2\kappa ^2T_0^2}{105^2\mathrm{\Omega }_8^2}}\omega ^{10}\left[\delta _{il}\widehat{k}_j\widehat{k}_m\widehat{k}_i\widehat{k}_j\widehat{k}_l\widehat{k}_m\right]Q^{ij}Q^{lm}`$
Integrating (59) over the $`S^8`$ sphere, we get the total radiated power
$$P=\frac{1}{3^25^27^211}\frac{2\kappa ^2T_0^2}{\mathrm{\Omega }_8}\omega ^{10}\left[Q^{ij}Q_{ij}^{}\frac{1}{9}|Q_i^i|^2\right]$$
(60)
Note the similarity between the expressions (59) and (60) and the corresponding expressions (108) and (111) for the gravitational quadrupole radiation.
We now need to compute the quadrupole moment $`Q^{ij}`$ for the rotating ellipsoidal membrane. One can check that it splits into a block-diagonal form with three two-by-two matrices corresponding to the three coordinate pairs 12, 34 and 56, with all other entries in $`Q^{ij}`$ being zero. The 12 block is given by
$$\left(\begin{array}{cc}\hfill Q_{11}& Q_{12}\hfill \\ \hfill Q_{21}& Q_{22}\hfill \end{array}\right)=\frac{1}{6}N\omega _1R_1^2\left(\begin{array}{cc}\hfill \mathrm{sin}2\omega _1t& \mathrm{cos}2\omega _1t\hfill \\ \hfill \mathrm{cos}2\omega _1t& \mathrm{sin}2\omega _1t\hfill \end{array}\right)$$
(61)
The 34 and 56 blocks are similarly given in terms of the two other frequencies and radii. One can see from (61) that the quadrupole radiation has double the frequency of the rotation frequency. Thus, the rotating ellipsoidal membrane emits quadrupole radiation at frequencies $`2\omega _1`$, $`2\omega _2`$ and $`2\omega _3`$.
Considering the fourier component at frequency $`2\omega _1`$, we get that the non-zero components of $`Q^{ij}`$ are
$$\left(\begin{array}{cc}\hfill Q_{11}& Q_{12}\hfill \\ \hfill Q_{21}& Q_{22}\hfill \end{array}\right)=\frac{1}{12}N\omega _1R_1^2\left(\begin{array}{cc}\hfill i& \hfill 1\\ \hfill 1& \hfill i\end{array}\right)$$
(62)
From formula (60) we then get
$$P_1=\frac{2^8}{3^45^27^211}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}\omega _1^{12}R_1^4$$
(63)
The total power emitted via quadrupole radiation in the $`C^{(1)}`$-field is therefore
$$P=\frac{2^8}{3^45^27^211}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}\left(\omega _1^{12}R_1^4+\omega _2^{12}R_2^4+\omega _3^{12}R_3^4\right)$$
(64)
As one can see from (64) the total power has the dependence $`\omega ^{12}R^4`$ on the frequencies and radii. Instead, electric quadrupole radiation in 3+1 dimensions has the dependence $`\omega ^6R^4`$. But, these two cases are in perfect correspondence with each other since the propagator (104) in 9+1 dimensions has an $`\omega ^3`$ dependence in the leading $`1/r^4`$ term. This gives an extra $`\omega ^6`$ for all types of radiation when comparing with 3+1 dimensions. For example, scalar radiation has the dependence $`\omega ^2`$ on the frequency in 3+1 dimensions and $`\omega ^8`$ in 9+1 dimensions. That radiation in 9+1 is suppressed by the factor $`\omega ^6`$ in comparison with 3+1 dimensions, can be understood as the fact that there are 6 more directions available for a particle to propagate in.
### 5.2 Ramond-Ramond three-form radiation
We would like to compute the radiation due to the coupling of the membrane to the $`C^{(3)}`$ RR potential. At this point we make use of the couplings we derived in Section 2.1. The dynamics of the $`C^{(3)}`$-field is governed by the action
$`S_{C^{(3)}}`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa ^2}}{\displaystyle d^{10}x\sqrt{g}\frac{1}{24!}F_{\lambda \mu \nu \rho }F^{\lambda \mu \nu \rho }}`$ (65)
$`+T_0{\displaystyle \frac{1}{2\pi l_s^2}}{\displaystyle 𝑑tTr\left(C_{0ij}[X^i,X^j]+C_{ijk}[X^i,X^j]\dot{X}^k\right)}`$
where
$$F_{\lambda \mu \nu \rho }^{(4)}=4_{[\lambda }^{}C_{\mu \nu \rho ]}^{(3)}.$$
(66)
We denote the current by $`J^{ijk}`$ with a convenient normalization
$$J^{ijk}=\frac{1}{2\pi l_s^2}Tr\left([X^{[i},X^j]\dot{X}^{k]}\right),$$
(67)
where we now have to impose anti-symmetrization with respect to $`ijk`$ indices.
We should be working in Lorenz gauge, in order to ensure the explicit transversality of the radiated wave:
$$^\mu C_{\mu \nu \rho }=0$$
(68)
This gives the wave equation
$$_\lambda ^\lambda C_{ijk}=2\kappa ^2T_0J^{ijk}\delta (\stackrel{}{x})$$
(69)
In the region far from the source, we have the plane wave solution
$$C_{\mu \nu \rho }(\stackrel{}{x},t)=\widehat{C}_{\mu \nu \rho }\mathrm{exp}(ik_\lambda x^\lambda )+\widehat{C}_{\mu \nu \rho }^{}\mathrm{exp}(ik_\lambda x^\lambda )$$
(70)
with the conditions
$$k_\mu k^\mu =0,k^\mu \widehat{C}_{\mu \nu \rho }=0$$
(71)
In order to compute the energy carried away by this wave we will need the energy-momentum tensor:
$$2\kappa ^2T_{\mu \nu }=\frac{1}{2}\frac{1}{3!}F_\mu ^{\lambda \rho \sigma }F_{\nu \lambda \rho \sigma }\frac{1}{2}g_{\mu \nu }\frac{1}{24!}F^2$$
(72)
We now set $`k^0=\omega `$ and $`\stackrel{}{k}=\omega \stackrel{}{x}/r`$. We see from (71) that fields that contain time-components can be found by using the gauge condition $`\widehat{C}_{0jk}=\widehat{k}_i\widehat{C}_{ijk}`$. The field strength is gotten by appropriate differentiation of the plane wave solution (70), because in the radiation zone the leading order derivative always comes from the exponential:
$`F_{ijkl}`$ $`=`$ $`4_{[i}C_{jkl]}=i\omega \left[\widehat{k}_i\widehat{C}_{jkl}\widehat{k}_j\widehat{C}_{ikl}\widehat{k}_k\widehat{C}_{jli}\widehat{k}_l\widehat{C}_{jki}\right]e^{i\omega (rt)}+\text{c.c.}`$
$`F_{0jkl}`$ $`=`$ $`i\omega \left[\widehat{C}_{jkl}\widehat{k}_m\widehat{k}_l\widehat{C}_{jkm}\widehat{k}_m\widehat{k}_k\widehat{C}_{ljm}\widehat{k}_m\widehat{k}_j\widehat{C}_{klm}\right]e^{i\omega (rt)}+\text{c.c.}`$ (73)
After averaging the energy-momentum tensor over a space-time region of size $`\omega ^1`$
$$2\kappa ^2T_{\mu \nu }=k_\mu k_\nu \frac{1}{3!}\widehat{C}^{\mu \nu \rho }\widehat{C}_{\mu \nu \rho }^{}=k_\mu k_\nu \frac{1}{3!}\left[\widehat{C}_{ijk}\widehat{C}_{ijk}^{}3\widehat{k}^l\widehat{C}_{ljk}\widehat{k}^m\widehat{C}_{mjk}^{}\right]$$
(74)
Note that this form of $`T_{\mu \nu }`$ was to be expected since the only Lorentz invariant scalar one can construct from $`\widehat{C}^{\mu \nu \rho }`$ is $`\widehat{C}^{\mu \nu \rho }\widehat{C}_{\mu \nu \rho }^{}`$. The $`F^2`$ term in (72) vanishes just like in ordinary Maxwell plane wave, as can be ascertained by direct contraction from (5.2).
The propagator in $`9+1`$ dimensions, including the normalization was derived in the Appendix A, and gives the following form for the C-field in the radiation zone:
$$C_{ijk}(\stackrel{}{r},t)=2\kappa ^2T_0\frac{i\omega ^3}{105\mathrm{\Omega }_8}\frac{J_{ijk}}{r^4}e^{i\omega (rt)}+\text{c.c.},$$
(75)
from which we read off the polarization $`\widehat{C}_{ijk}`$
$$\widehat{C}_{ijk}=2\kappa ^2T_0\frac{i\omega ^3}{105\mathrm{\Omega }_8}\frac{J_{ijk}}{r^4}$$
(76)
We are now ready to compute the energy flux in the out direction
$$\frac{dP}{d\mathrm{\Omega }_8}=2\kappa ^2T_{0i}\widehat{k}_i=\frac{2\kappa ^2T_0}{105^2\mathrm{\Omega }_8^2}\omega ^8\frac{1}{3!}\left[J^{ijk}J_{ijk}^{}3\widehat{k}_l\widehat{k^m}J^{lij}J_{mij}^{}\right]$$
(77)
For our particular system the non-zero components of Q are
$`Q_{135}={\displaystyle \frac{1}{3}}N\alpha R_1R_2R_3\mathrm{cos}\omega _1t\mathrm{cos}\omega _2t\mathrm{cos}\omega _3t`$
$`Q_{246}={\displaystyle \frac{1}{3}}N\alpha R_1R_2R_3\mathrm{sin}\omega _1t\mathrm{sin}\omega _2t\mathrm{sin}\omega _3t`$ (78)
together with $`Q_{146},Q_{136},Q_{235},Q_{236},Q_{145},Q_{245}`$. We see that it is possible to produce higher harmonics by multiplying the above trigonometrics. For generic values of $`\omega _i`$ one has four different frequencies:
$$\omega _1+\omega _2+\omega _3,\omega _1+\omega _2\omega _3,\omega _1\omega _2+\omega _3,\mathrm{and}\omega _1+\omega _2+\omega _3.$$
The computations are simplest for the highest mode, and in any case this is the one that carries the most power. We pick out this mode $`\omega =\omega _1+\omega _2+\omega _3`$, and also set $`R^3=R_1R_2R_3`$
$`Q_{135}={\displaystyle \frac{1}{24}}N\alpha R^3e^{i\omega t},`$ $`Q_{235}=Q_{145}=Q_{136}={\displaystyle \frac{i}{24}}N\alpha R^3e^{i\omega t}`$
$`Q_{246}={\displaystyle \frac{i}{24}}N\alpha R^3e^{i\omega t},`$ $`Q_{245}=Q_{146}=Q_{236}={\displaystyle \frac{1}{24}}N\alpha R^3e^{i\omega t}`$ (79)
From these it is easy to obtain the corresponding $`J`$’s by differentiation with respect to time. Substituting the fields into (77) we finally get the power radiated into unit solid angle as a function of direction, here parametrized by direction vector $`\widehat{k}`$
$`{\displaystyle \frac{dP}{d\mathrm{\Omega }_8}}={\displaystyle \frac{2\kappa ^2T_0^2}{105^2\mathrm{\Omega }_8^2}}\omega ^{10}N^2\alpha ^2R^6{\displaystyle \frac{1}{8}}\left[1{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{6}{}}}\widehat{k}_i^2\right]`$ (80)
This dependence on the direction is typical of rotating sources. The polarization of the wave (5.2) is such that when looking at the wave in a completely transverse direction (the 789 hyperplane), the wave is “circularly” polarized. In the 123456 hyperplane, the wave is “linearly” polarized and has half the power. In fact, we can rewrite the directional dependence in (80) in terms of the angle $`\widehat{\theta }`$ between $`\widehat{𝐤}`$ and the 789 hyperplane. The result is simply $`[1\mathrm{sin}^2\widehat{\theta }/2]`$. So indeed this is similar to the elliptically polarized field produced by a rotating, rather than oscillating dipole.
The total power emitted at the frequency $`\omega =\omega _1+\omega _2+\omega _3`$ is, after integrating over $`S^8`$
$$P=\frac{1}{12}\frac{2\kappa ^2T_0^2}{105^2\mathrm{\Omega }_8}\omega ^{10}N^2\alpha ^2R^6$$
(81)
### 5.3 Gravitational radiation
In this section we compute the power radiated from the rotating ellipsoidal membrane in the form of gravitational field. In the Appendix A we derived the formula (111) giving the radiated power for a given energy-momentum tensor, in terms of its fourier components. For the rotating ellipsoidal membrane the energy momentum tensor is
$$T_{\mu \nu }=T_0(2\pi l_s^2)^2Tr\left(F_\mu ^\lambda F_{\nu \lambda }\frac{1}{4}\eta _{\mu \nu }F^2\right)$$
(82)
We quote the $`T_{00}`$ component first, since it is the only one that has an interesting time-independent piece, namely the total energy
$$T_{00}=\frac{1}{4}NT_0\underset{i=1}{}\omega _i^2R_i^2$$
(83)
The spatial non-zero components are split into three blocks: 12, 34 and 56. Each one is oscillating at its own frequency, and since we are interested in the fourier decomposition of $`T_{\mu \nu }`$, we can start with the only block that contributes to radiation at frequency $`2\omega _1`$
$$\left(\begin{array}{c}T_{11}T_{12}\\ T_{21}T_{22}\end{array}\right)=\frac{NT_0}{6}R_1^2\omega _1^2\left(\begin{array}{c}\mathrm{cos}2\omega _1t\mathrm{sin}2\omega _1t\\ \mathrm{sin}2\omega _1t\mathrm{cos}2\omega _1t\end{array}\right)$$
(84)
Note that the energy-momentum tensor naturally has twice the frequency of the underlying object. This is a general feature related to the fact that there is only one kind gravitational charge.
Other blocks are of the same form, so for non-coincident frequencies only one block corresponds to each frequency: the 34 block to frequency $`2\omega _2`$ and 56 block to $`2\omega _3`$. Note also that it is traceless, and so the power will be given by the first term alone in the formula (111) which we rewrite for convenience below:
$$P=\frac{2\kappa ^2\omega ^8}{3^25711\mathrm{\Omega }_8}\left(T^{ij}T_{ij}\frac{1}{9}|T_i^i|^2\right)$$
(85)
Finally, the total power emitted at frequency $`2\omega _1`$ is
$$P_1=\frac{2\kappa ^2(2\omega _1)^8}{3^25711\mathrm{\Omega }_8}\frac{1}{9}N^2T_0^2\omega _1^4R_1^4$$
(86)
The exact same formula holds for radiation at the other frequencies. There can be no interference effect between different frequencies, thus the total emitted power is given by summing the conributions
$$P=\frac{2^6}{3^25711}\frac{2\kappa ^2N^2T_0^2}{\mathrm{\Omega }_8}(\omega _1^{12}R_1^4+\omega _2^{12}R_2^4+\omega _3^{12}R_3^4)$$
(87)
We note that each term is of the form $`P\kappa ^2\omega ^8(I\omega ^2)^2`$ where the moment of inertia $`INT_0R^2`$. Overall, the radiation is extremely suppressed by twelve powers of the frequency. In Appendix A we explain that this differs from $`3+1`$ dimensional power of six, solely because of extra six powers coming from the propagator in $`9+1`$ dimensions.
## 6 Discussion
We have shown that our rotating ellipsoidal membrane solution is classically stable. However the system constantly loses energy due to the quasi-classical radiation of various supergravity waves. This immediately poses the question of what will happen to the rotating ellipsoidal membrane after all available kinetic energy has been radiated. There are two possibilities: Either it will decay into $`N`$ free D0-branes or it will eventually become completely stable. We stress that the energy is positive, and so existence of a ground state is not obvious. If the system is to be quantum-mechanically stable, the size must be of order $`l_s`$, since at larger length-scales the classical approximation is valid and it does radiate energy. Therefore, if indeed this scenario is actually realized, there would be a particle-like ground state with positive energy.
In order to clarify what we mean by this fundamental particle ground state, let us employ the analogy with the archetypical example of quantum mechanics: the hydrogen atom. In this paper we found a rotating solution, similar to the motion of the electron in the central electrostatic field, however the electron should lose energy and emit photons, eventually falling onto the nucleus. This is the paradox which was resolved by the Bohr-Sommerfeld quantization of the atom. In this paper we computed a similar instability due to emission of various supergravity waves.
In the final quantum theory, the electron indeed emits quanta of light, corresponding to transitions between discrete energy levels. But there exists a ground state which is completely stable.
The Bohr-Sommerfeld quantization scheme tells us that the angular momentum should be quantized in units of the Planck constant. We thus have
$`M_{12}={\displaystyle \frac{1}{3}}NT_0\omega _1R_1^2=L_1,`$
$`M_{34}={\displaystyle \frac{1}{3}}NT_0\omega _2R_2^2=L_2,`$
$`M_{56}={\displaystyle \frac{1}{3}}NT_0\omega _3R_3^2=L_3,`$ (88)
where $`L_i`$ are integer values of the projection of the angular momentum. Roughly speaking, for $`L=L_1=L_2=L_3`$, the allowed frequency spectrum is
$$\omega _L^3=2\pi L\frac{64}{N^21}\frac{1}{NT_0(2\pi l_s^2)^2},$$
(89)
so that the leading behaviour is
$$\omega _L\sqrt[3]{g_sL}\frac{1}{Nl_s},$$
(90)
from this the quantized radius and energy are
$$R_L\sqrt[3]{g_sL}l_s\text{and}EL\omega L\sqrt[3]{g_sL}\frac{1}{Nl_s}$$
(91)
One question that we might ask is whether the angular momentum of the hypothetical ground state would be zero. The likely answer, by analogy with hydrogen atom, is positive.
## 7 Acknowledgments
We thank J. Ambjørn, E. Cheung, J. Correia, P. Di Vecchia, N. Obers, J. L. Petersen, G. K. Savvidy, W. Taylor and Z. Yin for useful discussions. KS would like to thank everyone at NBI for warm welcome and excellent working environment.
## Appendix A Gravitational radiation in 9+1 dimensions
We consider the region far away from the source. We have the metric
$$g_{\mu \nu }=\eta _{\mu \nu }+h_{\mu \nu }$$
(92)
where $`\eta _{\mu \nu }=\text{diag}(1,+1,+1,\mathrm{},+1)`$ and $`|h_{\mu \nu }|1`$. Indices are always lowered and raised with $`\eta _{\mu \nu }`$ and $`\eta ^{\mu \nu }`$. The Einstein equations are
$$R_{\mu \nu }^{(1)}\frac{1}{2}\eta _{\mu \nu }\eta ^{\rho \sigma }R_{\rho \sigma }^{(1)}=T_{\mu \nu }+t_{\mu \nu }$$
(93)
where the left-hand side are first order in $`h_{\mu \nu }`$ and $`t_{\mu \nu }`$ is the energy-momentum tensor of the gravitational field. The linearized Einstein equations are
$$R_{\mu \nu }^{(1)}\frac{1}{2}\eta _{\mu \nu }\eta ^{\rho \sigma }R_{\rho \sigma }^{(1)}=T_{\mu \nu }$$
(94)
Choosing the harmonic gauge
$$^\mu h_{\mu \nu }=\frac{1}{2}_\nu h_\lambda ^\lambda $$
(95)
We have from (94) the linearized Einstein equations
$$\frac{1}{2}_\lambda ^\lambda h_{\mu \nu }=2\kappa ^2\left(T_{\mu \nu }\frac{1}{8}\eta _{\mu \nu }T_\lambda ^\lambda \right)$$
(96)
Away from the source it is simply the wave equation
$$_\lambda ^\lambda h_{\mu \nu }=0$$
(97)
We consider a plane gravitational wave
$$h_{\mu \nu }=e_{\mu \nu }\mathrm{exp}(ik_\lambda x^\lambda )+e_{\mu \nu }^{}\mathrm{exp}(ik_\lambda x^\lambda )$$
(98)
From EOM (97) and the gauge condition (95) we get
$$k_\mu k^\mu =0,k^\mu e_{\mu \nu }=\frac{1}{2}k_\nu e_\lambda ^\lambda $$
(99)
We also have that $`e_{\mu \nu }=e_{\nu \mu }`$. Using the additional residual gauge freedom one can see that there are 35 independent components of $`e_{\mu \nu }`$. To second order in $`h_{\mu \nu }`$ the energy-momentum tensor of the gravitional field is
$$\kappa ^2t_{\mu \nu }=R_{\mu \nu }^{(2)}\frac{1}{2}\eta _{\mu \nu }\eta ^{\rho \sigma }R_{\rho \sigma }^{(2)}$$
(100)
where we used that $`R_{\mu \nu }^{(1)}=0`$ from (97). Averaging over a region much larger than $`|k|^1`$ we compute
$$<R_{\mu \nu }^{(2)}>=\frac{1}{2}k_\mu k_\nu \left(e^{\lambda \rho }e_{\lambda \rho }\frac{1}{2}|e_\lambda ^\lambda |^2\right)$$
(101)
Using this with (99) and (100) we get the energy-momentum tensor of the gravitational wave as
$$<t_{\mu \nu }>=\frac{1}{2\kappa ^2}k_\mu k_\nu \left(e^{\lambda \rho }e_{\lambda \rho }\frac{1}{2}|e_\lambda ^\lambda |^2\right)$$
(102)
In a 9+1 dimensional Minkowski space the solution to the wave equation with an oscillating $`\delta `$-function source
$$_\lambda ^\lambda f(r,t)=\delta (\stackrel{}{x})e^{i\omega t}$$
(103)
is
$$f(r,t)=i\frac{\omega ^3}{105\mathrm{\Omega }_8}\frac{1}{r^4}e^{i\omega r}\left(1+\frac{6i}{\omega r}\frac{15}{(\omega r)^2}\frac{15i}{(\omega r)^3}\right)e^{i\omega t}$$
(104)
when demanding that only the outgoing wave is present in the solution<sup>6</sup><sup>6</sup>6The solution in a $`d+1`$ dimensional space-time is $`f(r,t)r^{(d2)/2}H_{(d2)/2}^{(1)}(\omega r)e^{i\omega t}`$, where $`H_n^{(1)}`$ is the Henkel function of the first kind of order $`n`$.. Using this, we can solve (96) for a single frequency $`\omega `$ as
$`h_{\mu \nu }(\stackrel{}{x},t)`$ $`=`$ $`{\displaystyle \frac{2i}{105\mathrm{\Omega }_8}}{\displaystyle \frac{\omega ^3}{r^4}}e^{i\omega r}e^{i\omega t}{\displaystyle d^9\stackrel{}{x}^{}\left(T_{\mu \nu }(\stackrel{}{x}^{},\omega )\frac{1}{8}\eta _{\mu \nu }T_\lambda ^\lambda (\stackrel{}{x}^{},\omega )\right)e^{i\omega \stackrel{}{x}\stackrel{}{x}^{}/r}}`$ (105)
$`+\text{c.c.}`$
for $`r`$ large. Setting $`k^0=\omega `$ and $`\stackrel{}{k}=\omega \stackrel{}{x}/r`$ we have from (98) that
$$e_{\mu \nu }=\frac{4i\kappa ^2}{105\mathrm{\Omega }_8}\frac{\omega ^3}{r^4}\left(T_{\mu \nu }(\stackrel{}{k},\omega )\frac{1}{8}\eta _{\mu \nu }T_\lambda ^\lambda (\stackrel{}{k},\omega )\right)$$
(106)
with
$$T_{\mu \nu }(\stackrel{}{k},\omega )=d^9\stackrel{}{x}^{}T_{\mu \nu }(\stackrel{}{x}^{},\omega )e^{i\stackrel{}{k}\stackrel{}{x}^{}}$$
(107)
Thus, from (102) and (106) we get that the power emitted per solid angle is
$`{\displaystyle \frac{dP}{d\mathrm{\Omega }_8}}`$ $`=`$ $`r^8\widehat{k}_i<t^{0i}>={\displaystyle \frac{1}{2\kappa ^2}}\omega ^2r^8\left(e^{\lambda \rho }e_{\lambda \rho }{\displaystyle \frac{1}{2}}|e_\lambda ^\lambda |^2\right)`$ (108)
$`=`$ $`{\displaystyle \frac{8\kappa ^2}{105^2\mathrm{\Omega }_8^2}}\omega ^8\left(T^{\mu \nu }T_{\mu \nu }{\displaystyle \frac{1}{8}}|T_\lambda ^\lambda |^2\right)`$
$`=`$ $`{\displaystyle \frac{8\kappa ^2}{105^2\mathrm{\Omega }_8^2}}\omega ^8\mathrm{\Lambda }_{ijlm}(\widehat{k})T^{ij}T^{lm}`$
where $`\widehat{k}=\stackrel{}{k}/\omega `$ and
$$\mathrm{\Lambda }_{ijlm}(\widehat{k})=\delta _{il}\delta _{jm}2\widehat{k}_j\widehat{k}_m\delta _{il}+\frac{7}{8}\widehat{k}_i\widehat{k}_j\widehat{k}_l\widehat{k}_m\frac{1}{8}\delta _{ij}\delta _{lm}+\frac{1}{8}\widehat{k}_l\widehat{k}_m\delta _{ij}+\frac{1}{8}\widehat{k}_i\widehat{k}_j\delta _{lm}$$
(109)
After averaging the above over the 8-sphere
$$\frac{1}{\mathrm{\Omega }_8}𝑑\mathrm{\Omega }_8\mathrm{\Lambda }_{ijlm}(\widehat{k})=\frac{7}{8911}\left[89\delta _{il}\delta _{jm}10\delta _{ij}\delta _{lm}+\delta _{im}\delta _{jl}\right]$$
(110)
we finally get that the power emitted at the single frequency $`\omega `$ is
$$P=\frac{2\kappa ^2\omega ^8}{3465\mathrm{\Omega }_8}\left(T^{ij}T_{ij}\frac{1}{9}|T_i^i|^2\right)$$
(111)
Here we would like to remark on the overall dependence of the radiated power on frequency $`\omega `$. The spatial components of the energy-momentum tensor above (102) are typically proportional to $`I\omega ^2`$, and so power scales as $`P_{grav}^{(10)}\omega ^{12}`$. This is in contrast to $`3+1`$ dimensions where the corresponding quantity scales as $`P_{grav}^{(4)}\omega ^6`$. However this is more of an artifact of the propagator (104) as even $`l=0`$ scalar wave is radiated in 10 dimensions as $`P_{scalar}^{(10)}\omega ^8`$, whereas in $`3+1`$dim it’s $`P_{scalar}^{(4)}\omega ^2`$. Thus the extra powers of $`\omega `$ in (111) are completely due to the $`\omega ^3`$ behavior in the propagator (104).
|
warning/0002/hep-ph0002236.html
|
ar5iv
|
text
|
# Double inflation in supergravity and the primordial black hole formation
## I Introduction
In the framework of supergravity the reheating temperature of inflation should be low enough to avoid overproduction of gravitinos . The new inflation model generally predicts a very low reheating temperature and hence it is the most attractive among many inflation models . However, the new inflation suffers from a fine-tuning problem about the initial condition; i.e., for a successful new inflation, the initial value of the inflaton should be very close to the local maximum of the potential in a large region whose size is much longer than the horizon of the universe.
A framework of a double inflation was proposed to solve the initial value problem of the new inflation model<sup>*</sup><sup>*</sup>*Different models of double inflation were studied by various authors . . It was shown that the above serious problem is naturally solved by supergravity effects if there exists a preinflation (e.g., hybrid inflation ) with a sufficiently large Hubble parameter before the new inflation .
In this double inflation model, density fluctuations produced by both inflations are cosmologically relevant if the $`e`$-fold number of the new inflation is smaller than $`60`$ (the total $`e`$-fold number $`60`$ is required to solve flatness and horizon problems in the standard big bang cosmology ). In this case, the preinflation should account for the density fluctuations on large cosmological scales \[including the Cosmic Background Explorer (COBE) scales\] while the new inflation produces density fluctuations on smaller scales. Although the amplitude of the fluctuations on large scales should be normalized to the COBE data , fluctuations on small scales are free from the COBE normalization and can have arbitrary power matched to observations. In Ref., a cosmological implication of the double inflation for the large-scale structure formation was discussed. In this paper, we study primordial black hole (PBH) formation in the double inflation model. In Refs. , the production of black hole MACHO was investigated in the double inflation model for a special caseDifferent models for the PBH formation have been studied in Ref. .. Here, we consider a wide range of parameter space where PBHs are formedIn this paper we investigate the PBHs with mass $`10^{20}10^5M_{}`$. The upper bound of $`10^5M_{}`$ is not rigorous but much heavier black holes can be produced in the present model if we take the appropriate model parameters. However, the mass of the black holes should be less than the galactic mass $`10^{12}M_{}`$, otherwise the power spectrum conflicts the observations ( e.g. distribution of galaxies ). The lower bound $`10^{20}M_{}`$ comes from the requirement that $`e`$-fold number of the new inflation should be larger than $`0`$.. In particular, we show that the double inflation creates small PBHs evaporating now if those PBHs are produced during matter-dominated (MD) era, i.e., before the end of reheating process after the new inflation. We stress that these evaporating PBHs may account for antiproton fluxes observed by the BESS experiment .
Throughout this paper the gravitational scale ($`2.4\times 10^{18}`$ GeV) is taken to be unity.
## II Black hole formation
In a radiation-dominated (RD) universe, PBHs are formed if the density fluctuations $`\delta `$ at horizon crossing satisfy a condition $`1/3\delta 1`$ , where $`\delta `$ is the over density at the horizon scale. Masses of the black holes $`M`$ are roughly equal to the horizon mass,
$$M4\pi \sqrt{\frac{3}{\rho }}0.066M_{}\left(\frac{T}{\mathrm{GeV}}\right)^2\left(\frac{g_{}}{50}\right)^{1/2},$$
(1)
where $`\rho `$, $`T`$, and $`g_{}`$ are the total cosmic density, temperature, and statistical degrees of freedom at the horizon crossing, respectively.
The horizon length at the black hole formation epoch ($`T=T_{}`$) corresponds to the scale $`L_{}`$ in the present universe given by
$$L_{}\frac{a(T_0)}{a(T_{})}H^1(T_{})0.064\mathrm{pc}\left(\frac{T_{}}{\mathrm{GeV}}\right)^1\left(\frac{g_{}}{50}\right)^{1/6},$$
(2)
where $`T_0`$ is the temperature of the present universe. The comoving wave number corresponding to this length scale, $`k_{}2\pi /L_{}`$, is
$$k_{}1.0\times 10^8\mathrm{Mpc}^1\left(\frac{g_{}}{50}\right)^{1/6}\left(\frac{T_{}}{\mathrm{GeV}}\right).$$
(3)
Thus, we can write the PBH mass as a function of comoving wave number as
$$M_{}6.4\times 10^{14}M_{}\left(\frac{g_{}}{50}\right)^{1/6}\left(\frac{k_{}}{\mathrm{Mpc}^1}\right)^2.$$
(4)
The mass fraction $`\beta _{}(=\rho _{BH}/\rho )`$ of PBHs of mass $`M_{}`$ is given by
$$\beta _{}(M_{})=_{1/3}^1\frac{d\delta }{\sqrt{2\pi }\sigma (M_{})}\mathrm{exp}\left(\frac{\delta ^2}{2\sigma ^2(M_{})}\right)\sigma (M_{})\mathrm{exp}\left(\frac{1}{18\sigma ^2(M_{})}\right),$$
(5)
where $`\sigma (M_{})`$ is the mass variance at the horizon crossing. Notice that the mass fraction $`\beta _{}(M_{})`$ drops off sharply as $`\sigma (M)`$ decreases. The density of the black holes of mass $`M_{}`$, $`\rho _{BH}(M_{})`$, is given by
$$\frac{\rho _{BH}(M_{})}{s}\frac{3}{4}\beta _{}(M_{})T_{},$$
(6)
where $`s`$ is the entropy density. Since $`\rho _{BH}/s`$ is constant at $`T<T_{}`$, we estimate the density parameter $`\mathrm{\Omega }_{BH}(M_{})`$ of the black holes in the present universe as
$$\mathrm{\Omega }_{BH}(M_{})h^22.1\times 10^8\beta _{}(M_{})\left(\frac{T_{}}{\mathrm{GeV}}\right),$$
(7)
where $`h`$ is the present Hubble constant in units of 100 km/sec/Mpc. We write it as a function of PBH mass or PBH scale as
$$\mathrm{\Omega }_{BH}(M_{})h^25.4\times 10^7\beta _{}(M_{})\left(\frac{g_{}}{50}\right)^{1/4}\left(\frac{M_{}}{M_{}}\right)^{1/2},$$
(8)
or
$$\mathrm{\Omega }_{BH}(M_{})h^22.1\beta _{}(M_{})\left(\frac{g_{}}{50}\right)^{1/6}\left(\frac{k_{}}{\mathrm{Mpc}^1}\right).$$
(9)
As for the mass of the PBHs produced during the RD era, we have a lower limit $`M_R`$. The mass of PBH produced at the reheating epoch is given by \[see Eq.(1)\]
$$M_R0.066M_{}\left(\frac{T_R}{\mathrm{GeV}}\right)^2\left(\frac{g_{}}{50}\right)^{1/2},$$
(10)
where $`T_R`$ is the reheating temperature. As seen later $`T_R`$ is less than $`10^6`$GeV in our inflation model, and hence $`M_R`$ is larger than $`10^{13}M_{}`$. Thus, the PBHs lighter than $`M_R`$ should be produced during the MD era.
In a MD universe, a relation between the comoving scale $`L_{}`$ and the horizon mass $`M_{}`$ is
$$L_{}=L_R\left(\frac{M_{}}{M_R}\right)^{1/3},$$
(11)
where $`L_R`$ is the comoving scale of the horizon at the reheating epoch. Thus, the mass $`M_{}`$ of the PBH produced during the MD epoch is given by
$$M_{}6.3\times 10^{22}M_{}\left(\frac{T_R}{\mathrm{GeV}}\right)\left(\frac{k_{}}{\mathrm{Mpc}^1}\right)^3.$$
(12)
We see that small PBHs of mass, $`M_{}10^{19}M_{}`$ for example, would be produced for $`T_R10^6`$GeV and $`k_{}10^{16}`$Mpc<sup>-1</sup>. The condition for the PBH formation in a MD universe is discussed in Ref. , where the mass fraction of PBHs of mass $`M_{}`$ is estimated as
$$\beta _{}(M_{})2\times 10^2\sigma (M_{})^{13/2}.$$
(13)
Notice that the mass fraction has a weeker dependence on $`\sigma `$ than in the RD case \[see Eq.(5)\]
During the MD era, the mass fraction $`\beta _{}`$ stays constant and hence the density of the black holes of mass $`M_{}`$, $`\rho _{BH}(M_{})`$, is given by
$$\frac{\rho _{BH}(M_{})}{s}\frac{3}{4}\beta _{}(M_{})T_R.$$
(14)
We can write the present density parameter $`\mathrm{\Omega }_{BH}(M_{})`$ for the black holes of mass $`M_{}`$ as
$$\mathrm{\Omega }_{BH}(M_{})h^22.1\times 10^8\beta _{}(M_{})\left(\frac{T_R}{\mathrm{GeV}}\right).$$
(15)
## III Double Inflation Model
We adopt a double inflation model proposed in Ref.. The model consists of two inflationary stages; the first one is called preinflation and we take a hybrid inflation (see also Ref.) as the preinflation. We also assume that the second inflationary stage is realized by a new inflation model and its $`e`$-fold number is smaller than $`60`$. Thus, the density fluctuations on large scales are produced during the preinflation and their amplitude should be normalized to the COBE data . On the other hand, the new inflation produces fluctuations on small scales. Since the amplitude of small scale fluctuations is free from the COBE normalization, we expect that the new inflation can produce large density fluctuations enough to form PBHs. We choose the predicted power spectrum to be almost scale invariant ($`n_s1`$) on large cosmological scales which is favored for the structure formation of the universe . On the other hand, the new inflation gives the power spectrum which has large amplitude and shallow slope ($`n_s<1`$) on small scales. Thus, this power spectrum has a large and sharp peak on the scale corresponding to a turning epoch from the preinflation to the new inflation, and we expect that PBHs are produced at that scale.
As for the detailed argument of the dynamics of our model, see Refs..
### A Preinflation
First, let us briefly discuss a hybrid inflation model . The hybrid inflation model contains two kinds of superfields: one is $`S(x,\theta )`$ and the others are a pair of $`\mathrm{\Psi }(x,\theta )`$ and $`\overline{\mathrm{\Psi }}(x,\theta )`$. Here $`\theta `$ is the Grassmann number denoting superspace. The model is based on the U$`(1)_R`$ symmetry under which $`S(\theta )e^{2i\alpha }S(\theta e^{i\alpha })`$ and $`\mathrm{\Psi }(\theta )\overline{\mathrm{\Psi }}(\theta )\mathrm{\Psi }(\theta e^{i\alpha })\overline{\mathrm{\Psi }}(\theta e^{i\alpha })`$. The superpotential is given by
$$W(S,\mathrm{\Psi },\overline{\mathrm{\Psi }})=\mu ^2S+\lambda S\overline{\mathrm{\Psi }}\mathrm{\Psi }.$$
(16)
The $`R`$-invariant Kähler potential is given by
$$K(S,\mathrm{\Psi },\overline{\mathrm{\Psi }})=|S|^2+|\mathrm{\Psi }|^2+|\overline{\mathrm{\Psi }}|^2+\mathrm{},$$
(17)
where the ellipsis denotes higher-order terms which we neglect in the present analysis for simplicity. We gauge the U$`(1)`$ phase rotation:$`\mathrm{\Psi }e^{i\delta }\mathrm{\Psi }`$ and $`\overline{\mathrm{\Psi }}e^{i\delta }\overline{\mathrm{\Psi }}`$. To satisfy the $`D`$-term flatness condition we take always $`\mathrm{\Psi }=\overline{\mathrm{\Psi }}`$ in our analysis.
We define $`N_{\mathrm{COBE}}`$ as the $`e`$-fold number corresponding to the COBE scale and the COBE normalization leads to a condition for the inflaton potential,
$$\left|\frac{V^{3/2}}{V^{}}\right|_{N_{\mathrm{COBE}}}5.3\times 10^4,$$
(18)
where $`V`$ is the inflaton potential obtained from Eqs.(16) and (17). In the hybrid inflation model, density fluctuations are almost scale invariant;
$$n_{\mathrm{pre}}1+2\left(\frac{V^{\prime \prime }}{V}\right)3\left(\frac{V^{}}{V}\right)^2|_{N_{\mathrm{COBE}}}1\frac{1}{N_{\mathrm{COBE}}}1,$$
(19)
where $`n_{\mathrm{pre}}`$ is a spectral index for a power spectrum of density fluctuations.
### B New inflation
Now, we consider a new inflation model. We adopt an inflation model proposed in Ref. . The inflaton superfield $`\varphi (x,\theta )`$ is assumed to have an $`R`$ charge $`2/(n+1)`$ and U$`(1)_R`$ is dynamically broken down to a discrete $`Z_{2nR}`$ at a scale $`v`$, which generates an effective superpotential ,
$$W(\varphi )=v^2\varphi \frac{g}{n+1}\varphi ^{n+1}.$$
(20)
The $`R`$-invariant effective Kähler potential is given by
$$K(\varphi ,\chi )=|\varphi |^2+\frac{\kappa }{4}|\varphi |^4+\mathrm{},$$
(21)
where $`\kappa `$ is a constant of order $`1`$. We require that supersymmetry breaking effects make the potential energy at a vacuum vanish, and we have a relation between $`v`$ and the gravitino mass $`m_{3/2}`$ as (for details, see Ref. )
$$m_{3/2}\left(\frac{n}{n+1}\right)|v|^2\left|\frac{v^2}{g}\right|^{\frac{1}{n}}.$$
(22)
The inflaton $`\varphi (x)`$ (the scalar component of $`\varphi (x,\theta )`$) has a mass $`m_\varphi `$ in the vacuum with
$$m_\varphi n|g|^{1/n}|v|^{22/n}.$$
(23)
The inflaton $`\varphi `$ may decay into ordinary particles through gravitationally suppressed interactions, which yields reheating temperature $`T_R`$ given by<sup>§</sup><sup>§</sup>§ The decay rate of the inflaton $`\varphi `$ is discussed in Ref.
$`T_R0.1m_\varphi ^{3/2}`$ $``$ $`2.4\times 10^{17}\mathrm{GeV}n^{3/2}|g|^{3/2n}|v|^{3(11/n)}`$ (24)
$``$ $`10^6\mathrm{GeV}\mathrm{for}m_{3/2}1\mathrm{T}\mathrm{e}\mathrm{V},n3.`$ (25)
An important point on the above density fluctuations is that it results in a tilted spectrum with spectral index $`n_{\mathrm{new}}`$ given by (see Refs. )
$$n_{\mathrm{new}}12\kappa .$$
(26)
### C Initial value and fluctuations of the inflaton $`\phi `$
The crucial point observed in Ref. is that the preinflation sets dynamically the initial condition for the new inflation. We identify the inflaton field $`\phi (x)/\sqrt{2}`$ with the real part of the field $`\varphi (x)`$. It gets an effective mass $`m_{\mathrm{eff}}\mu ^2`$ during the preinflation . Thus, this inflaton $`\phi `$ tends to the potential minimum,
$$\phi _{\mathrm{min}}\frac{\sqrt{2}}{\sqrt{\lambda }}v\left(\frac{v}{\mu }\right).$$
(27)
Notice that $`\phi _{\mathrm{min}}`$ deviates from zero due to the presence of a linear term $`v^2\mu ^2S\phi `$ (see Ref. ). Thus, at the end of the preinflation the $`\phi `$ settles down to this $`\phi _{\mathrm{min}}`$.
After the preinflation, the universe becomes MD because of the oscillation of the inflaton for preinflation. During the MD era between the two inflations, the energy density scales as $`a^3`$, and the new inflaton oscillates around $`\phi =0`$ with its amplitude decreasing as $`a^{3/4}`$. Since the scale factor increases by a factor $`(\mu /v)^{4/3}`$ during this era, the mean initial value $`\phi _b`$ of $`\phi `$ at the beginning of the new inflation is written as
$$\phi _b\frac{\sqrt{2}}{\sqrt{\lambda }}v\left(\frac{v}{\mu }\right)^2.$$
(28)
Therefore, the amplitude of fluctuations with comoving wavelength corresponding to the horizon scale at the beginning of the new inflation is given by
$$\delta \phi \frac{H_{\mathrm{pre}}}{2\pi }\left(\frac{H_{\mathrm{pre}}}{m_{\mathrm{eff}}}\right)^{\frac{1}{2}}\left[\left(\frac{\mu }{v}\right)^{2/3}\right]^{3/2}\left[\left(\frac{\mu }{v}\right)^{4/3}\right]^{3/4}\frac{H_{\mathrm{pre}}}{3^{1/4}2\pi }\left(\frac{v}{\mu }\right)^2,$$
(29)
where $`H_{\mathrm{pre}}`$ is the Hubble parameter during the hybrid inflation, $`H_{\mathrm{pre}}^2\mu ^4/3`$. The fluctuations given by Eq. (29) are a little less than newly induced fluctuations at the beginning of the new inflation \[$`\delta \phi _{\mathrm{new}}v^2/(2\pi \sqrt{3}`$)\]. Moreover, the fluctuations produced during the preinflation are more suppressed for smaller wavelength. Thus, we assume that the fluctuations of $`\phi `$ induced in the preinflation are negligible compared with fluctuations produced by the new inflation. As mentioned before, the new inflation gives the tilted spectrum on small scales \[see Eq. (26)\] and hence the fluctuations at the scale corresponding to the beginning of the new inflation is dominant.
Now let us estimate $`e`$-fold number which corresponds to our current horizon. The $`e`$-fold number is given by
$$N_{\mathrm{tot}}=62\mathrm{ln}\frac{k}{a_0H_0}\mathrm{ln}\frac{10^{16}\mathrm{GeV}}{V_k^{1/4}}+\mathrm{ln}\frac{V_k^{1/4}}{V_{\mathrm{end}}^{1/4}}\frac{1}{3}\mathrm{ln}\frac{V_{\mathrm{end}}^{1/4}}{\rho _{\mathrm{reh}}^{1/4}},$$
(30)
where $`V_k`$ is a potential energy when a given scale $`k`$ leaves the horizon, $`V_{\mathrm{end}}`$ that when the inflation ends, and $`\rho _{\mathrm{reh}}`$ energy density at the time of reheating.
We take $`V_kV_{\mathrm{end}}`$, and $`\rho _{\mathrm{reh}}^{1/4}\mathrm{a}\mathrm{few}\times T_{\mathrm{reh}}`$. For $`k=a_0H_0`$ (i.e., the present horizon scale), we have
$$N_{\mathrm{tot}}67.1+\left(\frac{5}{3}\frac{1}{n}\right)\mathrm{ln}v+\frac{1}{2}\mathrm{ln}n+\frac{1}{2n}\mathrm{ln}g.$$
(31)
In estimating $`N_{\mathrm{COBE}}`$ we must take into account the fact that the fluctuations induced at $`e`$-fold number less than $`(2/3)\mathrm{ln}(\mu /v)`$ before the end of the hybrid inflation reenter the horizon before the new inflation starts. Such fluctuations are cosmologically irrelevant since the new inflation produce much larger fluctuations . Thus, $`N_{\mathrm{COBE}}`$ is given by
$`N_{\mathrm{COBE}}`$ $`=`$ $`N_{\mathrm{tot}}N_{\mathrm{new}}+{\displaystyle \frac{2}{3}}\mathrm{ln}{\displaystyle \frac{\mu }{v}}`$ (32)
$``$ $`67.1+\left({\displaystyle \frac{5}{3}}{\displaystyle \frac{1}{n}}\right)\mathrm{ln}v+{\displaystyle \frac{1}{2}}\mathrm{ln}n+{\displaystyle \frac{1}{2n}}\mathrm{ln}gN_{\mathrm{new}}+{\displaystyle \frac{2}{3}}\mathrm{ln}{\displaystyle \frac{\mu }{v}}.`$ (33)
The COBE normalization in Eq. (18) should be imposed by using this $`N_{\mathrm{COBE}}`$.
### D Numerical Results
We estimate density fluctuations in the double inflation by calculating evolution of $`\phi `$ and $`\sigma `$ numerically. For given parameters $`\kappa `$ and $`\lambda `$, we obtain the break scale $`k_b`$ and the amplitude of density fluctuations produced at the beginning of new inflation $`\delta _b`$. Here, $`k_b^1`$ is the comoving scale corresponding to the Hubble radius at the beginning of the new inflation (a turning epoch). We can understand the qualitative dependence of $`(k_b,\delta _b)`$ on $`(\kappa ,\lambda )`$ as follows: When $`\kappa `$ is large, the slope of the potential for the new inflation is too steep, and the new inflation cannot last for a long time. Therefore, the break occurs at smaller scales. As for $`\delta _b`$, we can see from Eq.(18) that as $`\lambda `$ gets larger, $`\mu `$ also gets large. In addition, we can show that
$$\delta _b\left(\frac{\delta \rho }{\rho }\right)_{\mathrm{new},k_b}\frac{\sqrt{\lambda }\mu ^2}{\kappa }\frac{\lambda ^{3/2}}{\kappa },$$
(34)
for a given $`v`$ (see Ref. ). Thus, we have larger $`\delta _b`$ for larger $`\lambda `$.
## IV Mass Variance and Density Fluctuations
Our double inflation model predicts the amplitude of density fluctuations $`\delta _b`$ as a function of inflation parameters. On the other hand, the black hole abundance is expressed as a function of mass variance $`\sigma `$ at the time when the corresponding scale enters the horizon. Therefore, when we compare the observations with the prediction of our model, we need a relation between the mass variance $`\sigma `$ and the fluctuations $`\delta _b`$.
For the power spectrum with the break scale $`k_b^1`$ which enters the horizon during the RD epoch, we have a relation between the mass variance and the amplitude of fluctuations asIn Refs. , an incorrect relation $`\delta _b\sigma _b/6`$ was used.
$$\delta _b\sigma _b/0.65,$$
(35)
(the numerical factor depends on the tilted spectral index $`n_s`$, and within the parameter range we consider, this factor lies between $`0.620.67`$.). For the power spectrum with the break scale $`k_b^1`$ which enters the horizon during the MD epoch, we have a relation between the mass variance and the amplitude of fluctuations as
$$\delta _b\sigma _b/2.3,$$
(36)
(again, the numerical factor depends on the tilted spectral index $`n_s`$, and within the parameter range we consider, this factor lies between $`2.12.8`$.).
First, let us consider PBH dark matter with $`\mathrm{\Omega }_{BH}1`$ which are produced during the RD epoch. Since the density fluctuations at the break scale is dominant, and the mass fraction $`\beta _{}`$ has a sharp peak at that scale, only the PBHs of mass corresponding to the break scale are formed. For PBHs produced during the RD epoch (after reheating process), we have, from Eq. (8),
$$\mathrm{\Omega }_{BH}h^25.4\times 10^7\beta _{}\left(\frac{g_{}}{50}\right)^{1/4}\left(\frac{M_{}}{M_{}}\right)^{1/2}.$$
(37)
For example, if we require that the black holes with mass $`M_{}`$ ($`=`$MACHOs) be dark matter in the present universe, i.e. $`\mathrm{\Omega }_{BH}h^20.25`$, we obtain $`\beta _{}5\times 10^9`$, and from Eq.(5) we obtain
$$\sigma (M_{})0.06.$$
(38)
From Eq.(35), we see that the break amplitude is $`\delta _b0.06/0.650.092`$, and we find from Eq.(4) that the break scale is $`2.5\times 10^7\mathrm{Mpc}^1`$. In Fig.1, we plot the numerical results of our double inflation model for $`n=4,g=1`$, and $`v=10^7`$. We see a wide range of parameter space which may account for DM ($`\mathrm{\Omega }_{BH}0.11`$).
Next, let us consider another interesting mass range of PBHs which are evaporating now ($`M_{\mathrm{evap}}3\times 10^{19}M_{}`$). The PBHs of such light mass are produced during the MD epoch and from Eq.(15) we find
$$\mathrm{\Omega }_{BH}(M_{})h^24.2\times 10^6\sigma (M_{})^{13/2}\left(\frac{T_R}{\mathrm{GeV}}\right).$$
(39)
It has been reported, recently, that the BESS experiment has observed antiproton fluxes, which may be explained by the evaporation of PBHs if $`\mathrm{\Omega }_{BH}h^22\times 10^9`$ . In order to explain the BESS result by evaporating PBHs we need
$$\sigma (M_{\mathrm{evap}})4.4\times 10^3\left(\frac{T_R}{\mathrm{GeV}}\right)^{2/13}.$$
(40)
From Eq.(24), we estimate the required fraction of the evaporating PBHs as
$$\sigma (M_{\mathrm{evap}})9.3\times 10^6n^{3/13}|g|^{3/13n}v^{6(11/n)/13}.$$
(41)
Since the mass variance $`\sigma (M)`$ scales as $`\sigma (M)M^{(1n_s)/6}`$ during the MD epoch, we obtain the mass variance at the break scale as
$$\sigma _b9.3\times 10^6n^{3/13}|g|^{3/13n}v^{6(11/n)/13}\left(\frac{M_b}{M_{\mathrm{evap}}}\right)^{(1n_s)/6},$$
(42)
and the amplitude $`\delta _b`$ is $`\delta _b\sigma _b/2.3`$ and $`n_s12\kappa `$. In Fig.2, we plot an example of the numerical results of our double inflation modelFor the case of $`n=4`$, we do not find a consistent parameter region with the BESS experiment. for $`n=3,g=10^4`$, and $`v=10^6`$. As shown in the figure, we have a set of inflation parameters $`(\kappa ,\lambda `$) which may account for the BESS experiment.
## V Conclusions and Discussions
In this paper we have studied the formation of PBHs by taking a double inflation model in supergravity. We have shown that in a wide range of parameter space PBHs are produced of various masses. These PBHs are interesting since, for example, they may be identified with MACHOs ($`MM_{}`$) in the halo of our galaxy. Or, they may be PBHs which are just evaporating now ($`M10^{19}M_{}`$). Such black holes are one of the interesting candidates for the sources of antiproton fluxes recently observed in the BESS detector .
The dark matter PBHs play a role of the cold dark matter on the large scale structure formation. The scales of the fluctuations for PBH formation themselves are much smaller than the galactic scale and thus we cannot see any signals for such fluctuations in $`\delta T/T`$ measurements. However, the PBHs may be a source of gravitational waves. If the PBHs dominate dark matter of the present universe, some of them likely form binaries. Such binary black holes coalesce and produce significant gravitational waves which may be observable in future detectors.
## Acknowledgment
T. K. is grateful to K. Sato for his continuous encouragement. A part of work is supported by Grant-in-Aid of the Ministry of Education and by Grant-in-Aid, Priority Area “Supersymmetry and Unified Theory of Elementary Particles” (#707).
|
warning/0002/gr-qc0002024.html
|
ar5iv
|
text
|
# Statistical mechanics of generally covariant quantum theories: A Boltzmann-like approach
## 1 Introduction
General relativity has modified our understanding of the physical world in depth and has altered some among the most fundamental notions we use to describe it. During the last ten years, the effort to understand the combined consequences of this conceptual revolution and quantum mechanics has lead to loop quantum gravity, a predictive quantum theory of the gravitational field, whose theoretical results can be, in principle, empirically tested . There are other areas in our understanding of nature, however, where the consequences of the general relativistic conceptual revolution have not been fully explored yet. Among these is statistical mechanics. To be precise, thermodynamics and statistical mechanics on a fixed curved spacetime have been much studied (see, for instance, ); but not much is known on the possibility of developing thermodynamics and statistical mechanics of a fully general covariant system, in particular, a system including the gravitational field.
Here, we begin to address this issue. Specifically, we study the following problem. Consider a simple physical system, $`s`$, with a finite number $`D`$ of degrees of freedom. Assume that $`s`$ is described by a fully constrained Hamiltonian system. That is, its dynamics is not given by a Hamiltonian, but rather by $`M`$ first class constraints. Physically, this means that we do not understand the dynamics of $`s`$ in terms of the evolution of $`D`$ dependent Lagrangian variables (or $`2D`$ phase space variables) as functions of a single preferred independent external time variable $`t`$; rather, we understand the dynamics as the relative evolution of $`2(D+M)`$ phase space variables with respect to one another – all the variables being on the same footing. The dynamics fixes relations between these variables, so that by knowing some of them we can predict the others. Such a simple system encodes an critical feature of general relativistic systems: the absence of a preferred time variable, and the relational aspect of evolution. Now, consider a macroscopic system $`S`$ composed by a large number of component systems, each one identical to $`s`$, and interacting weakly. Can we use statistical mechanics to describe macroscopic properties of $`S`$? Notice that there is no time variable in the description of $`S`$, therefore no notion of thermalization ‘in time’; there is also no notion of energy, and thus no obvious way to define a canonical or microcanonical ensemble. If we arbitrarily choose one variable in $`S`$ as the physical time (that is, if we ‘deparameterize’ the system), and then use conventional statistical techniques, our results are going to depend on the choice of time, and therefore to be possibly unphysical. Is there anything we can nevertheless say, about the macroscopic behavior of this system? Can we still apply thermodynamical or statistical mechanical techniques?
These questions are relevant in a strong-field gravitational context, whenever a preferred time and a conserved energy are not defined. Of course, if we consider a system with a notion of time and with conserved energy, we expect that temperature and energy will recover their traditional role. This is the case, for instance, of an asymptotically flat gravitational field; in this case the Hamiltonian is given by suitable boundary terms and the observables at infinity evolve in the Lorentz time of the asymptotic metric. The general theory we present here will have to yield standard results in this case. However, what about the situations in which there is no conserved energy and no preferred time? For instance, as far as we know, our universe might very well not be asymptotically flat. Alternatively, we may be interested in a system with a strong (dynamical) gravitational field, and have no access to an external asymptotic region. In particular, consider a “high temperature” early-universe regime. This is usually described in terms of fluctuations around a background metric; is there a genuinely general covariant description of this physics? And what is temperature in this context, if we do not fix a background metric? Certainly, it is difficult to even define what statistical mechanics is if we do not have some notion of energy conservation; but does this mean that in all gravitational systems in which there is no conserved energy (most of them!), we have to renounce using statistical methods?
These questions have not yet been addressed in the literature, as far as we know. An attempt to study certain aspects of the foundations of general covariant statistical theory is in Refs. . In these works, the question addressed is whether a preferred time flow, having the thermodynamical properties that we ascribe to physical time, can be derived from the statistical mechanics of a covariant system. The answer is positive, and the flow turns out to be dependent on the statistical state. The relation flow/state reflects a very general operator algebra structure (Tomita-Takesaki theorem), and raises intriguing physical issues, in particular in view of powerful mathematical uniqueness results about the flow (Connes’ Cocycle Radon-Nikodym theorem). Here, on the other hand, we are not concerned with the emergence of a time flow. Instead, we address directly the issue of a statistical description independent from any notion of time.
Furthermore, Refs. take Gibbs’ (and Einstein’s !) point of view on statistical mechanics: a statistical state is described by a distribution over the phase space $`\mathrm{\Gamma }`$ of the composite system $`S`$ (in Ehrenfest’s terminology, over the $`\mathrm{\Gamma }`$-space ). The state represents the distribution of $`S`$’s microstates over many imaginary copies of the system, all in the same macrostate. Here, on the contrary, we use Boltzmann’s original point of view : we assume that $`S`$ is composed by a large number of identical subsystems $`s`$. The statistical state is then described by a distribution over the phase space $`\gamma `$ of the component system $`s`$ (over the $`\mu `$-space, in Ehrenfest’s terminology). This gives, for each state of $`s`$, the expected number of component systems that are in that state.
Of course, we do not expect any of the well known subtleties and conceptual difficulties of statistical mechanics to be solved by applying it to covariant systems. Here we are not concerned with the old problems in the foundations of statistical mechanics, but only with the specific new problems –and new beauties– that emerge in trying to extend the general relativistic revolution to statistical physics.
Our main result is the following. We argue that, under appropriate conditions, the statistical mechanics of a system $`S`$ composed by many constrained systems $`s`$ is well defined. In particular, statistical mechanics is not necessarily tied to the concept of energy, or to a preferred time flow. Accordingly, general covariant statistical mechanics is not governed by the notion of temperature. Instead, intensive macroscopic parameters are determined by the properties of Boltzmann’s thermalizing interaction among the individual component systems. In the course of the paper, we develop the basis of covariant quantum statistical mechanics and define the intensive and extensive thermodynamical quantities.
We begin by recalling the properties and the physical interpretation of the parameterized systems in Section 2. We then give the main discussion on the foundations of covariant statistical mechanics in Section 3, and a simple example in Section 4. We discuss the statistical mechanics of a gas of free relativistic particles in Section 5, and we comment and summarize in Section 6.
## 2 Presymplectic systems
We consider fully constrained systems, with a finite number of degrees of freedom, and with first class constraints . Their dynamics is obtained from the action
$`S[q^i,p_i,\lambda ^m]`$ $`=`$ $`{\displaystyle 𝑑\tau \left\{\frac{dq^i}{d\tau }p_i\lambda ^mC_m(q^i,p_i)\right\}},`$ (1)
which is invariant under arbitrary reparametrizations of the parameter $`\tau `$. The parameter $`\tau `$ is unphysical and unobservable, like the time coordinate in general relativity. The unreduced, or extended phase space $`\gamma _{ex}`$ is coordinatized by the canonical pairs $`(q^i,p_i)`$; $`i=1,2,\mathrm{},N`$. The canonical 2-form on $`\gamma _{ex}`$ is $`\omega _{ex}=dp_idq^i`$. The pair $`(\mathrm{\Gamma }_{ex},\omega _{ex})`$ forms a symplectic space. The variation of the action with respect to the canonical coordinates $`q^i,p_i`$ gives the equations of motion
$`{\displaystyle \frac{dq^i}{d\tau }}`$ $`=`$ $`\lambda ^m{\displaystyle \frac{C_m(q^i,p_i)}{p_i}},`$
$`{\displaystyle \frac{dp_i}{d\tau }}`$ $`=`$ $`\lambda ^m{\displaystyle \frac{C_m(q^i,p_i)}{q^i}},`$ (2)
while the variation of the action with respect to the Lagrange multipliers $`\lambda ^m`$ gives the constraint equations
$`C_m`$ $`=`$ $`C_m(q^i,p_i)=0,m=1,2,\mathrm{},M.`$ (3)
Thus, the dynamics of the system with respect to $`\tau `$ is the unfolding of the gauge symmetry generated by the first class constraints, i.e., dynamics is gauge.
The first class constraints satisfy, in general, a “non-Lie” algebra
$`\{C_m,C_n\}`$ $`=`$ $`C_{mn}^l(q^i,p_i)C_l,`$ (4)
and the number of independent physical degrees of freedom of the theory is $`D=NM`$. The constraint surface $`\gamma `$ in $`\gamma _{ex}`$ defined by the constraint equations (3) is a $`(2D+M)`$-dimensional manifold. The restriction $`\omega `$ of $`\omega _{ex}`$ to the constraint surface $`\gamma `$ is of rank $`2D`$. The $`M`$ null directions of $`\omega `$ are the infinitesimal transformations generated by the constraints. They define the gauge orbits on $`\gamma `$. The physical phase space $`\gamma _{ph}`$ is the space of these orbits. This is the space of the physically distinct solutions of the equations of motion.
The space $`(\gamma ,\omega )`$ is a presymplectic space, which contains the full dynamical information about the system. Hence dynamical systems in this form are also called ‘presymplectic systems’. $`\gamma `$ can be parameterized by the set of independent coordinates $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a,t^m)`$, where $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a),a=1,2,\mathrm{},D`$ are canonical variables that coordinatize the physical phase space $`\gamma _{ph}`$, and $`t^m,m=1,2,\mathrm{},M`$ coordinatize the orbits. In general this coordinatization can hold only locally, and different charts may be needed to cover the entire space.
Any conventional dynamical system with phase space $`(\gamma _{ph},\omega _{ph}=d\stackrel{~}{p}_ad\stackrel{~}{q}^a)`$, and Hamiltonian $`H=H(\stackrel{~}{p}_a,\stackrel{~}{q}^a)`$ can be represented as a presymplectic system as
$$(\gamma =\gamma _{ph}\times R,\omega =\omega _{ph}H(\stackrel{~}{p}_a,\stackrel{~}{q}^a)dt),$$
(5)
where $`t`$ is the coordinate in $`R`$, and corresponds to the external time variable. The difference between the conventional formulation and the presymplectic formulation is only in the fact that this time variable is treated on the same footing as the other variables. As a concrete example, we may imagine that $`H`$ is the harmonic oscillator Hamiltonian describing the small oscillations of a pendulum, while $`t`$ is the reading of a physical clock. Then the presymplectic system (5) describes how two equal-footing physical variables (the pendulum amplitude and the clock reading) evolve with respect to one another. In general covariant systems, such as any general relativistic system, this ‘equal footing’ status between all physical variables is an essential feature of the theory. It expresses the major physical discovery of general relativity: the complete relativity of spacetime localization.
Note that the canonical coordinates $`\stackrel{~}{q}^a`$, and $`\stackrel{~}{p}_a`$ are the physical observables of the system. They are gauge-invariant. They satisfy $`\{\stackrel{~}{q}^a,\stackrel{~}{p}_b\}=\delta _b^a`$ on the physical phase space. In these coordinates, the physical symplectic form on $`\gamma _{ph}`$ is $`\omega _{ph}=d\stackrel{~}{p}_ad\stackrel{~}{q}^a`$. The general solution of the equations of motion is simply given by the embedding equations of the orbits in $`\gamma _{ex}`$, that is
$`q^i`$ $`=`$ $`q^i(t^m;\stackrel{~}{q}^a,\stackrel{~}{p}_a),`$ (6)
$`p_i`$ $`=`$ $`p_i(t^m;\stackrel{~}{q}^a,\stackrel{~}{p}_a).`$ (7)
Each set $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a)`$ determines a solution; along each solution, the quantities $`(q^i,p_i)`$ depend on the $`M`$ parameters $`t^m`$ (instead than just on a single time variable) because of the gauge freedom in the evolution. The inverse relations of (6)-(7) give the dependence of the physical observables $`\stackrel{~}{q}_a`$, and $`\stackrel{~}{p}_a`$ from the original coordinates
$`\stackrel{~}{q}^a`$ $`=`$ $`\stackrel{~}{q}^a(q^i,p_i),`$ (8)
$`\stackrel{~}{p}_a`$ $`=`$ $`\stackrel{~}{p}_a(q^i,p_i),`$ (9)
as well as the orbit coordinates $`t^m`$
$`t^m`$ $`=`$ $`t^m(q^i,p_i).`$ (10)
The quantities (8) and (9,) commute with all the constraints, and provide a complete set (in the sense of Dirac) of gauge-invariant observables. Every other physical observable can be obtained from them.
Let us recall how evolution can be obtained from the basic observables (8) and (9) . If we plug the gauge variables (10) into the full solution (6) and (7) we obtain the equations
$`q^i`$ $`=`$ $`q^i(t^m(q^i,p_i);\stackrel{~}{q}^a,\stackrel{~}{p}_a),`$ (11)
$`p_i`$ $`=`$ $`p_i(t^m(q^i,p_i);\stackrel{~}{q}^a,\stackrel{~}{p}_a).`$ (12)
In general, $`2NM`$ of these equations are independent. For each physical state of the system, determined by the value of $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a)`$, these equations define an $`M`$ dimensional subspace in the phase space. Therefore each state determines a set of relations on the original phase space variables. These relations represent the dynamical information on the system; they provide the full solution of the dynamics in a gauge-invariant fashion .
In particular, we might arbitrarily choose a set of M coordinates $`q^m`$ (or momenta $`p^m`$; or a combination of both) as independent ‘clock and position’ variables, and express the evolution of the remaining set of coordinates and momenta as functions of these $`q^m`$ for any physical state $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a)`$. For each fixed numerical value $`\widehat{q}^m`$ of the coordinates $`q^m`$, we have a well defined gauge-invariant observables in $`\gamma _{ph}`$. For instance, let us chose (arbitrarily) $`q^1`$ as a dependent ‘partial’ observable<sup>3</sup><sup>3</sup>3A ‘partial’ observable is a physical quantity to which we associate a number, such as time $`t`$, position $`x`$ or electric field $`E`$. A ‘complete observable’ is a physical quantity that can be predicted if the state is known, or, equivalently, that gives us information on the state, for instance the value $`E(t,x)`$ of the electric field in a certain point $`x`$ at a certain time $`t`$. For the notions of partial observable and complete observable, see ., and the next $`M`$ of the $`q^i`$’s, as independent ‘partial’ observables, or ‘clock and position variables’. That is, let us choose $`m=2,\mathrm{},M+1`$. Pick $`M`$ fixed numerical values $`\widehat{q}^m`$ for the $`M`$ variables $`q^m`$. Generically, this fixes uniquely a point on every orbit. The value $`Q_{\widehat{q}^m}^1`$ of $`q^1`$ on this point depends on the orbit, and can be obtained from (8-10). Let it be
$$Q_{\widehat{q}^m}^1=Q_{\widehat{q}^m}^1(\stackrel{~}{q}^a,\stackrel{~}{p}_a).$$
(13)
Here all the $`q^i`$ are partial observables, while $`Q_{\widehat{q}^m}^1`$ is a complete observable. The function (13) is gauge-invariant, well defined on $`\gamma _{ph}`$ and expresses the relative evolution of $`q^1`$, as a function of the $`q^m,m=2,\mathrm{},M`$. It is called an ‘evolving constant of the motion’, or simply a ‘relational observable’ .
The quantum theory can be constructed by imposing the quantum constraints on the unconstrained Hilbert space $``$ (or some suitable extension of the same if the constraints have continuum spectrum). The space of solutions of the constraint equations is the physical Hilbert space $`_{phys}`$ of the theory. (If $`_{phys}`$ is not a subspace of $``$ a scalar product is determined in $`_{phys}`$ by the requirement that the self-adjoint observables in $``$ which are well defined on $`_{phys}`$ be still self-adjoint.) Generically, we expect that out of the operators corresponding to the set of $`2D`$ gauge invariant observables $`(\stackrel{~}{q}^a,\stackrel{~}{p}_a)`$, we can define $`D=NM`$ commuting operators, $`\widehat{O}_a`$, $`a=1,2,\mathrm{},D`$ forming a complete Dirac set. Assuming for simplicity these have discrete spectrum, a basis of physical states is labeled by their quantum numbers $`n_a`$, $`a=1,2,\mathrm{},D`$. A general physical state is killed by all the constraints $`\widehat{C}_m|\psi =|0`$, and can be written as
$`|\psi `$ $`=`$ $`{\displaystyle \underset{n_1,\mathrm{},n_D}{}}c_{n_1,\mathrm{},n_D}|n_1,\mathrm{},n_D.`$ (14)
Physical evolution is described by (Heisenberg) operators corresponding to relational classical quantities such as (13). In constructing these operators, ordering and consistency problem might, in general, be serious.
## 3 Covariant statistical mechanics
Can we use statistical mechanics methods in a covariant, presymplectic framework? Energy plays an important role in statistical mechanics, and here there is no energy. Statistical mechanics relies on the idea that systems thermalize to equilibrium in time. What is thermalization in a covariant context, in which there is no external time variable? To address these questions, our strategy will be to recall Boltzmann’s logic, to rephrase it in the language of the presymplectic formulation of a conventional system, and from here, to extend it to presymplectic systems that do not correspond to a conventional system.
Consider a Boltzmann gas in a closed box. The gas is composed by a very large number $`𝒩`$ of identical molecules. Begin by considering each molecule as free. Let $`\gamma `$ be the phase space of a single molecule. For instance, if we neglect rotational and vibrational motion, we may assume $`\gamma `$ to be six dimensional. Since the molecule is assumed to be free, its motion is very simply described by a free motion in $`\gamma `$. The phase space $`\mathrm{\Gamma }`$ of the entire gas has dimension $`6𝒩`$. The motion of the entire gas is described by a simple motion in $`\mathrm{\Gamma }`$ as well. Under these assumptions, the gas does not thermalize, and we cannot use statistical methods. For instance if we started with all the molecules bouncing up and down within the right half of the box, they would continue to do so forever, never expanding to the left-hand part of the box. To have thermal behavior, we need the particles to interact. However, taking the actual physical interaction among the molecules into account complicates the dynamical problem dramatically, and puts it far outside our theoretical capabilities.
Boltzmann’s genius found a way in between, by postulating a ‘small,’ ‘thermalizing’ interaction among the molecules. The molecules bounce, attract and repel in a non-trivial manner. In the theoretical description, we simply assume that each molecule is still free most of the time, but, once in a while, it interacts with another molecule. We are not concerned with the details of this interaction, except for the assumption that the interaction is maximally thermalizing, that is, it conserves a minimal number of physical quantities. Under this assumption, motion in $`\mathrm{\Gamma }`$ becomes ergodic and we have thermalization. As time goes on, the state of the gas will fill up all allowed regions in $`\mathrm{\Gamma }`$. Of course, there are quantities that must be conserved in any interactions, due to the homogeneity properties of the spacetime in which the gas lives, such as momentum and energy. The presence of the box walls forces the total momentum to be zero, and the only non-trivial conserved quantity is the total energy. Anything else is washed away by the thermalizing interaction. We assume that time averages are the same as ensemble averages, and that under the action of the thermalizing interaction all microstates of the gas become equiprobable, with the only constraint given by the value of the total energy. Thus macroscopic (microcanonical) states can be labeled by a single parameter, their total energy. As is well known, the quantitative consequences of this very delicate argument, considered borderline fantasy by Boltzmann’s contemporaries, are strikingly accurate in a truly impressive range of physical contexts.
In the course of the dynamics, the motion of a single molecule can be followed within its phase space $`\gamma `$. This motion is free for most of the time, but at certain times it gets suddenly altered: when the molecule interacts with another molecule. Assuming equiprobability, a simple calculation shows then that the time averaged distribution of the states of a single molecule, and thus the distribution of the molecules over the states, is given by $`\rho e^{\beta H}`$, where $`H`$ is the free Hamiltonian of the particle and $`\beta `$, the (inverse) temperature, can be computed from the total energy.
Let us now describe the same system in the presymplectic framework. First, let the particles be free. The key difference with the previous description is that a point in the physical phase space $`\gamma _{ph}`$ does not represent anymore the state of the particle at some time. Rather, it represents a single full solution of the equations of motion. (It is like a classical analog of a Heisenberg state, versus a Schrödinger state.) Thus, the particle motion is now described by a single, non moving, point in $`\gamma _{ph}`$, which represents a full orbit in $`\gamma `$. Similarly, the motion of the entire gas is given by a single non-moving point in $`\mathrm{\Gamma }_{ph}`$, corresponding to a full gauge orbit in $`\mathrm{\Gamma }`$. As there is no time, there is no time for moving around.
However, the magic, once again, happens when we turn Boltzmann’s ‘small’ interaction on. The dynamics of the full system is still given by a single non-moving point in $`\mathrm{\Gamma }_{ph}`$, or, equivalently, by a single orbit in $`\mathrm{\Gamma }`$. However, what about the dynamics of a single molecule? Since the phase space $`\gamma _{ph}`$ is defined by the dynamics of the system, and not just its kinematic as in the conventional case, it seems that the motion of a single molecule cannot be described in $`\gamma _{ph}`$ at all in the interacting situation, because $`\gamma _{ph}`$ is the space of the free motions of the molecule. It seems to be a core difficulty. However, there must be a way out, since, after all, we are describing the same physics as before. Indeed, the way out is provided precisely by the assumptions about Boltzmann’s thermalizing interaction. Observe that the orbits in $`\mathrm{\Gamma }`$ do correspond to free motions of the single particles, interrupted by interactions. Each such orbit gives, for every particle, a collection of free motions, namely a collection of points in $`\gamma _{ph}`$. In other words, what the interaction does is simply make the (timeless) state in $`\gamma _{ph}`$ diffused. A full orbit in $`\mathrm{\Gamma }_{ph}`$ determines a distribution of points in $`\gamma _{ph}`$. Under our assumptions, the density of these points must clearly be given by $`\rho e^{\beta H}`$!
What conclusion can we draw from this exercise of re-expressing Boltzmann’s ideas in a timeless language? The first conclusion is that we can still think in terms of the Boltzmann’s distribution on the states of the subsystem, even in a timeless context. It is true that nothing moves in the physical phase space of a fully parameterized system, and so it seems that nothing can ever thermalize. But the effect of the interaction between the subsystems can be represented precisely by a distribution on the space of timeless non-interacting states<sup>8</sup><sup>8</sup>8In some sense, we are dealing here with a ‘covariant ideal gas’. As in standard ideal gases, there is no specific interaction in the formulae involved. Nevertheless, the ideal gas thermalizes thanks to the Boltzmann thermalizing interaction, which is implicitly assumed. If specific interaction terms among the single components $`s`$ were allowed, then it would be interesting to study what thermalization could mean in that context, as well as its relationship with the many-body forces required to get separability of the whole system $`S`$ (the cluster decomposition property) . This issue is not addressed here and we leave it for future developments..
The second conclusion regards the energy. Why does the energy still play a role, when the presymplectic formalism treats the time variable, and thus the energy, just as one among other variables? The answer, from the above discussion, is not that the energy has any special importance by itself. Rather, it is that we have simply fed into the formalism the information that the small interaction between the subsystems washes away everything excepts energy. But there is nothing sacred about energy conservation. Energy conservation is just a consequence of invariance under time shift, which, in turn, is a feature of the homogeneity of the Minkowski solution under time shifts. We have learned from general relativity that the Minkowski gravitational field is just one among many possible fields. There is no fundamental energy conservation in nature.
On the other hand, the discussion above leads us to see precisely under which conditions we can still use Boltzmann’s statistical mechanics in a covariant context. We can, anytime we have a system $`S`$ that can be seen as composed by a large number of identical subsystems $`s`$, whose dynamics is given by a free part which we understand well, plus a ‘small interaction’ that can thermalize the macrosystem, and conserves, say, only some global quantities $`O_l`$. We can formalize our conclusions as follows.
Let $`(\gamma ,\omega )`$ be the presymplectic space describing $`s`$. Let $`S`$, with presymplectic space $`(\mathrm{\Gamma },\mathrm{\Omega })`$ be composed by a large number $`𝒩`$ of systems $`s_{(n)},n=1,\mathrm{},𝒩`$, all identical to $`s`$, having presymplectic spaces $`(\gamma _{(n)},\omega _{(n)})`$. By this we mean
$`\mathrm{\Gamma }`$ $`=`$ $`\times _n\gamma _{(n)},`$ (15)
$`\mathrm{\Omega }`$ $`=`$ $`{\displaystyle \underset{n}{}}\omega _{(n)}+\omega _{int},`$ (16)
where $`\times `$ indicates the Cartesian product and $`\omega _{int}`$ gives the interaction between the subsystems. Next, let us assume that there are $`L`$ quantities $`O_l`$, defined on $`\gamma _{ph}`$ (and thus on $`\gamma `$) such that the corresponding global quantities
$$𝒪_l=\underset{n}{}O_l^{(n)},l=1,\mathrm{},L<D,$$
(17)
are invariant along the orbits of $`\mathrm{\Omega }`$, that is
$$X(𝒪_l)=0,l=1,\mathrm{},L<D,$$
(18)
for all vector fields $`X`$ in $`\mathrm{\Gamma }`$ such that
$$\mathrm{\Omega }(X)=0.$$
(19)
We call these quantities ‘conserved’. Finally, we assume that $`\omega _{int}`$ suitably thermalizes all other variables besides the $`O_l`$’s. This means, precisely as above, that all allowable (combined) states of the $`s_{(n)}`$ systems are, on average, equally covered, in moving along a generic orbit in $`\mathrm{\Gamma }`$.
Under these conditions, we can straightforwardly construct a covariant statistical formalism. A state in $`\mathrm{\Gamma }_{ph}`$, determines, for each $`s_{(n)}`$ a distribution $`\rho `$ on $`\gamma _{ph}`$, which gives the distribution of ‘initial states’ of the component system as we move around the corresponding orbit of the interacting composite system. For a generic state in $`\mathrm{\Gamma }_{ph}`$, this distribution can be computed using conventional statistical techniques, in particular, by assuming that the distribution is the one that maximizes the number of possible microstates compatible with the given macrostate. The result is straightforward: the (unnormalized) probability distribution on the phase space is
$$\rho =e^{\gamma ^lO_l},$$
(20)
where $`\gamma ^l`$ are the intensive thermodynamical parameters that determine the equilibrium of the members of the ensemble with respect to the transfer of the quantities $`[\widehat{O}_l]`$.
Instead of detailing the classical theory, we discuss directly the quantum theory. We use von Neumann’s density operator formalism . We ask the ensemble to satisfy a maximum entropy principle. In other words, we ask the quantum statistical entropy $`S`$ per constituent member of the ensemble given by
$`S=k\text{Tr}\left(\widehat{\rho }\mathrm{ln}\widehat{\rho }\right),`$ (21)
to be a maximum under the constraints
$`\text{Tr}(\widehat{\rho }\widehat{O}_l)`$ $`=`$ $`\overline{O}_l,`$
$`\text{Tr}\widehat{\rho }`$ $`=`$ $`1,`$ (22)
where $`\widehat{O}_l`$ are the quantum operator corresponding to the conserved quantities $`O_l`$ and $`\overline{O}_l`$ are fixed average values. $`\widehat{\rho }`$ is the density operator. The density operator $`\widehat{\rho }`$ that fulfills these requirements is
$`\widehat{\rho }`$ $`=`$ $`𝒵^1e^{\gamma ^l\widehat{O}_l},`$ (23)
with
$`𝒵=\text{Tr}e^{\gamma ^l\widehat{O}_l},`$ (24)
the partition function. The thermodynamical parameters $`\gamma ^l`$ can be obtained from the conditions (22) provided that the matrix $`\frac{[\widehat{O}_i]}{\gamma ^j}`$ have non-vanishing determinant. Clearly, they are the parameters that measure the equilibrium of the members of the ensemble with respect to the transfer of the quantities $`[\widehat{O}_l]`$.
In the conventional case of a non-covariant system formulated in covariant terms, only one non-trivial quantity is conserved, the energy, and we obtain the standard results. In particular, we can consider a gravitational system with an asymptotically flat gravitational field; in this case the Hamiltonian is given by suitable boundary terms and the observables at infinity evolve in the Lorentz time of the asymptotic metric. The system is Lorentz invariant for the asymptotic Lorentz transformations, and therefore, in particular, invariant for time translations. Therefore all interactions conserve the asymptotic Lorentz energy and the theory considered here reduces to the standard results. A specific example of this is given by all the literature on black hole thermodynamics an statistical mechanics, in which, in general, the gravitational field is assumed to be asymptotically flat.
## 4 An example
As a simple example, we take as component system $`s`$ a model with two non-commuting Hamiltonian constraints and one physical degree of freedom which was studied in . This model mimics the constraint structure of general relativity. We refer to for all details.
The model we consider is defined by the action
$`S[\stackrel{}{u},\stackrel{}{v},N,M,\lambda ]={\displaystyle \frac{1}{2}}{\displaystyle 𝑑t\left[N(𝒟\stackrel{}{u}^2+\stackrel{}{v}^2)+M(𝒟\stackrel{}{v}^2+\stackrel{}{u}^2)\right]},`$ (25)
where
$$𝒟\stackrel{}{u}=\frac{1}{N}(\dot{\stackrel{}{u}}\lambda \stackrel{}{u}),𝒟\stackrel{}{v}=\frac{1}{M}(\dot{\stackrel{}{v}}+\lambda \stackrel{}{v});$$
(26)
the two Lagrangian dynamical variables $`\stackrel{}{u}=(u^1,u^2)`$ and $`\stackrel{}{v}=(v^1,v^2)`$ are two-dimensional real vectors; $`N`$, $`M`$ and $`\lambda `$ are Lagrange multipliers. The squares are taken in $`R^2`$: $`\stackrel{}{u}^2=\stackrel{}{u}\stackrel{}{u}=(u^1)^2+(u^2)^2`$. The action can be put in the form (1),
$`S[\stackrel{}{u},\stackrel{}{v},\stackrel{}{p},\stackrel{}{\pi },\lambda ^m]={\displaystyle 𝑑\tau \left\{\dot{\stackrel{}{u}}\stackrel{}{p}+\dot{\stackrel{}{v}}\stackrel{}{\pi }\lambda ^mC_m(q^i,p_i)\right\}}.`$ (27)
The canonical variables $`(\stackrel{}{u},\stackrel{}{v},\stackrel{}{p},\stackrel{}{\pi })`$ define the eight dimensional extended phase space $`\gamma _{ex}`$, with symplectic form $`\omega _{ex}=d\stackrel{}{p}d\stackrel{}{u}+d\stackrel{}{\pi }d\stackrel{}{v}`$, also $`\lambda ^1=N`$, $`\lambda ^2=M`$, and $`\lambda ^3=\lambda `$. The constraints
$`C_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\stackrel{}{p}^2\stackrel{}{v}^2),`$
$`C_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\stackrel{}{\pi }^2\stackrel{}{u}^2),`$
$`C_3`$ $`=`$ $`\stackrel{}{u}\stackrel{}{p}\stackrel{}{v}\stackrel{}{\pi },`$ (28)
are first class and define a five dimensional constraint surface $`\gamma `$ in $`\gamma _{ex}`$, and $`\omega =\omega _{ex}|_\gamma `$.
A complete set of gauge-invariant quantities is given by the two continuous quantities $`JR^+,\varphi S_1`$ and two discrete quantity $`ϵ,ϵ^{}=\pm 1`$, defined by
$`ϵ`$ $`=`$ $`{\displaystyle \frac{u^1p^2p^1u^2}{|u^1p^2p^1u^2|}},`$
$`ϵ^{}`$ $`=`$ $`{\displaystyle \frac{\pi ^1v^2v^1\pi ^2}{|\pi ^1v^2v^1\pi ^2|}},`$
$`J`$ $`=`$ $`|u^1p^2p^1u^2|,`$
$`\varphi `$ $`=`$ $`\mathrm{arctan}{\displaystyle \frac{u^1v^2p^1\pi ^2}{u^1v^1p^1\pi ^1}},`$ (29)
These can be taken as coordinates of the physical gauge-invariant phase space. The quantity $`J`$ resembles an angular momentum, and thus it is called as such.
Let us now consider a large number of systems of this kind which are interacting weakly. The composite system dynamics is given by the presymplectic system (16). Let us assume, as an example, that $`\omega _{int}`$ is a sum of binary interactions in which the sum of the two angular momenta, while all other quantities are thermalized. This defines the statistical mechanics of the composite system.
In the quantum theory, we take a complete Dirac set of commuting operators $`\widehat{J}`$, $`\widehat{ϵ}`$ and $`\widehat{ϵ^{}}`$. Their spectrum, worked out in , is
$`\widehat{J}|m,ϵ,ϵ^{}_N`$ $`=`$ $`J_m|m,ϵ,ϵ^{}_N=m\mathrm{}|m,ϵ,ϵ^{}_N,`$
$`\widehat{ϵ}|m,ϵ,ϵ^{}_N`$ $`=`$ $`ϵ|m,ϵ,ϵ^{}_N,`$
$`\widehat{ϵ^{}}|m,ϵ,ϵ^{}_N`$ $`=`$ $`ϵ^{}|m,ϵ,ϵ^{}_N.`$ (30)
$`m`$ is a positive integer, and $`ϵ`$ and $`ϵ^{}`$ take values 1 and -1. The states $`|m,ϵ,ϵ^{}_N`$ form a normalized basis in the physical Hilbert space of the theory.
$`{}_{N}{}^{}m,ϵ,ϵ^{}|\stackrel{~}{m},\stackrel{~}{ϵ},\stackrel{~}{ϵ}^{}_{N}^{}=\delta _{m,\stackrel{~}{m}^{}}\delta _{ϵ,\stackrel{~}{ϵ}}\delta _{ϵ^{},\stackrel{~}{ϵ}^{}}.`$ (31)
The quantity $`J`$ is represented by the operator $`\widehat{J}`$, which is a kind of angular momentum as we have mentioned. If we assume that the thermalizing interaction conserves the total value of $`J`$, we have immediately the density operator
$`\widehat{\rho }`$ $`=`$ $`{\displaystyle \frac{e^{\gamma \widehat{J}}}{𝒵}}.`$ (32)
Using $`\text{Tr}\widehat{\rho }=1`$, we get the partition function
$`𝒵={\displaystyle \underset{m=1}{\overset{\mathrm{}}{}}}\omega _me^{\gamma J_m}={\displaystyle \frac{4}{e^\gamma \mathrm{}1}},`$ (33)
with $`\omega _m=4`$ because there are 4 states for a given $`m`$, and also $`e^\gamma \mathrm{}<1`$ has been used (i.e., positive ‘temperature’ $`\gamma `$ has been assumed). The angular momentum per constituent is given by
$`L:=[\widehat{J}]`$ $`=`$ $`{\displaystyle \frac{\text{Tr}(e^{\gamma \widehat{J}}\widehat{J})}{𝒵}}={\displaystyle \frac{}{\gamma }}\mathrm{ln}𝒵={\displaystyle \frac{e^\gamma \mathrm{}\mathrm{}}{(e^\gamma \mathrm{}1)}},`$ (34)
and the entropy per constituent
$`S`$ $`=`$ $`k\sigma =k\text{Tr}\left(\widehat{\rho }\mathrm{ln}\widehat{\rho }\right)=k\mathrm{ln}𝒵+k\gamma L.`$ (35)
The parameter $`\gamma `$ characterizes the equilibrium state of the system with the reservoir and plays here the role of a temperature. If we had an empirical thermodynamics of this system, we could identify this parameter with an empirically determined thermodynamical quantity.
## 5 Gas of free relativistic particles
Before concluding, we discuss a simple case, in which some of the ideas presented above can play a role: the case of a gas of free relativistic particles. This is really an oversimplified situation, which can be treated with simpler tools; but the illustration of this case may be instructive, and can be seen as a check that the theory described here is in agreement with other methods in the cases in which other methods can be applied. For simplicity, we remain here in the classical context.
Consider thus a gas of relativistic particles. Can we associate a temperature to this gas? What is the statistical state describing these particles? The Hamiltonian description of a single relativistic particle can be formulated in a manifestly Lorentz covariant fashion as follows. The phase space is coordinatized by the coordinates $`x^\mu `$ and their conjugate momenta $`p_\mu `$ –that is, the symplectic form is $`\omega _{ext}=dx^\mu dp_\mu `$– and the dynamics is given by the constraint $`C=p^2m^2`$. The seven dimensional constraint surface $`\gamma `$ defined by $`C=0`$ with its induced restriction $`\omega `$ of $`\omega _{ext}`$ form the presymplectic space $`(\gamma ,\omega )`$ describing the dynamics of the particle.
Now, one may say that for this system we know that the time is $`t=x^0`$, and the energy is $`E=p_0`$. Therefore we can apply standard statistical mechanics with no difficulties. However, there are two distinct problems. The first is that the entire dynamics of the system is contained in the geometry of $`\gamma `$, which has no a priori specification of which variable is time and which variable is energy. Thus, can we do thermodynamics just on the basis of the actual dynamical laws, without specifying which one is the time parameter? However, there is also a second problem, much more concrete. Suppose we say that $`x^0`$ is the time variable, and $`p_0`$ is the energy. You, on the other hand, use a different Lorentz reference frame, and therefore for you the time is $`x^{}{}_{}{}^{0}=\mathrm{\Lambda }_\mu ^0x^\mu `$ and the energy is $`p_0^{}=\mathrm{\Lambda }_0^\mu p_\mu `$, where $`\mathrm{\Lambda }`$ is a Lorentz transformation. If I write a Boltzmann statistical state using my definition of energy, and you in yours, do we define the same statistical state? It is easy to see that the answer is no, because
$$\rho (x,p)=e^{\beta p_0}\rho ^{}(x,p)=e^{\beta ^{}p_0^{}}.$$
(36)
whatever is $`\beta ^{}`$. So, which one is the correct equilibrium state?
Let us address both problems in terms of the general theory developed above. The key point is that if the gas is formed by particles that are really free, they will never thermalize. Some thermalizing interaction is needed in order to reach an equilibrium state. Thus, we need some additional physical input (this is the key point). On physical grounds, we may for instance observe that our gas of relativistic particles thermalizes by means of relativistic elastic scattering. Therefore, the dynamics of a single particle is not really free: the presymplectic space describing the dynamics of the system is the cartesian product of the spaces of the particles, the total presymplectic form is the sum of the individual presymplectic forms plus the interaction term, as in Eqs.(15 and 16). What are the conserved quantities $`O_l`$, in the sense of Section III? Namely, what are the quantities that are exchanged, but whose total value is conserved, in such an interaction? Clearly, they are the momenta $`p_\mu `$. Therefore, according to the general theory of Section III, the (unnormalized) probability distribution of the states of the single particle is
$$\rho (x,p)=e^{\gamma _\mu p^\mu },$$
(37)
where the quantities $`\gamma _\mu `$ are the intensive parameters describing the system. Straightforward application of standard statistical techniques tells us then that the average 4-momentum is proportional to $`\gamma ^\mu `$:
$$\overline{p}^\mu =\frac{_C𝑑x𝑑pp^\mu e^{\gamma _\mu p^\mu }}{_C𝑑x𝑑pe^{\gamma _\mu p^\mu }}=\frac{d}{d\gamma _\mu }\mathrm{ln}_C𝑑pe^{\gamma _\mu p^\mu }=\frac{\gamma ^\mu }{|\gamma |^2}.$$
(38)
Therefore, $`\gamma ^\mu `$ must be timelike, and therefore there is a preferred Lorentz frame in which $`\stackrel{}{\gamma }=0`$ and $`\gamma _0=\beta `$. This is the frame in which the center of mass of the cloud is at rest. In this frame, the (unnormalized) probability distribution of the states of the single particle is
$$\rho (x,p)=e^{\beta p_0}.$$
(39)
We can learn various lessons from this. First, there is no Lorentz invariant thermal state: a thermal state is in equilibrium in a preferred Lorentz frame of reference, and therefore breaks Lorentz invariance. Second, we can say that the form of the statistical state is physically determined by the fact that the thermalizing interaction conserves $`p^\mu `$, and not by the fact that preferred phase space coordinates play an a priori role of time and energy.
The argument above can indeed be sharpened by a more detailed analysis of the physics of the system. Note that if the only interaction is elastic scattering among the particles, then the gas will diffuse and fail to reach equilibrium. Thus, we need something that keeps the gas contained, in order to have a meaningful thermodynamics. One possibility to keep the gas contained is to put it in a box. The position of the box will then break Lorentz invariance and pick the preferred Lorentz frame in which the box does not move. Alternatively, we may think that the particles are gravitationally bound. To have Lorentz invariance, we need a field theory for the gravitational field. (We can disregard here the difficulties of having point particles, or rigid particles, in general relativity, which play no role in this context). In this case, we can approximate the dynamics of a single particle as the dynamic of a particle in the mean gravitational field of the others. At equilibrium, this gravitational field will be stationary in a a preferred Lorentz frame, the one of the center of mass of the cloud. Now, in both cases a single particle is not longer free, but rather is subjected to an interaction which is not longer Lorentz invariant. In fact, in both cases the interaction preserves energy but not momentum (in the second case, energy is kinetic plus potential.) Therefore in both cases there is a preferred $`p_0`$ which is conserved in the course of the thermalizing interaction. It is clearly this energy the one that enters Eq.(39), because this is the energy which is totally conserved and freely exchanged in the system, and thus which becomes equipartioned. In other words, this is the quantity that is conserved in the thermalizing interaction, and that becomes the extensive thermodynamical parameter governing the system.
The general lesson should be clear at this point. Even if we do not have an a priori recognition of which function on the extended phase space represents time (or energy), we can nevertheless run the statistical mechanics formalism on the basis of the quantities that are preserved in the thermalizing interaction.
## 6 Conclusions and perspectives
We have argued that quantum statistical mechanical techniques can be applied to a macroscopic generally covariant system composed of a large number of generally covariant subsystems. This can be done without arbitrarily selecting a variable as the time variable, and in spite of the absence of a notion of energy.
We recall that in the literature there are two main schools of thought in relation to the ‘problem of time’ in generally relativistic theories. One tries to single out the ‘correct’ time variable among the variables of the covariant theory. The choice determines a preferred Hamiltonian (energy), and thus an unambiguous concept of temperature. The opposite point of view, which we consider more fruitful and we have developed here, takes general covariance more seriously, and keeps all variables on the same footing. From this point of view, temperature plays no fundamental role in the statistical analysis. It may not be defined, or, if it is defined at all, temperature is just one of the intensive macroscopic parameters characterizing the equilibrium configuration of the system.
Our basic idea is that if the macroscopic system can be viewed as being formed from weakly interacting systems, then a full solution of the equation of motion of the macroscopic system determines a distribution of solutions of the equations of motion of the components. The properties of the interaction determine which global quantities are conserved and thus the extensive macroscopic parameters. In turn, these determine intensive thermodynamical parameters that describe the macrostate. The other microscopic degrees of freedom are thermalized away by the interaction. The fact that a preferred single notion of temperature does not necessarily arise is not surprisingly, given the weak and always contingent role that energy plays in general relativistic theories. The statistical mechanics of generally covariant theories does not depend on the notion of energy. Rather, ensembles are determined by the properties of the thermalizing interaction.
What is a thermometer in this context? In the usual context a thermometer is a physical system which has the property of having a macroscopic variable $`h`$ (the height of the mercury column) directly coupled to the average energy. By looking at $`h`$ we measure directly the average energy and therefore the temperature. If local energy is not conserved then a conventional thermometer will keep measuring its own average energy, but this will give little information on the system, because individual subsystems will not thermalize to the same reading of the thermometer. On the other hand, if other intensive parameters $`\gamma ^l`$ are conjugate to other conserved quantities $`O_l`$, then in principle specific “thermometers” measuring the average value of $`O_l`$ may exist. These should play the same role as conventional thermometers.
Our approach has been very abstract, and applications in realistic general relativistic contexts may not be trivial. A first naive idea is to obtain a system composed of subsystems by partitioning space into small patches, each with its own gravitational degrees of freedom. This procedure, however, might interfere badly with general covariance. The example considered in the text suggests to look at the strong coupling limit of general relativity, which is precisely given by a collection of finite dimensional covariant systems. In this context, we recall that near singularities –such as the cosmological one– notions of temperature and entropy are usually very badly defined. Alternatively, one could think of somehow Fourier expanding the gravitational field, and partitioning it in momentum space, following fluid techniques. Perhaps a realistic context in which a covariant statistical mechanics may find application is where matter and strong gravitational fields are both present. The presence of matter could lead to a natural physical way of partitioning the degrees of freedom. In general, any context in which thermal energy can be lost substantially in the gravitational field would, in principle, require a covariant thermodynamics.
On the purely theoretical side, a natural open issue is the quantum statistical mechanics of constrained systems with symmetries in the global quantum states. That is, the covariant version of Fermi-Dirac, and Bose-Einstein statistics.
The relation between coherent states in standard quantum mechanics and thermodynamics suggests that coherent state quantization of constrained systems might bring a better understanding of the thermodynamics of generally covariant systems , and also shed light on the tantalizing issue of the classical limit states in quantum gravity. In particular, consider the spacetime described by a (generic) statistical mixture of gravitational states $`|\psi `$. In loop quantum gravity, $`|\psi `$ are superpositions of $`s`$-knots, or abstract spin networks. In the basis in which $`\widehat{\rho }`$ is diagonal, we have
$`\widehat{\rho }={\displaystyle \underset{\psi }{}}P(\psi )\psi \psi `$ (40)
and we can compute, for instance, the density matrix yielding average macroscopic values of the geometry. In particular, we can use area and volume operators associated with a compact region of space and require something like
$`\overline{A}=\text{Tr}\widehat{\rho }\widehat{A},`$
$`\overline{V}=\text{Tr}\widehat{\rho }\widehat{V}.`$ (41)
This may determine a statistical state of the kind
$`\widehat{\rho }={\displaystyle \frac{e^{\alpha \widehat{A}\beta \widehat{V}}}{𝒵}},`$ (42)
which might describe the physical state of spacetime better than a somewhat arbitrary pure state.
In closing, let us emphasize again that we have not discussed here the statistical mechanics of matter interacting with a fixed gravitational field. Rather, we have considered the full quantum statistical mechanics of spacetime itself. Similarly, we have not discussed the statistical mechanics of black holes, which focuses on (the perturbations of the gravitational field around) certain preferred black hole configuration, and is generally in the asymptotically flat context, in which boundary quantities determine time and conserved energy. Finally, we think that the proposal of the statistical and algebraic origin of the time flow might be reconsidered at the light of the physical ideas discussed here. The magic secret coffer of the relations between gravity, quantum and thermodynamics is still far from being fully open.
## Acknowledgments
Warm thanks to Ted Newman for many discussions and many helpful criticisms. Thanks to an anonymous referee for useful and detailed criticisms to the first version of this work. MM thanks financial support provided by the Sistema Nacional de Investigadores (SNI) of the Secretaría de Educación Pública (SEP) of Mexico. The summer stay of MM at the Department of Physics and Astronomy of the University of Pittsburgh, where this paper was finished, is supported by the Mexican Academy of Sciences and The United States-Mexico Foundation for Science. Also MM thanks all the members of the Department of Physics and Astronomy of the University of Pittsburgh for their warm hospitality. This work was partially supported by NSF grant PHY-9900791.
## References
|
warning/0002/astro-ph0002463.html
|
ar5iv
|
text
|
# Some Aspects of Inverse Compton Emission from 3K Background Photons
## 1 Introduction
The recent history of inverse Compton emission from cosmic background photons and relativistic electrons (IC/3K) began during the period when the origin of extended X-ray emission from clusters of galaxies was unclear. Felten and Morrison (1966), Blumenthal & Gould (1970), and Jones, O’Dell & Stein (1974) were among those who worked out the details of IC/3K, but not until iron lines were detected in the X-ray spectra was it generally accepted that the bulk of the X-ray emission was caused by the thermal bremsstrahlung process.
Although there were many attempts to isolate IC/3K emission from clusters, these were thwarted by the strong thermal emission which is quasi ubiquitous in clusters. The recent breakthrough came from the dual detections of IC/3K from the lobes of Fornax A: spatially with the ROSAT PSPC (Feigelson et al. 1995) and spectrally with ASCA (Kaneda et. al. 1995). In the last two years, there have been papers published on other sources which are relatively free of hot gas emission and from cluster radio halos, both from EUV excesses and from the harder spectral component at energies for which the thermal X-rays are dropping exponentially.
## 2 Salient Features
In this section, we list a few of the attributes of IC/3K emission which are often overlooked.
### 2.1 Every non-thermal radio source also emits IC/3K emission
IC/3K emission is mandatory: every (non-thermal) radio source in the universe is an emitter. This statement entails only the assumption that the 3K background is indeed universal. Thus the only problem is to see if it is strong enough for detection with a given system, and if it can be separated from other emissions.
### 2.2 Electron Energy and the observing frequency
At a given (X-ray) energy, we will be sampling the relativistic electron spectrum at the same value of electron energy ($`\gamma `$) for every source in the universe. This is because the peak frequency, $`\nu _o`$, of the cosmic background spectrum (where the major contribution to the photon energy density occurs) increases as (1+z). A given electron with Lorentz energy, $`\gamma `$, emits most of its radiation at $`\gamma ^2\times \nu _o`$ and the emitted photons at that energy are redshifted by (1+z) when observed on Earth. Thus soft X-rays (1-2 keV) sample $`\gamma `$ 1000 no matter what the redshift. Attempts to integrate these electrons in order to explain a significant part of the soft X-ray background have not been successful.
### 2.3 The importance of determining the amplitude of the electron spectrum at low energies
For B$`<`$5$`\mu `$G, low energy electrons (e.g. $`\gamma <`$1000) are not visible via their synchrotron (radio) emission because the ionosphere blocks frequencies below $``$ 20 MHz. Since these electrons normally have long lifetimes in the weak magnetic fields typically found in extended radio sources, they accumulate over the life of a radio source. In a sense we are sampling an encapsulated history of the radio source: the electron spectrum at low energies tells us about all the electrons produced over the source’s lifetime.
## 3 Observing Strategies - avoid hot gas!
* By going to higher energy: e.g. BeppoSax and RXTE
* By getting outside clusters: e.g. Fornax A lobes and relic radio galaxies such as 0917+75 (Harris et. al. 1995).
* Choose radio sources with large numbers of low energy electrons. This can be achieved by choosing sources likely to have relatively more low energy electrons and to have weak magnetic fields since these require more electrons to get a given radio intensity. Therefore, steep spectrum, low brightness radio sources are best.
## 4 Rewards
Although most of us like to believe that a successful detection of IC/3K yields the average magnetic field strength directly, what we really measure is the amplitude of the electron spectrum at some low energy. To obtain the average field strength, there is still the uncertainty of the form of the electron spectrum between the direct measurement afforded by the IC/3K observation and the segment of the electron spectrum responsible for the radio emission. This is illustrated in figure 1. Since we can’t be sure that a single power law extrapolates from the radio derived segment of the spectrum to the X-ray derived amplitude, the average value of B is uncertain (even if the ROSAT point were not an upper limit).
The electron spectrum below $`\gamma `$=1000 is of particular interest because in weak field regions (and z$`<`$0.5), E<sup>2</sup> halflives can approach 10<sup>9</sup> yr. Therefore, the total number of electrons in this energy range serves as a diagnostic of the total energy of a source during its life. While soft X-rays (e.g. 1.5 keV) provide a direct measurement of the number of electrons at $`\gamma `$=1000, EUVE data can provide estimates at $`\gamma `$=300. In Table 1 we have calculated log N (100$`<\gamma <`$1000) for a variety of radio sources by extrapolating their observed spectra to lower energies.
Table 1 Number of low energy electrons from extrapolated radio spectra
| Source | Component | z | $`\alpha _r`$ | B(eq) | logV | logN | $`\tau `$ |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | | | | ($`\mu `$G) | (cm<sup>3</sup>) | | (yrs) |
| FRII | | | | | | | |
| Cyg A | 2 lobes | 0.057 | 0.6 | 25 | 69.83 | 63.59 | 1E7 |
| RELICS | | | | | | | |
| 0917 | total | 0.12 | 1.0 | 0.7 | 72.77 | 64.23 | 8E8 |
| Coma | total | | 1.18 | 0.3 | 72.19 | 64.53 | 1E9 |
| CenB | 1 lobe | 0.012 | 0.8 | 1.5 | 71.14 | 63.32 | 9E8 |
| 1358+30 | | 0.11 | 0.72 | 0.9 | 72.00 | 63.22 | 8E8 |
| 1401-33 | | 0.0136 | 1.44 | 8 | 71.67 | 64.91 | 1E9 |
| 3C 326 | giantRG | 0.089 | 0.82 | 0.8 | 73.46 | 64.75 | 9E8 |
| IC2476 | | 0.027 | 1.1 | 0.7 | 71.57 | 63.80 | 1E9 |
| FRI | | | | | | | |
| 3C 31 | | 0.016 | 0.63 | 1.5 | 70.61 | 61.97 | 9E8 |
| 0715 | 1arm | 0.07 | 1.1 | 4 | | 62.40 | |
| 1718 | halo | 0.162 | 1.3 | 6 | | 62.90 | |
| HALOS | | | | | | | |
| Coma | total | 0.023 | 1.34 | 0.5 | 72.64 | 65.28 | 1E9 |
Notes to Table 1
z is the redshift
$`\alpha _r`$ is the radio spectral index
B(eq) is the equipartition field
V is the source volume
N is the integrated number of electrons for 100$`<\gamma <`$1000
$`\tau `$ is the halflife for $`\gamma `$=1000 electrons.
While there are considerable uncertainties on the magnetic field and volume estimates, a substantial difference is seen for log N amongst Cyg A and relics; FRIs; and the Coma halo. These values support the notion that the relics were once FR II radio galaxies and that if cluster halos arise from the remnants of FR I galaxies such as tailed radio galaxies, it would require of order 1000 such contributions in 10<sup>9</sup> years.
Once an estimate of the average magnetic field strength is obtained, inferences can be made on the total energy density, $`u(tot)`$, and thus on the composition of the relativistic particles. There are two contributions to the particle energy density, $`u(p)`$, which are not ’counted’ by the radio observations: low energy electrons and relativistic protons. For low brightness radio sources, it is probably the case that most of the source is ’relaxed’ in the sense that there are not active shock regions (e.g. hotspots) and equipartition between magnetic field energy density, $`u(B)`$, and $`u(p)`$ is a reasonable expectation. Thus knowledge of $`u(B)`$ translates to an estimate of $`u(p)`$ and $`u(tot)`$. If the required $`u(p)`$ is large enough, the presence of relativistic protons may be indicated. The major uncertainty in this process is probably the unknown filling factor, $`\varphi `$, and if $`\varphi `$ is substantially less than unity, the additional question: ”Do the particles and field occupy the same volume?”.
## 5 Discussion
### 5.1 Distinction between radio lobes (including relics) and cluster halos
While we have some evidence that relativistic and thermal plasmas are spatially distinct in the case of lobes of radio galaxies (e.g. Cygnus A, Carilli et al. 1994), we do not have any confidence as to the situation for cluster halos. This is partly caused by the uncertainty concerning the genesis of radio halos in clusters. If relativistic plasma is completely mixed with the thermal plasma, then it is reasonable to assign a value close to unity for the filling factor, $`\varphi `$, but if the actual synchrotron emitting regions are to be distinct from the thermal plasma (in the sense of a magnetic boundary separating the two), then $`\varphi `$ is most likely quite small. In the former case we should probably include the thermal energy density in any application of the equipartition condition, whereas in the latter case, we would be justified in balancing the relativistic particle energy density with the magnetic field energy density.
In a sense, we have two distinct problems. For relics and radio lobes such as Fornax A, we do not have to contend with excessive thermal emission and can get a fairly clean measurement if IC/3K intensity and distribution. But for clusters, it is only with difficulty that we avoid the thermal contamination.
What are the implications of the fact that some estimates of the Coma field strength are in reasonable accord with the minimum energy B field for $`\varphi `$=1 and $`u(p)`$=$`u(B)`$? If the synchrotron emitting plasma were distinct from the cluster gas and $`\varphi <<`$1, then the equipartition field B(eq) would be much larger than the IC/3K estimate, and that does not seem to be the case.
### 5.2 Current IC/3K detections and the implied B fields
Using the relationships of Harris & Grindlay (1979), we have calculated the value of the average magnetic field strengths in some of the sources for which IC/3K detections have been claimed or an upper limit has been measured. These are presented in Table 2 together with magnetic field estimates published by the authors.
### 5.3 Comparison of B field estimates
The B field estimates for clusters discussed at this workshop can be roughly divided into 3 categories, depending on the method used and the type of source.
B=0.1 to 0.2 $`\mu `$G
These sorts of values were put forward on the basis of the excess X-ray flux observed over that expected from the hot gas in the Coma cluster. Both the EUV excess and the BeppoSax measurements yield B values $`<`$0.2 $`\mu `$G. Fusco-Femiano et al. (1999) give fx(20-80keV)=2.2$`\times `$10<sup>-11</sup> ergs cm<sup>-2</sup> s<sup>-1</sup> and B$``$ 0.15 $`\mu `$G. Note however that Henriksen (1998) using ASCA and HEAO-I data obtained an upper limit of fx(20-60keV)$`<`$2.9$`\times `$10<sup>-12</sup> ergs cm<sup>-2</sup> s<sup>-1</sup> and B$`>`$0.26 $`\mu `$G.
B$``$ few $`\mu `$G
Magnetic field strengths of a few $`\mu `$G have been derived from equipartition calculations for the Coma radio halo, from the IC/3K estimates for the lobes of Fornax A, and from some estimates for clusters from Faraday rotation.
B = 5 to 20 $`\mu `$G
These stronger B field estimates generally come from the Faraday rotation measurements of unresolved sources in or behind clusters. The greatest uncertainty here is the scale size of magnetic field cells (for which the fields can reverse direction). If there are a large number of cells along the line of sight which have sufficient field strength and electron density to make a substantial contribution to the Faraday rotation, then the required B field will be significantly larger than for the case of only a few cells.
While we are not in a position to resolve the disparity in these B field estimates, we suspect that typical field strengths in clusters and relaxed radio lobes probably lie within the range 0.5$`<B<`$3$`\mu `$G. If this were to be the case, then published EUV and hard X-ray excesses are either wrong or arise from an emission process other than IC/3K and clusters have large scale coherent fields so that there are relatively few field reversals along the line of sight.
## 6 Conclusion
We have a fairly good idea of the problems involved in using IC/3K emission to derive physical quantities. Most of these difficulties are not going to go away soon, but the quality of the X-ray data should improve significantly with the advent of several missions with larger collecting area, better resolution, and extended frequency coverage.
Table 2 Magnetic field estimates from IC/3K detections Luminosities B Fields(A) B Fields(RX) Log Lr Log Lx Lr/Lx B(eq) B(ic) B(eq) B(ic) (erg/s) ($`\mu `$G) ($`\mu `$G) CLUSTERS Coma 1 40.85 43.71 0.001 0.15 0.2 0.05 RADIO LOBES Fornax A Keast 41.40 40.95 3.24 2.4 1.0 1.8 Kwest 41.64 40.89 5.62 3.5 1.0 2.6 Fwest 41.78 41.08 5.01 3.0 1.9 1.4 2.0 RELICS A 85 0038-096 41.80 42.95 0.07 0.9 2.2 0.4 Cen B $`<`$lobe$`>`$ 42.20 41.50 5.01 3.1 1.4 1.7 0917 total 41.83 $`<`$42.30 $`>`$0.34 0.7 $`>`$0.7 0.7 $`>`$0.7
Notes to Table 2
Luminosities are indicative; the radio band was generally 10<sup>7</sup> to 10<sup>10</sup> Hz (except as noted below). The X-ray bands depend on the telescopes used. B(eq) is the equipartition field (filling factor of unity and no protons) and B(ic) is the field strength derived from the IC/3K measurements. Both the original authors’ values (A) and those calculated from the equations in Harris and Grindlay (1979) \[’RX’\] are given.
COMA 1
Entries are based on the composite radio spectrum presented by Fusco-Femiano et al. (1999); ’FF’ hereafter. The general properties of the radio spectrum of the halo were checked against earlier presentations. Note however that whereas earlier work found $`\alpha _r`$ around 1.2 to 1.3, FF argue for curvature above 100 MHz, thus deducing a spectral break with an injection spectrum characterized by $`\alpha `$=0.96. For RX inputs, we take the synchrotron band to be 30-630MHz, with a fiducial flux density of S(100MHz)=10 Jy (from their figure). The X-ray data are from BeppoSax (ibid.). FF combine two detector results and make a spectral fit with two components (thermal and a power law). For the power law, they get $`\alpha _x`$ =0.6$`\pm `$0.4. For the volume, we take the diameter used by FF: 25’ (1 Mpc).
Fornax A
We used two papers: Kaneda et al. (1995) - ASCA spectral analysis (”Keast” and ”Kwest”), and Feigelson et al. (1995)- spatial analysis from the ROSAT PSPC. Feigelson only did the West lobe (”Fwest”).
A85
The results are based on Bagchi et al. (1998) from a wavelet analysis of ROSAT PSPC data.
CENTAURUS B (a.k.a. PKS1343-601)
This is a relic radio galaxy, somewhat obscured by the Milky Way. Radio data are from McAdam (1991) and the X-ray data are from Tashiro et al. (1998). Since the X-ray data are for the entire source, and since the lobes are not the same size and brightness, we model an ’average lobe’, taking the size, r=300” (that of the S lobe), and divide both radio and X-ray fluxes by 2.
|
warning/0002/cs0002013.html
|
ar5iv
|
text
|
# Computing and Comparing Semantics of Programs in Multi-valued Logics11footnote 1A preliminary version of this paper appeared in the form of an extended abstract in the conference Mathematical Foundations of Computer Science (MFCS’99)
## 1 Introduction
The different semantics that can be assigned to a logic program correspond to different assumptions made concerning the atoms whose logical values cannot be inferred from the rules. For example, the well founded semantics corresponds to the assumption that every such atom is false (Closed World Assumption), while the Kripke-Kleene semantics corresponds to the assumption that every such atom is unknown. In general, the usual semantics of logic programs are given in the context of three-valued logics, and are of two kinds: those based on the stable models or on the well-founded semantics , and those based on the Kripke-Kleene semantics .
We refer to semantics of the first kind as $`pessimistic`$, in the sense that it privileges negative information: if in doubt, then assume false; and we refer to semantics of the second kind as $`skeptical`$, in the sense that it privileges neither negative nor positive information: if in doubt, then assume nothing. To illustrate these semantics, consider the following program:
$$𝒫\{\begin{array}{ccc}\text{charge(X)}\hfill & \hfill & \neg \text{innocent(X)}\text{suspect(X)}\hfill \\ \text{free}(𝒳)\hfill & \hfill & \text{innocent(X)}\text{suspect(X)}\hfill \\ \text{innocent(X)}\hfill & \hfill & \text{free}(𝒳)\hfill \\ \text{suspect(John)}\hfill & \hfill & \end{array}$$
The only assertion made in the program is that John is suspect, but we know nothing as to whether he is innocent.
If we follow the pessimistic approach, then we have to assume that John is not innocent, and we can infer that John must not be freed, and must be charged. If, on the other hand, we follow the skeptical approach, then we have to assume nothing about the innocence of John, and we can infer nothing as to whether he must be freed or charged.
However, in the context of three-valued logic, one can envisage a third semantics, that we shall call $`optimistic`$: if in doubt, then assume true. If we follow this approach, then we have to assume that John is innocent, and we can infer that John must be freed, and must not be charged.
Now, the optimistic approach can be seen as a counterpart of the pessimistic approach. To find a counterpart for the skeptical approach, one has to adopt a multi-valued logic. In such a logic, one can envisage an $`inconsistent`$ semantics: if in doubt, then assume both false and true. Table 1 summarizes the four possible semantics of $`𝒫`$, where $``$, $`𝒯`$, $`𝒰`$ and $``$ stand for false, true, unknown and inconsistent, respectively.
| Approach | suspect(John) | innocent(John) | free(John) | charge(John) |
| --- | --- | --- | --- | --- |
| Pessimistic | $`𝒯`$ | $``$ | $``$ | $`𝒯`$ |
| Optimistic | $`𝒯`$ | $`𝒯`$ | $`𝒯`$ | $``$ |
| Skeptical | $`𝒯`$ | $`𝒰`$ | $`𝒰`$ | $`𝒰`$ |
| Inconsistent | $`𝒯`$ | $``$ | $``$ | $``$ |
Table 1 - The four possible semantics of $`𝒫`$
In this paper, we define the semantics of a program $`𝒫`$ using a parameter $`\alpha `$ whose value can be any of the above four logical values. Once fixed, the value of $`\alpha `$ represents the “default value” for those atoms of $`𝒫`$ whose values cannot be inferred from the rules. We define a simple operator that allows us to compute this parameterized semantics, and also to compare and combine semantics obtained for different values of $`\alpha `$. We show that our semantics extends the semantics proposed by Fitting , and captures the usual semantics of conventional logic programs thereby unifying their computation. As a side-result, we propose a new semantics for logic programs, that can be roughly described as a “compromise” between pessimistic and optimistic semantics.
Motivation for this work comes from the area of knowledge acquisition, where contradictions may occur during the process of collecting knowledge from different experts. Indeed, in multi-agent systems, different agents may give different answers to the same query. It is then important to be able to process the answers so as to extract the maximum of information on which the various agents agree, or to detect the items on which the agents give conflicting answers.
Motivation also comes from the area of deductive databases. Updates leading to a certain degree of inconsistency should be allowed because inconsistency can lead to useful information, especially within the framework of distributed databases. In particular, Fuhr and Rölleke showed in that hypermedia retrieval requires the handling of inconsistent information.
The use of multi-valued logics is justified by the fact that it provides a more natural modeling framework for the application areas just mentioned. Moreover, as Arieli and Avron showed in , the use of four values is preferable to the use of three even for tasks that can in principle be handled using only three values.
The remaining of the paper is organized as follows. In section 2, we recall very briefly definitions and notations from three-valued and multi-valued logics, namely, stable models and well-founded semantics, Kripke-Kleene semantics, Belnap’s logic, bilattices and Fitting’s programs. We then proceed, in section 3, to define our parameterized semantics of a Fitting program $`𝒫`$. This is done by defining a parameterized operator whose fixpoints we call the $`\alpha `$-fixed models of $`𝒫`$. Our treatment in this section is inspired by . If the value of the parameter $`\alpha `$ is false, then the $`\alpha `$-fixed models correspond to the stable models proposed by Fitting. We also present an algorithm for computing the $`\alpha `$-fixed semantics of $`𝒫`$. In section 4, we restrict our attention to conventional logic programs. We show that their $`\alpha `$-fixed models capture the three-valued stable models, the well-founded semantics, and the Kripke-Kleene semantics. We also provide a comparative study of the $`\alpha `$-fixed models for the four values of the parameter $`\alpha `$, and propose a “compromise” between pessimistic and optimistic semantics that in certain cases may lead to the definition of a new semantics. Section 5 contains concluding remarks and suggestions for further research.
## 2 Preliminaries
### 2.1 Three-valued logics
#### 2.1.1 Stable models and well founded semantics
Gelfond and Lifschitz introduced the notion of stable model , in the framework of classical logic under the closed world assumption. This notion was then extended to three-valued logics and partial interpretations: Van Gelder, Ross and Schlipf introduced the well-founded semantics , and Przymusinski defined the three-valued stable models . In fact, as shown in , Przymusinski’s extension captures both the bi-valued stable models and the well-founded semantics.
In Przymusinski’s approach, a conjunctive logic program is a set of clauses of the form $`AB_1\mathrm{}B_n\neg C_1\mathrm{}\neg C_m`$, where $`B_1,\mathrm{},B_n,C_1,\mathrm{},C_m`$ are atoms. In this context, a valuation is a mapping that assigns to each ground atom a truth value from the set {$`false,unknown,true`$}. A valuation can be extended to ground litterals and conjunctions of ground litterals in the usual way. To define the stable models and well-founded semantics of a program $`𝒫`$, one uses the extended Gelfond-Lifschitz transformation $`GL_𝒫`$ which assigns to each valuation $`v`$ another valuation $`GL_𝒫(v)`$ defined as follows :
1. Transform $`𝒫`$ into a positive program $`𝒫_{/v}`$ by replacing all negative literals by their values from $`v`$.
2. Compute the least fixpoint of an immediate consequence operator $`\mathrm{\Phi }`$ defined as follows :
* if the ground atom $`A`$ is not in the head of any rule of Inst-$`𝒫_{/v}`$, then $`\mathrm{\Phi }_{𝒫_{/v}}(v)(A)=false`$; here, Inst-$`𝒫_{/v}`$ denotes the set of all instantiations of rules of $`𝒫_{/v}`$;
* if the rule “$`A`$” occurs in Inst-$`𝒫_{/v}`$, then $`\mathrm{\Phi }_{𝒫_{/v}}(v)(A)=true`$;
* else $`\mathrm{\Phi }_{𝒫_{/v}}(v)(A)=\{v(B)|AB`$ Inst-$`𝒫_{/v}`$, where $``$ is the extension of classical disjunction defined by:
| $`false`$ | $``$ | $`unknown`$ | $`=`$ | $`unknown;`$ |
| --- | --- | --- | --- | --- |
| $`true`$ | $``$ | $`unknown`$ | $`=`$ | $`true;`$ |
| $`unknown`$ | $``$ | $`unknown`$ | $`=`$ | $`unknown.`$ |
The valuation $`v`$ is defined to be a three-valued stable model of $`𝒫`$ if $`GL_𝒫(v)=v`$. The least three-valued stable model coincides with the well-founded semantics of $`𝒫`$, as defined by Van Gelder et als .
It follows from the definition of $`\mathrm{\Phi }`$ above that this approach gives greater importance to negative information, so it is a pessimistic approach.
#### 2.1.2 Kripke-Kleene semantics
Working with three-valued logic, Fitting introduced the Kripke-Kleene semantics . The program $`𝒫`$ has the same definition as for stable models, but the operator $`\mathrm{\Phi }`$ is now defined as follows :
given a valuation $`v`$ and a ground atom $`A`$ in Inst-$`𝒫`$,
* if there is a rule in Inst-$`𝒫`$ with head $`A`$, and the truth value of the body under $`v`$ is $`true`$, then $`\mathrm{\Phi }_𝒫(v)(A)=true`$;
* if there is a rule in Inst-$`𝒫`$ with head $`A`$, and for every rule in Inst-$`𝒫`$ with head $`A`$ the truth value of the body under $`v`$ is false, then $`\mathrm{\Phi }_𝒫(v)(A)=false`$;
* else $`\mathrm{\Phi }_𝒫(v)(A)=unknown`$.
It follows that this approach gives greater importance to the lack of information since $`unknown`$ is assigned to the atoms whose logical values cannot be inferred from the rules, so it is a skeptical approach.
### 2.2 Multi-valued logics
#### 2.2.1 Belnap’s logic
In , Belnap defines a logic called $`𝒪𝒰`$ intended to deal with incomplete and inconsistent information. Belnap’s logic uses four logical values, that we shall denote by $``$, $`𝒯`$, $`𝒰`$ and $``$ , i.e. $`𝒪𝒰`$ = {$``$, $`𝒯`$, $`𝒰`$, $``$}. These values can be compared using two orderings, the knowledge ordering and the truth ordering.
In the knowledge ordering, denoted by $`_k`$, the four values are ordered as follows: $`𝒰`$ $`_k`$ $``$, $`𝒰`$ $`_k`$ $`𝒯`$, $``$ $`_k`$ $``$, $`𝒯`$ $`_k`$ $``$. Intuitively, according to this ordering, each value of $`𝒪𝒰`$ is seen as a possible knowledge that one can have about the truth of a given statement. More precisely, this knowledge is expressed as a set of classical truth values that hold for that statement. Thus, $``$ is seen as {$`false`$}, $`𝒯`$ is seen as {$`true`$}, $`𝒰`$ is seen as $`\mathrm{}`$ and $``$ is seen as {$`false`$, $`true`$}. Following this viewpoint, the knowledge ordering is just the set inclusion ordering.
In the truth ordering, denoted by $`_t`$, the four logical values are ordered as follows: $``$ $`_t`$ $`𝒰`$, $``$ $`_t`$ $``$, $`𝒰`$ $`_t`$ $`𝒯`$, $``$ $`_t`$ $`𝒯`$. Intuitively, according to this ordering, each value of $`𝒪𝒰`$ is seen as the degree of truth of a given statement. $`𝒰`$ and $``$ are both less false than $``$, and less true than $`𝒯`$, but $`𝒰`$ and $``$ are not comparable.
The two orderings are represented in the double Hasse diagram of Figure 1.
Each of the orderings $`_t`$ and $`_k`$ gives $`𝒪𝒰`$ a lattice structure. Meet and join under the truth ordering are denoted by $``$ and $``$, and they are natural generalizations of the usual notions of conjunction and disjunction. In particular, $`𝒰`$$``$$``$= $``$ and $`𝒰`$$``$$``$= $`𝒯`$. Under the knowledge ordering, meet and join are denoted by $``$ and $``$, and are called the $`consensus`$ and $`gullibility`$, respectively:
* $`xy`$ represents the maximal information on which $`x`$ and $`y`$ agree, whereas
* $`xy`$ adds the knowledge represented by $`x`$ to that represented by $`y`$.
In particular, $``$$``$$`𝒯`$= $`𝒰`$ and $``$$``$$`𝒯`$= $``$.
There is a natural notion of $`negation`$ in the truth ordering denoted by $`\neg `$, for which we have: $`\neg `$ $`𝒯`$= $``$, $`\neg `$ $``$= $`𝒯`$, $`\neg `$ $`𝒰`$= $`𝒰`$, $`\neg `$ $``$= $``$. There is a similar notion for the knowledge ordering, called $`conflation`$, denoted by -, for which: - $`𝒰`$= $``$, - $``$= $`𝒰`$, - $``$= $``$, - $`𝒯`$= $`𝒯`$.
The operations $`,,\neg `$ restricted to the values $`𝒯`$ and $``$ are those of classical logic, and if we add to these operations and values the value $`𝒰`$, then they are those of Kleene’s strong three-valued logic.
#### 2.2.2 Bilattices
In , bilattices are used as truth-value spaces for integration of information coming from different sources. The bilattice approach is a basic contribution to many-valued logics. Bilattices and their derived sublogics are useful in expressing uncertainty and inconsistency in logic programming and databases . The simplest non-trivial bilattice is called FOUR, and it is basically Belnap’s four-valued logic .
###### Definition 1
A bilattice is a triple $`,_t,_k`$, where $``$ is a nonempty set and $`_t`$, $`_k`$ are each a partial ordering giving $``$ the structure of a lattice with a top and a bottom.
In a bilattice $`,_t,_k`$, meet and join under $`_t`$ are denoted $``$ and $``$, and meet and join under $`_k`$ are denoted $``$ and $``$. Top and bottom under $`_t`$ are denoted $`𝒯`$ and $``$, and top and bottom under $`_k`$ are denoted $``$ and $`𝒰`$. If the bilattice is complete with respect to both orderings, infinitary meet and join under $`_t`$ are denoted $``$ and $``$, and infinitary meet and join under $`_k`$ are denoted $``$ and $``$.
###### Definition 2
A bilattice $`,_t,_k`$ is called distributive if all 12 distributive laws connecting $``$, $``$, $``$ and $``$ hold. It is called infinitely distributive if it is a complete bilattice in which all infinitary, as well as finitary, distributive laws hold.
An example of a distributive law is $`x(yz)=(xy)(xz)`$. An example of an infinitary distributive law is $`x\{y_i|iS\}=\{xy_i|iS\}`$.
###### Definition 3
A bilattice $`,_t,_k`$ satisfies the interlacing conditions if each of the operations $``$, $``$, $``$ and $``$ is monotone with respect to both orderings. If the bilattice is complete, it satisfies the infinitary interlacing conditions if each of the infinitary meet and join is monotone with respect to both orderings.
An example of an interlacing condition is: $`x_1_ty_1`$ and $`x_2_ty_2`$ implies $`x_1x_2_ty_1y_2`$. An example of an infinitary interlacing condition is: $`x_i_ty_i`$ for all $`iS`$ implies $`\{x_i|iS\}_t\{y_i|iS\}`$. A distributive bilattice satisfies the interlacing conditions.
$`𝒪𝒰`$ is an infinitary distributive bilattice which satisfies the infinitary interlacing laws. A bilattice is said to be nontrivial if the bilattice FOUR can be isomorphically embedded in it.
A way for constructing a bilattice is proposed in . Consider two lattices $`L_1,_1`$ and $`L_2,_2`$. We can see $`L_1`$ as the set of values used for representing the degree of belief (evidence, confidence, etc.) of an information and $`L_2`$ as the set of values used for representing the degree of doubt (counter-evidence, lack of confidence, etc.) of the information.
The structure $`L_1\times L_2,_t,_k`$ where:
* $`x,y_tz,w`$ iff $`xz`$ and $`wy`$,
($`x,yz,w`$ = $`min(x,z),max(y,w)`$), and
* $`x,y_kz,w`$ iff $`xz`$ and $`yw`$
($`x,yz,w`$ = $`min(x,z),min(y,w)`$)
is a bilattice satisfying the interlacing conditions; it also satisfies the infinitary interlacing conditions if $`L_1`$ and $`L_2`$ are complete. Moreover, it is infinitely distributive if $`L_1`$ and $`L_2`$ are complete and infinitely distributive.
By abuse of notation we will sometimes talk about the bilattice $``$ when the orders are irrelevant or understood from the context. From now on, we assume that the bilattices we use are infinitely distributive, satisfy the infinitary interlacing conditions and have a negation unless explicitly stated otherwise.
#### 2.2.3 Fitting programs
Conventional logic programming has the set {$``$, $`𝒯`$} as its intended space of truth values, but since not every query may produce an answer, partial models are often allowed (i.e. $`𝒰`$ is added). If we want to deal with inconsistency as well, then $``$ must be added. Thus Fitting asserts that $`𝒪𝒰`$ can be thought as the “home” of ordinary logic programming and extends the notion of logic program so that a bilattice $``$ other than $`𝒪𝒰`$ can be thought of as the space of truth values.
###### Definition 4
(Fitting program)
* A formula is an expression built up from literals and elements of $``$, using $`,,,,,`$.
* A clause is of the form $`P(x_1,\mathrm{},x_n)\varphi (x_1,\mathrm{},x_n)`$, where the atomic formula $`P(x_1,\mathrm{},x_n)`$ is the head, and the formula $`\varphi (x_1,\mathrm{},x_n)`$ is the body. It is assumed that the free variables of the body are among $`x_1,\mathrm{},x_n`$.
* A program is a finite set of clauses with no predicate letter appearing in the head of more than one clause (this apparent restriction causes no loss of generality ).
We shall refer to such an extended logic program as a Fitting program. Fitting also defined the family of conventional logic programs. A conventional logic program is one whose underlying truth-value space is the bilattice $`𝒪𝒰`$ and which does not involve $`,,,𝒰,`$. Such programs can be written in the customary way, using commas to denote conjunction.
## 3 Parameterized semantics for Fitting programs
In the following, $`\alpha 𝒪𝒰`$, $`𝒫`$ is a Fitting program, $`𝒱`$($``$) is the set of all valuations in $``$ and Inst-$`𝒫`$ is the set of all ground instances of rules of $`𝒫`$. Some of the results in this section are inspired by which deals only with the case $`\alpha `$ = $``$.
### 3.1 Immediate Consequence Operators
First, we extend the two orderings on $`𝒪𝒰`$ to the space of valuations $`𝒱()`$.
###### Definition 5
Let $`v_1`$ and $`v_2`$ be in $`𝒱`$($``$), then
* $`v_1_tv_2`$ if and only if $`v_1(A)_tv_2(A)`$ for all ground atoms $`A`$;
* $`v_1_kv_2`$ if and only if $`v_1(A)_kv_2(A)`$ for all ground atoms $`A`$.
Under these two orderings $`𝒱`$($``$) becomes a bilattice, and we have $`(vw)(A)=v(A)w(A)`$, and similarly for the other operators. $`𝒱`$($``$) is infinitely distributive, satisfies the infinitely interlacing conditions and has a negation and a conflation.
The actions of valuations can be extended from atoms to formulas as follows:
* $`v(XY)=v(X)v(Y)`$, and similarly for the other operators,
* $`v((x)\varphi (x))=_{t=closedterm}v(\varphi (t))`$, and
* $`v((x)\varphi (x))=_{t=closedterm}v(\varphi (t))`$.
The predicate $`equal(x,y)`$ is a predefined predicate defined by: for all valuations $`v`$,
* $`v(equal(x,y))=𝒯`$ if $`x=y`$,
* $`v(equal(x,y))=`$$``$ if $`xy`$, and
* $`v(\beta )=\beta `$ for all $`\beta `$ in $``$.
The following contrajoin operation assigns a truth value to a ground atom $`A`$ independently of the truth value assigned to the negation of $`A`$.<sup>2</sup><sup>2</sup>2Our contrajoin operation is exactly the same as pseudovaluation in . However, we prefer the term contrajoin of v and w as it is more indicative of the fact that an operation is performed on valuations v and w.
###### Definition 6 (contrajoin)
Let v and w be in $`𝒱`$($``$).
The contrajoin of $`v`$ and $`w`$, denoted $`vw`$, is defined as follows:
v$``$w(A)=v(A) and v$``$w($`\neg `$A)=$`\neg `$w(A), for each ground atom $`A`$.
Contrajoin operations are extended to formulas by induction. The idea is that $`v`$ represents the information about $`A`$, and $`w`$ the information about $`\neg A`$. For example, if $`v(innocent(John))`$ = $`𝒯`$ $`and`$ $`w(innocent(John))`$ = $`𝒰`$ then $`vw(innocent(John))`$ = $`𝒯`$, whereas $`\neg (vw(\neg innocent(John))`$ = $`𝒰`$.
We can now define a new operator $`\mathrm{\Psi }_𝒫^\alpha `$ which is inspired by . It infers new information from a contrajoin operation in a way that depends on the value of the parameter $`\alpha `$.
###### Definition 7
Let v and w be in $`𝒱`$($``$). The valuation $`\mathrm{\Psi }_𝒫^\alpha (v,w)`$ is defined as follows:
1. if the ground atom $`A`$ is not the head of any rule of Inst-$`𝒫`$, then $`\mathrm{\Psi }_𝒫^\alpha (v,w)(A)=\alpha `$
2. if A $``$ B occurs in Inst-$`𝒫`$, then $`\mathrm{\Psi }_𝒫^\alpha (v,w)(A)=vw(B)`$.
Clearly, the valuation $`\mathrm{\Psi }_𝒫^\alpha (v,w)`$ is in $`𝒱()`$, and as the interlacing conditions are satisfied by $`𝒱()`$, we can prove the following proposition.
###### Proposition 1
Let $`𝒫`$ be a Fitting program.
(1) Under the knowledge ordering, $`\mathrm{\Psi }_𝒫^\alpha `$ is monotonic in both arguments;
(2) Under the truth ordering, $`\mathrm{\Psi }_𝒫^\alpha `$ is monotonic (and moreover continuous) in its first argument, and anti-monotonic in its second argument.
Proof.The proof makes use of the following lemma which is an immediate consequence of the definition of contrajoin.
###### Lemma 1
Let $`v_1,v_2,w_1,w_2𝒱()`$. We have:
(1) if $`v_1_kv_2`$ and $`w_1_kw_2`$, then $`v_1w_1_kv_2w_2`$;
(2) if $`v_1_tv_2`$ and $`w_2_tw_1`$, then $`v_1w_1_tv_2w_2`$;
Now, suppose $`v_1_kv_2`$ and let $`A`$ be a ground atom. We want to show that $`\mathrm{\Psi }_𝒫^\alpha (v_1,w)(A)_k\mathrm{\Psi }_𝒫^\alpha (v_2,w)(A)`$. If A does not occur as the head of any member of Inst-$`𝒫`$, then $`\mathrm{\Psi }_𝒫^\alpha (v_1,w)(A)=\mathrm{\Psi }_𝒫^\alpha (v_2,w)(A)=\alpha `$. If $`AB`$ Inst-$`𝒫`$, then $`\mathrm{\Psi }_𝒫^\alpha (v_1,w)(A)=v_1w(B)`$, and similarly for $`v_2`$, so, by part one of the previous lemma, $`\mathrm{\Psi }_𝒫^\alpha (v_1,w)(A)_k\mathrm{\Psi }_𝒫^\alpha (v_2,w)(A)`$. The proof of the monotonicity in the second argument is similar. Item (2) of Proposition 1 is established by a similar argument using part 2 of the lemma.
Before we continue, we recall that according to the Knaster-Tarski theorem, a monotone operator $`f`$ on a complete lattice $`L`$ has a least fixpoint $`l`$ and a greatest fixpoint $`g`$. There are two ways of constructing these fixpoints, and each leads to a technique for proving certain properties.
Following the first way, the least fixpoint of $`f`$ is shown to be $`\{xL|f(x)x\}`$. It follows that if $`f(x)x`$, then $`lx`$. The greatest fixpoint of $`f`$ is shown to be $`\{xL|xf(x)\}`$. It follows that if $`xf(x)`$, then $`xg`$.
Following the second way, one produces a (generally transfinite) sequence of members of $`L`$ as follows: $`f_0`$ is the least member of $`L`$. For an ordinal $`n`$, $`f_{n+1}`$ is set to be $`f(f_n)`$, and for a limit ordinal $`\lambda `$, $`f_\lambda `$ is set to be $`_{n<\lambda }f_n`$. The limit of this sequence is the least fixpoint of $`f`$. This yields another method of proof: by transfinite induction. If it can be shown that each member of the sequence $`f_n`$ has some property, then the least fixpoint $`l`$ also has the property. For the greatest fixpoint, we construct a similar sequence: $`f_0`$ is the greatest member of $`L`$. For an ordinal $`n`$, $`f_{n+1}`$ is set to be $`f(f_n)`$, and for a limit ordinal $`\lambda `$, $`f_\lambda `$ is set to be $`_{n<\lambda }f_n`$.
It follows from Proposition 1 that the function $`\lambda x.\mathrm{\Psi }_𝒫^\alpha (x,v)`$ has a least fixpoint and a greatest fixpoint for each ordering. We define now a new operator $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ which associates each valuation $`v`$ with one of these fixpoints depending on the value of $`\alpha `$. $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v)`$ is the iterated fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^\alpha (x,v)`$ obtained from an initial valuation $`v_\alpha `$ defined by: $`v_\alpha (A)=\alpha `$, for all ground atoms $`A`$.
###### Definition 8
Let v be in $`𝒱()`$. Define $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v)`$ to be the limit of the sequence of valuations ($`a_n`$) defined as follows:
* $`a_0=v_\alpha `$;
* $`a_n=\mathrm{\Psi }_𝒫^\alpha (a_{n1},v)`$, for a successor ordinal n;
* $`a_\lambda =\{\begin{array}{ccc}_{n<\lambda }\mathrm{\Psi }_𝒫^\alpha (a_n,v)\text{ for }\alpha =\hfill & & \\ _{n<\lambda }\mathrm{\Psi }_𝒫^\alpha (a_n,v)\text{ for }\alpha =𝒯\hfill & & \\ _{n<\lambda }\mathrm{\Psi }_𝒫^\alpha (a_n,v)\text{ for }\alpha =𝒰\hfill & & \\ _{n<\lambda }\mathrm{\Psi }_𝒫^\alpha (a_n,v)\text{ for }\alpha =\hfill & & \end{array}`$ , for a limit ordinal $`\lambda `$.
In fact, we fix the truth value of negative literals with $`v`$, then we compute the semantics of the positive program thus obtained (in a similar manner to that of Gelfond-Lifschitz transformation).
We remark that $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}(v)`$ is the least fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,v)`$ and $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}`$ the greatest fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^{}(x,v)`$ under the knowledge ordering. $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}(v)`$ is the least fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^{}(x,v)`$ and $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒯}(v)`$ the greatest fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^𝒯(x,v)`$ under the truth ordering.
To illustrate this definition consider the following program $`𝒫`$ and let $`v`$ be the valuation which assigns to every ground atom the truth value $`𝒰`$:
$`𝒫\{\begin{array}{ccc}𝒜\hfill & \hfill & 𝒞\hfill \\ 𝒟\hfill & \hfill & \neg 𝒯\hfill \\ \hfill & \hfill & 𝒜\neg 𝒟\hfill \\ \hfill & \hfill & 𝒯\hfill \end{array}`$ Atom A B C D E $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}(v)`$ $``$ $`𝒯`$ $``$ $`𝒯`$ $`𝒰`$
To compute $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}(v)`$, we first replace all negative literals by the value $`𝒰`$, then we compute the least model of the positive program thus obtained (with respect to the truth ordering) beginning with the valuation which assigns to every ground atom the truth value $``$.
### 3.2 The family of $`\alpha `$-fixed models
We recall that a valuation $`v`$ is a model of a program $`𝒫`$ if and only if for all rules $`AB`$ in Inst-$`𝒫`$, $`v(A)_tv(B)`$ . By definition of $`\mathrm{\Psi }_𝒫^\alpha `$, a valuation $`v`$ that verifies $`\mathrm{\Psi }_𝒫^\alpha (v,v)=v`$ is a model of $`𝒫`$. Now, every fixpoint $`m`$ of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ verifies $`\mathrm{\Psi }_𝒫^\alpha (m,m)=\mathrm{\Psi }_𝒫^\alpha (\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(m),m)=\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(m)=m`$, therefore $`m`$ is a model of $`𝒫`$. So we can define four new families of models that we shall call $`\alpha `$-fixed models.
###### Definition 9 ($`\alpha `$-fixed models)
A valuation $`v𝒱()`$ is a $`\alpha `$-fixed model of a program $`𝒫`$ if and only if $`v`$ is a fixpoint of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$.
From now on, $``$-fixed models will be called $`pessimistic`$, $`𝒯`$-fixed models $`optimistic`$, $`𝒰`$-fixed models $`skeptical`$, and $``$-fixed models $`inconsistent`$. We can now study the family of $`\alpha `$-fixed models.
###### Theorem 1
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ is monotonic under $`_k`$, and anti-monotonic under $`_t`$.
Proof. Suppose $`v_1_kv_2`$. We want to show $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v_1)_k\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v_2)`$. Consider $`\alpha =`$. We define two transfinite sequences of valuations $`a_n`$ and $`b_n`$ as follows: $`a_0=b_0`$ is the always $`false`$ valuation, the least in the truth ordering; for all $`n+1`$ successor ordinals, $`a_{n+1}=\mathrm{\Psi }_𝒫^\alpha (a_n,v_1)`$ and $`b_{n+1}=\mathrm{\Psi }_𝒫^\alpha (b_n,v_2)`$; for a limit ordinal $`\lambda `$, $`a_\lambda =_{n<\lambda }a_n`$ and $`b_\lambda =_{n<\lambda }b_n`$. Both sequences are increasing in the truth ordering since $`\mathrm{\Psi }_𝒫^\alpha `$ is monotonic in its first argument. The sequence $`a_n`$ has $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v_1)`$ as its limit, while the sequence $`b_n`$ has $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v_2)`$ as its limit, so it is enough to establish that $`a_n_kb_n`$ for every ordinal $`n`$.
If $`n=0`$, $`a_0=b_0`$.
Suppose $`a_n_kb_n`$. Then $`a_{n+1}=\mathrm{\Psi }_𝒫^\alpha (a_n,v_1)_k\mathrm{\Psi }_𝒫^\alpha (b_n,v_2)=b_{n+1}`$, using the monotonicity of $`\mathrm{\Psi }_𝒫^\alpha `$ in both arguments under $`_k`$.
Finally, suppose $`a_n_kb_n`$ for every $`n<\lambda `$. $`𝒱()`$ satisfies the infinitary interlacing conditions so $`_{n<\lambda }a_n_k_{n<\lambda }b_n`$.
The result for $`\alpha =𝒯,𝒰,`$ is established similarly by replacing respectively $`a_0=b_0`$ (the valuation always $`false`$) by the valuation always $`true`$, always $`unknown`$, always $`inconsistent`$ and $``$ by $`,`$ and $``$, respectively.
Anti-monotonicity under the truth ordering is established by a similar argument.
Given the monotonicity of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ under the knowledge ordering and the complete lattice structure of $`𝒱()`$ under this ordering, we can apply the Knaster-Tarski theorem, and we obtain the following result:
###### Theorem 2
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ has a least fixpoint , denoted $`Fix_𝒰^\alpha `$, and a greatest fixpoint, denoted $`Fix_{}^\alpha `$, with respect to the knowledge ordering.<sup>3</sup><sup>3</sup>3Actually, $`Fix_𝒰^\alpha `$ and $`Fix_{}^\alpha `$ refer both to program $`𝒫`$, and should be denoted as $`Fix_{𝒫,𝒰}^\alpha `$ and $`Fix_{𝒫,}^\alpha `$, respectively. However, in order to simplify the presentation, we shall omit $`𝒫`$ in our notations.
We can remark that the computation of $`Fix_𝒰^\alpha `$, that we call $`\alpha `$-fixed semantics, is similar to the computation of the well-founded semantics via the Gelfond-Lifschitz transformation.
Four different semantics can now be associated to a Fitting program, one for each value of $`\alpha `$. The following example shows how these semantics can be used to provide different contexts, depending on the requirements.
Example. Let $`𝒫`$ be the following program:
$`𝒫\{\begin{array}{ccc}𝒞𝓁𝓁𝒶𝓊(𝒳,𝒴)\hfill & \hfill & 𝒞𝓁𝓁𝒶𝓊(𝒴,𝒳)\hfill \\ 𝒞𝓁𝓁𝒶𝓊(𝒶,𝒷)\hfill & \hfill & 𝒯\hfill \\ 𝒞𝓁𝓁𝒶𝓊(𝒶,𝒸)\hfill & \hfill & \hfill \end{array}`$
If we have to send information to persons that we are sure to be colleagues of $`b`$, we have to choose the pessimistic or skeptical semantics. Indeed, under this semantics, the only person that can be proved to be a colleague of $`b`$ is $`a`$.
Now, if we want to send information to persons that may be colleagues of $`b`$, then we have to choose the optimistic semantics. There are two persons that are or may be colleagues of $`b`$ : $`a`$ and $`c`$. The following table summarizes the results.
Semantics Coll(a,b) Coll(b,a) Coll(a,c) Coll(c,a) Coll(b,c) Coll(c,b) $`Fix_𝒰^{}`$ $`𝒯`$ $`𝒯`$ $``$ $``$ $``$ $``$ $`Fix_𝒰^𝒯`$ $`𝒯`$ $`𝒯`$ $``$ $``$ $`𝒯`$ $`𝒯`$ $`Fix_𝒰^𝒰`$ $`𝒯`$ $`𝒯`$ $``$ $``$ $`𝒰`$ $`𝒰`$ $`Fix_𝒰^{}`$ $`𝒯`$ $`𝒯`$ $``$ $``$ $``$ $``$
The behavior of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ with respect to the truth ordering is less simple because $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ is anti-monotonic under this ordering. However, there is a modification of the Knaster-Tarski theorem dealing with precisely this case:
###### Lemma 2 ()
Suppose that a function f is anti-monotonic on a complete lattice $``$. Then there are two elements $`\mu `$ and $`\nu `$ of $``$, called extreme oscillation points of f, such that the following hold:
\- $`\mu `$ and $`\nu `$ are the least and greatest fixpoint of $`f^2`$ (i.e. of $`f`$ composed with itself);
\- $`f`$ oscillates between $`\mu `$ and $`\nu `$ in the sense that $`f(\mu )=\nu `$ and $`f(\nu )=\mu `$;
\- if $`x`$ and $`y`$ are also elements of $``$ between which $`f`$ oscillates then $`x`$ and $`y`$ lie between $`\mu `$ and $`\nu `$.
As $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ is anti-monotonic and $`𝒱()`$ is a complete lattice under the truth ordering, it follows that $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ has two extreme oscillation points under this ordering:
###### Proposition 2
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ has two extreme oscillation points denoted $`Fix_{}^\alpha `$ and $`Fix_𝒯^\alpha `$, with $`Fix_{}^\alpha _tFix_𝒯^\alpha `$, under the truth ordering.
We can now extend the result of to any value of $`𝒪𝒰`$.
###### Theorem 3
Let $`𝒫`$ be a Fitting program. Then we have:
| $`Fix_𝒰^\alpha `$ | = | $`Fix_{}^\alpha `$ | $``$ | $`Fix_𝒯^\alpha `$ |
| --- | --- | --- | --- | --- |
| $`Fix_{}^\alpha `$ | = | $`Fix_{}^\alpha `$ | $``$ | $`Fix_𝒯^\alpha `$ |
| $`Fix_{}^\alpha `$ | = | $`Fix_𝒰^\alpha `$ | $``$ | $`Fix_{}^\alpha `$ |
| $`Fix_𝒯^\alpha `$ | = | $`Fix_𝒰^\alpha `$ | $``$ | $`Fix_{}^\alpha `$ |
Proof.The proof of this theorem is given in the Appendix.
The family of $`\alpha `$-fixed models of a program is bounded for each $`\alpha 𝒪𝒰`$ as follows: in the knowledge ordering, all $`\alpha `$-fixed models are between $`Fix_𝒰^\alpha `$ and $`Fix_{}^\alpha `$ which are the least and greatest $`\alpha `$-fixed models, respectively; in the truth ordering, all $`\alpha `$-fixed models are between $`Fix_{}^\alpha `$ and $`Fix_𝒯^\alpha `$ which are not necessarily $`\alpha `$-fixed models of $`𝒫`$.
It is interesting to note that for $`\alpha =`$ the first equality of Theorem 3 relates two different definitions of the well-founded semantics: the left-hand side, $`Fix_𝒰^\alpha `$, represents the definition of Przymusinski via three-valued stable models, whereas the right-hand side, $`Fix_{}^\alpha Fix_𝒯^\alpha `$, represents the definition of Van Gelder via alternating fixpoints . Working with bilattices, Fitting generalized the approach of Van Gelder in and that of Przymusinski in .
### 3.3 An algorithm for computing $`\alpha `$-fixed semantics
In this section, all literals are ground literals.
An interpretation $`=(T,F)`$ is a pair of sets of atoms where $`T`$ is the set of atoms considered as true and $`F`$ the set of atoms considered as false. The logical value of an atom $`A`$ with respect to $``$ is :
* $`𝒯`$ if $`AT`$ and $`AF`$,
* $``$ if $`AT`$ and $`AF`$,
* $`𝒰`$ if $`AT`$ and $`AF`$, and
* $``$ if $`AT`$ and $`AF`$.
A pseudo-interpretation $`𝒥=(T,F,T^{},F^{})`$ is composed of four sets of atoms and assigns to every literal $`L`$ a logical value as follows:
-if $`L`$ is a ground atomic formula of the form $`R(v_1,\mathrm{},v_n)`$ then its logical value with respect to $`𝒥`$ is the logical value of $`R(v_1,\mathrm{},v_n)`$ with respect to the interpretation $`(T,F)`$;
-if $`L`$ is a ground atomic formula of the form $`\neg R(v_1,\mathrm{},v_n)`$ then its logical value with respect to $`𝒥`$ is the negation of the logical value of $`R(v_1,\mathrm{},v_n)`$ with respect to the interpretation $`(T^{},F^{})`$;
The logical value of a formula with respect to a pseudo-interpretation $`𝒥`$ is given by the logical value of its literals with respect to $`𝒥`$ and the truth tables of the different operators.
The following algorithm uses a bottom-up approach to compute the $`\alpha `$-fixed semantics of a ground Fitting program $`𝒫`$ with no function symbol over the bilattice $`𝒪𝒰`$.
Algorithm: $`\alpha `$-fixed semantics
| 1. | begin |
| --- | --- |
| 2. | | Res\_True := $`\mathrm{}`$; |
| 3. | | Res\_False := $`\mathrm{}`$; |
| 4. | | Tmp\_Res := ({$`<>`$ },{ $`<>`$ }); |
| 5. | | match $`\alpha `$ with |
| 6. | | | $`\alpha `$ = $`𝒯`$ -$`>`$ |
| | | | | Init\_True := $`_𝒫`$; |
| | | | | Init\_False := $`\mathrm{}`$; |
| | | | | Not\_Head\_True := { all atoms in $`_𝒫`$ which are not heads of any rule in $`𝒫`$ }; |
| | | | | Not\_Head\_False := $`\mathrm{}`$; |
| 7. | | | $`\alpha `$ = $``$ -$`>`$ |
| | | | | Init\_True := $`\mathrm{}`$; |
| | | | | Init\_False := $`_𝒫`$; |
| | | | | Not\_Head\_True := $`\mathrm{}`$ ; |
| | | | | Not\_Head\_False := { all atoms in $`_𝒫`$ which are not heads of any rule in $`𝒫`$ }; |
| 8. | | | $`\alpha `$ = $``$ -$`>`$ |
| | | | | Init\_True := $`_𝒫`$; |
| | | | | Init\_False := $`_𝒫`$; |
| | | | | Not\_Head\_True := { all atoms in $`_𝒫`$ which are not heads of any rule in $`𝒫`$ } ; |
| | | | | Not\_Head\_False := { all atoms in $`_𝒫`$ which are not heads of any rule in $`𝒫`$ }; |
| 9. | | | $`\alpha `$ = $`𝒰`$ -$`>`$ |
| | | | | Init\_True := $`\mathrm{}`$; |
| | | | | Init\_False := $`\mathrm{}`$; |
| | | | | Not\_Head\_True := $`\mathrm{}`$ ; |
| | | | | Not\_Head\_False := $`\mathrm{}`$; |
| 10. | | while Tmp\_Res $``$ (Res\_True,Res\_False) do |
| 11. | | | Tmp\_Res $`=`$ (Res\_True,Res\_False); |
| 12. | | | Iter\_True := Init\_True; |
| 13. | | | Iter\_False := Init\_False; |
| 14. | | | Tmp\_Iter := ({$`<>`$ },{$`<>`$ }); |
| 15. | | | while Tmp\_Iter $``$ (Iter\_True,Iter\_False) do |
| 16. | | | | Tmp\_Iter $`=`$ (Iter\_True,Iter\_False); |
| 17. | | | | Im\_$`𝒯`$ := $`\mathrm{}`$; |
| 18. | | | | Im\_$``$ := $`\mathrm{}`$; |
| 19. | | | | Im\_$``$ := $`\mathrm{}`$; |
| 20. | | | | for all clauses $`C`$ in $`𝒫`$ match the logical value $`l`$ of the body of $`C`$ |
| | | | | with respect to the pseudo-interpretation |
| | | | | (Iter\_True, Iter\_False, Res\_True, Res\_False) |
| | | | | with |
| 21. | | | | | $`l=𝒯`$ -$`>`$ Im\_$`𝒯`$ := Im\_$`𝒯`$ $``$ { head($`C`$)} |
| 22. | | | | | $`l=`$ -$`>`$ Im\_$``$ := Im\_$``$ $``$ { head($`C`$)} |
| 23. | | | | | $`l=`$ -$`>`$ Im\_$``$ := Im\_$``$ $``$ { head($`C`$)} |
| 24. | | | | end for |
| 25. | | | | Iter\_True := Im\_$`𝒯`$ $``$ Im\_$``$ $``$ Not\_Head\_True; |
| 26. | | | | Iter\_False := Im\_$``$ $``$ Im\_$``$ $``$ Not\_Head\_False; |
| 27. | | | end while |
| 28. | | | Res\_True := Res\_True $``$ Iter\_True; |
| 29. | | | Res\_False := Res\_False $``$ Iter\_False; |
| 30. | | end while |
| 31. | | return (Res\_True,Res\_False); |
| 32. | end. |
Intuitively, the assignment of the logical value $`\alpha `$ to the atoms which are not heads of any rule is done through the sets of atoms Not\_Head\_True and Not\_Head\_False. The value of $`\alpha `$ also determines the initial value, (Init\_True, Init\_False) of the iterated computation of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(v)`$ performed by the while loop (lines 15 to 27). Here $`v`$ corresponds to the interpretation (Res\_True,Res\_False), and (Iter\_True, Iter\_False) to the value of a step of this computation. The first while loop (lines 10 to 30) calculates the sequence of iterated values of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ with $`𝒰`$ as initial value, and having $`Fix^\alpha `$ as limit.
This algorithm could be easily modified in order to verify if an interpretation is a $`\alpha `$-fixed model of a Fitting program $`𝒫`$.
## 4 Comparing the usual semantics of logic programs
In this section, we compare the $`\alpha `$-fixed models of conventional logic programs with the usual semantics, then we compare the different usual semantics among them.
The following theorem states that the family of stable models is included in the family of pessimistic fixed models (thus extending stable models from conventionnal logic programs to Fitting programs), and that the well-founded semantics and the Kripke-Kleene semantics are captured (and similarly extended) by our appproach.
###### Theorem 4
Let $`𝒫`$ be a conventional logic program.
(1) If v is a three-valued stable model of $`𝒫`$, then v is a pessimistic fixed model.
(2) If v is the well-founded semantics of $`𝒫`$, then v = $`Fix_𝒰^{}`$;
(3) If v is the Kripke-Kleene semantics of $`𝒫`$, then v= $`Fix_𝒰^𝒰`$.
Proof. The Gelfond-Lifschitz transformation $`GL_𝒫`$ is divided in two steps: firstly, it transforms the program $`𝒫`$ in a positive program $`𝒫_{/v}`$ by replacing negative literals by their value in the valuation $`v`$; then, it applies to this program the immediate consequence operator $`\mathrm{\Phi }_{𝒫}^{}{}_{/v}{}^{}`$. The valuation $`v`$ is a stable model if and only if $`GL_𝒫(v)=v`$.
We have
$`\mathrm{\Phi }_{𝒫}^{}{}_{/v}{}^{}(w)=\mathrm{\Psi }_𝒫^{}(w,v)`$,
so,
$`lfp_t\lambda w.\mathrm{\Phi }_{𝒫}^{}{}_{/v}{}^{}(w)=lfp_t\lambda w.\mathrm{\Psi }_𝒫^{}(w,v)`$.
Thus
$`GL_p=\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}`$
so, if $`v`$ is a stable model of $`𝒫`$, then it is a fixpoint of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}`$ and consequently, a pessimistic fixed model of $`𝒫`$.
Thus, (1) is established and (2) is immediate with this proof because the well-founded semantics of $`𝒫`$ and $`Fix_𝒰^{}`$ are the least fixpoints under the truth ordering of $`GL_𝒫`$ and $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{}`$, respectivly.
Concerning (3), we have $`\mathrm{\Psi }_𝒫^𝒰(v,v)=\mathrm{\Phi }_𝒫(v)`$ where $`\mathrm{\Phi }_𝒫`$ is the Kripke-Kleene operator. Let $`K_𝒫`$ be the Kripke-Kleene semantics, then we have
$`K_𝒫=lfp_k\lambda x.\mathrm{\Phi }_𝒫(x)=lfp_k\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,x)`$
Now, $`Fix_𝒰^𝒰`$ is a fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,x)`$,
so $`K_𝒫_kFix_𝒰^𝒰`$.
In the other direction, we have $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}(K_𝒫)=lfp_k(\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,K_𝒫)`$
Now, $`K_𝒫`$ is a fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,K_𝒫)`$, so $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}(K_𝒫)_kK_𝒫`$.
As $`Fix_𝒰^𝒰`$ is the least fixpoint of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}`$, we have $`Fix_𝒰^𝒰_KK_𝒫`$.
It is important to recall here that, in our approach, positive and negative information are treated separately during the computation of $`Fix_𝒰^𝒰`$. This is not the case with the computation of Kripke-Kleene semantics. Nevertheless, when we restrict our attention to conventional programs, the two methods compute the same semantics. Our approach unifies the computation of usual semantics, and thus allows us to compare them.
###### Theorem 5
Let $`𝒫`$ be a Fitting program. Then we have:
$`Fix_𝒰^𝒰`$ $`_k`$ $`Fix_𝒰^{}`$ and $`Fix_𝒰^𝒰`$ $`_k`$ $`Fix_𝒰^𝒯`$.
Proof. Let $`A`$ be a ground atom. If $`A`$ does not occurs as the head of any rule in Inst-$`𝒫`$, then
$`\mathrm{\Psi }_𝒫^𝒰(Fix_𝒰^{},Fix_𝒰^{})(A)=𝒰_kFix_𝒰^{}(A)=`$.
If $`A`$ occurs as the head of a rule in Inst-$`𝒫`$, then
$`\mathrm{\Psi }_𝒫^𝒰(Fix_𝒰^{},Fix_𝒰^{})(A)=Fix_𝒰^{}(A)`$
because $`Fix_𝒰^{}`$ is a fixed model. Thus, we have
$`\mathrm{\Psi }_𝒫^𝒰(Fix_𝒰^{},Fix_𝒰^{})_kFix_𝒰^{}`$.
As $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}(Fix_𝒰^{})`$ is the least fixpoint of $`\lambda x.\mathrm{\Psi }_𝒫^𝒰(x,Fix_𝒰^{})`$, we have
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}(Fix_𝒰^{})_kFix_𝒰^{}`$.
Now, $`Fix_𝒰^𝒰`$ is the least fixpoint of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{𝒰}`$, so $`Fix_𝒰^𝒰_kFix_𝒰^{}`$.
Similarly, $`Fix_𝒰^𝒰_kFix_𝒰^𝒯`$.
It follows from Theorem 5 that the skeptical semantics gives less information than the pessimistic and optimistic semantics. From this theorem, we can infer the following result:
###### Corollary 1
Let $`𝒫`$ be a Fitting program. Then we have:
$$Fix_𝒰^𝒰_kFix_𝒰^{}Fix_𝒰^𝒯$$
.
Proof. The proof is immediate using the preceding theorem and interla-
cing.
In the previous corollary, the equality is satisfied for positive programs, but if we accept negation then it is false in general.
This corollary suggests the possibility of defining a new semantics, namely $`Fix_𝒰^{}Fix_𝒰^𝒯`$, that is smaller than the pessimistic and optimistic semantics but greater than the skeptical semantics. The following example shows that this semantics can be useful in certain contexts.
$`𝒫:𝒜\neg `$
| Semantics of $`𝒫`$ | $`Fix_𝒰^{}`$ | $`Fix_𝒰^𝒯`$ | $`Fix_𝒰^𝒰`$ | $`Fix_𝒰^{}Fix_𝒰^𝒯`$ |
| --- | --- | --- | --- | --- |
| A | $`𝒯`$ | $`𝒯`$ | $`𝒰`$ | $`𝒯`$ |
| B | $``$ | $`𝒯`$ | $`𝒰`$ | $`𝒰`$ |
The program $`𝒫`$ seems to assert that $`A`$ is always true (because it is inferred from either $`B`$ or $`\neg B`$), and this conclusion is reached by both the optimistic and the pessimistic semantics. However, there is no reason why we should choose between $`B`$ $`true`$ and $`B`$ $`false`$ when we cannot assert anything about the value of $`B`$. It seems therefore more natural in this case to take the consensus between the pessimistic and optimistic semantics, which gives the value $`unknown`$ to $`B`$.
Although $`Fix_𝒰^{}Fix_𝒰^𝒯`$ seems to give an interesting new semantics, one has to check under what conditions $`Fix_𝒰^{}Fix_𝒰^𝒯`$ is actually a model. Assuming that it is a model, we can call it the consensus semantics.
## 5 Conclusion
We have defined parametrized semantics for the family of Fitting programs , and an algorithm for their computation. The family of Fitting programs is very general and includes the conventional logic programs. When we restrict the class of Fitting programs to the class of conventional logic programs, the new semantics coincide with the conventional ones. This allows us to compare conventional semantics in this new setting in which they are embedded. It also allows us to combine conventional semantics, and thus it suggests the possibility of defining new semantics such as the consensus semantics that we proposed in this paper.
Extending this work to logics with signs and annotations is a topic for future work.
### Appendix - Proof of Theorem 3
We need a proposition and a few lemma to prove the next result.
###### Lemma 3
Let $`x𝒱()`$$`x=(x𝒰)(x𝒰)`$ and $`𝒯=𝒰`$.
Proof.
| $`(x𝒰)(x𝒰)`$ | = | $`[x(x𝒰)][𝒰(x𝒰)]`$ |
| --- | --- | --- |
| | = | $`[(xx)(x𝒰)][(𝒰x)(𝒰𝒰)]`$ |
| | = | $`[xx][x𝒰]`$ |
| | = | $`x(x𝒰)`$ |
| | = | $`x`$. |
$``$ is the smallest member of $`𝒱()`$ under the truth ordering so $`_t𝒰`$ and using the interlacing conditions, we have : $`𝒯_t𝒰𝒯=𝒰`$.
Similarly, $`𝒰_t𝒯`$, so $`𝒰=𝒰_t𝒯`$.
The three following equations have similar proofs: $`𝒯=`$, $`𝒰=`$ and $`𝒰=𝒯`$.
###### Lemma 4
Let $`a,b,c𝒱()`$. If $`a_tb_tc`$, then
(1) $`(a𝒰)(c𝒰)_k𝒰`$ ;
(2) $`(a𝒰)c_kb`$;
(3) $`(a𝒰)(c𝒰)_kb`$ .
Proof. Since $`𝒰_k`$, by the interlacing conditions $`a𝒰_ka=`$. Similarly, $`c𝒰_k𝒯`$. Then by the interlacing conditions,
$`(a𝒰)(c𝒰)_k𝒯`$.
By the precedent lemma, part 1 is established.
Then, using the hypothesis and the interlacing,
$`(a𝒰)c_k(ab)c=bc_kb`$.
Finally, $`(a𝒰)(c𝒰)_k(a𝒰)(cb)=(a𝒰)b_kb`$.
Now, we can prove the result we need.
###### Lemma 5
Let $`a,b,c𝒱()`$. If $`a_tb_tc`$, then $`ac_kb`$.
Proof. Using the precedent lemmas and interlacing,
$`ac`$ = $`[(a𝒰)(a𝒰)]c`$ = $`[(a𝒰)c][(a𝒰)c]`$ $`_k`$ $`[(a𝒰)c]b`$ = $`[(a𝒰)((c𝒰)(c𝒰))]b`$ = $`[(a𝒰)(c𝒰)][(a𝒰)(c𝒰)]b`$ $`_k`$ $`b𝒰b`$ = $`b`$.
Using a similar proof, the following can also be shown:
(1) if $`a_tb_tc`$, then $`b_kac`$;
(2) if $`a_kb_kc`$, then $`ac_tb`$;
(3) if $`a_kb_kc`$, then $`b_tac`$.
###### Proposition 3
If $`f`$ is a monotone mapping on a complete lattice, then $`f`$ and $`f^2`$ have the same least and greatest fixpoints.
Proof. Let $`a`$ be the least fixpoint of $`f`$ and let $`b`$ be the least fixpoint of $`f^2`$.
Every fixpoint of $`f`$ is also a fixpoint of $`f^2`$ and $`b`$ is the least fixpoint of $`f^2`$, so $`ba`$.
If $`x`$ is a fixpoint of $`f^2`$, then $`f^2(f(x))=f(f^2(x))=f(x)`$ so $`f(x)`$is a fixpoint of $`f^2`$. $`b`$ is the least fixpoint of $`f^2`$, so $`f(b)`$ is a fixpoint of $`f^2`$ and $`bf(b)`$. By monotonicity, $`f(b)f^2(b)=b`$, so $`b=f(b)`$. Since $`a`$ is the least fixpoint of $`f`$, $`ab`$.
Now, we can prove the result concerning the structure of the family of $`\alpha _𝒫`$-fixed models.
Theorem 3 Let $`𝒫`$ be a Fitting program. Then we have:
| $`Fix_𝒰^\alpha `$ | = | $`Fix_{}^\alpha `$ | $``$ | $`Fix_𝒯^\alpha `$ |
| --- | --- | --- | --- | --- |
| $`Fix_{}^\alpha `$ | = | $`Fix_{}^\alpha `$ | $``$ | $`Fix_𝒯^\alpha `$ |
| $`Fix_{}^\alpha `$ | = | $`Fix_𝒰^\alpha `$ | $``$ | $`Fix_{}^\alpha `$ |
| $`Fix_𝒯^\alpha `$ | = | $`Fix_𝒰^\alpha `$ | $``$ | $`Fix_{}^\alpha `$ |
Proof. the proof is separated in several parts.
Part 1. We want to show that $`Fix_{}^\alpha Fix_𝒯^\alpha `$ and $`Fix_{}^\alpha Fix_𝒯^\alpha `$ are fixpoints of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ in order to have $`Fix_𝒰^\alpha _kFix_{}^\alpha Fix_𝒯^\alpha `$ and $`Fix_{}^\alpha Fix_𝒯^\alpha _kFix_{}^\alpha `$.
By monotonicity of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ under knowledge ordering, we have
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha Fix_𝒯^\alpha )_k\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha )=Fix_𝒯^\alpha `$,
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha Fix_𝒯^\alpha )_k\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_𝒯^\alpha )=Fix_{}^\alpha `$.
So $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha 𝒫`$-$`Fix_𝒯^\alpha )_kFix_{}^\alpha Fix_𝒯^\alpha `$.
Also, $`Fix_{}^\alpha _tFix_𝒯^\alpha `$, so, by interlacing,
$`Fix_{}^\alpha =Fix_{}^\alpha Fix_{}^\alpha _tFix_{}^\alpha Fix_𝒯^\alpha _tFix_𝒯^\alpha Fix_𝒯^\alpha =Fix_𝒯^\alpha `$,
and, by anti-monotonicity of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ under the truth ordering,
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_𝒯^\alpha )_t\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_𝒯^\alpha Fix_𝒯^\alpha )_t\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha )`$,
so,
$`Fix_{}^\alpha _t\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_𝒯^\alpha Fix_𝒯^\alpha )_tFix_𝒯^\alpha `$.
Using the precedent lemma,
$`Fix_{}^\alpha Fix_𝒯^\alpha _t\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }(Fix_{}^\alpha Fix_𝒯^\alpha )`$.
We have shown that $`Fix_{}^\alpha Fix_𝒯^\alpha `$ is a fixed point of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$.
The proof for $`Fix_{}^\alpha Fix_𝒯^\alpha `$ is dual.
$`Fix_𝒰^\alpha `$ and $`Fix_{}^\alpha `$ are the least and greatest fixpoints of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$, so part 1 is established.
Part 2. We show now the other direction.
$`Fix_{}^\alpha `$ and $`Fix_𝒯^\alpha `$ are the two extremal oscillation points of $`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ under the truth ordering, so
$`Fix_{}^\alpha _tFix_𝒰^\alpha _tFix_𝒯^\alpha `$,
Thus, using the precedent lemma,
$`Fix_{}^\alpha Fix_𝒯^\alpha _tFix_𝒰^\alpha `$.
The proof of the other inequality is dual. The two first equality of the theorem are established.
Part 3. In this part, we show the last two equality.
$`\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha }`$ is monotonic under the knowledge ordering and its least and greatest fixpoints are $`Fix_𝒰^\alpha `$ and $`Fix_{}^\alpha `$. Under the knowledge ordering, $`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2`$ is also monotonic and, using the precedent proposition, has the same least and greatest fixpoints. $`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2`$ is also monotonic under the truth ordering and its least and greatest fixpoints under this ordering are $`Fix_{}^\alpha `$ et $`Fix_𝒯^\alpha `$.
We have
$`Fix_𝒰^\alpha Fix_{}^\alpha _tFix_𝒰^\alpha `$
So
$`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2(Fix_𝒰^\alpha Fix_{}^\alpha )_t(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2(Fix_𝒰^\alpha )=Fix_𝒰^\alpha `$
and, similarly, $`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2(Fix_𝒰^\alpha Fix_{}^\alpha )_tFix_{}^\alpha `$.
Consequently, using the fact that $`Fix_{}^\alpha `$ is the least fixpoint of $`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2`$ under the truth ordering,
$`Fix_{}^\alpha _tFix_𝒰^\alpha Fix_{}^\alpha `$.
Further, $`Fix_{}^\alpha `$ is a fixpoint of $`(\mathrm{\Psi }_{}^{}{}_{𝒫}{}^{\alpha })^2`$, and $`Fix_𝒰^\alpha `$ and $`Fix_{}^\alpha `$ are its least and greatest fixpoints, so
$`Fix_𝒰^\alpha _kFix_{}^\alpha _kFix_{}^\alpha `$
Thus, using the lemma,
$`Fix_𝒰^\alpha Fix_{}^\alpha _tFix_{}^\alpha `$.
The last part is dual.
|
warning/0002/math-ph0002016.html
|
ar5iv
|
text
|
# Extensions of diffeomorphism and current algebras
## 1 Introduction
An extension $`\widehat{L}`$ of a Lie algebra $`L`$ by a module $`M`$ is an exact sequence
$$0M\stackrel{ı}{}\widehat{L}\stackrel{\pi }{}L0.$$
This means that $`ı`$ is injective, $`\pi `$ is surjective, and $`M`$ is an ideal in $`\widehat{L}`$. It is precisely this situation which is of interest in physics, because if $`L`$ is a classical symmetry algebra (realized in terms of Poisson brackets), quantum corrections are of order $`\mathrm{}`$ and thus generate an ideal. In particular, if $`M=`$ we say that the extension is central, which is the case that has attracted most attention in physics; suffice it to mention the ample applications of Virasoro and affine Kac-Moody algebras.
The best known non-central extension is the Mickelsson-Faddeev (MF) algebra , which is an abelian extension of the algebra $`map(N,𝔤)`$ of maps from $`N`$-dimensional spacetime to a finite-dimensional Lie algebra $`𝔤`$. $`map(N,𝔤)`$ also admits higher-dimensional generalizations of the Kac-Moody cocycle , whose Fock representations were first constructed in . Similarly, the diffeomorphism algebra in $`N`$ dimensions, $`diff(N)`$, has non-central extensions analogous to the Virasoro algebra . The representation theory of these algebras was developped in . It appears that representations of the MF algebra, if they exist, are not attainable by similar methods .
In , Dzhumadil’daev classified all extensions of $`L=diff(N)`$ when $`M`$ is a tensor module. He also covered the cases when $`L`$ is one of the algebras of divergence free, Hamiltonian and contact vector fields. Unfortunately, his paper is not easy to read for a physicist (at least not for this one), so one purpose of the present paper is to review his classification in a more physicist-friendly manner, using notation from tensor calculus. Moreover, his results are quite bewildering, since he obtain no less than seventeen different cocycles. However, it turns out that all of them can be grouped into four classes:
1. A cubic (in derivatives) cocycle, which splits into its traceless and trace parts.
2. Quartic cocycles, which follow from the MF extension for $`gl(N)`$ (recall that tensor fields are functions with values in $`gl(N)`$ modules). In fact, only the tensor part of the MF extension was used, which can be removed by a redefinition à la Cederwall et al. .
3. Quintic cocycles which only exist in two dimensions.
4. Special cases in one dimension.
Dzhumadil’daev also considered cocycles for the $`diff(N)`$ subalgebras $`svect(N)`$ (divergence-free vector fields), $`Ham(N)`$ (Hamiltonian vector fields) and $`K(N)`$ (contact vector fields). All such cocycles follow directly by restriction from the full diffeomorphism algebra.
I extend Dzhumadil’daev’s result by constructing some extensions where $`M`$ is not a tensor module. These cases include the higher-dimensional Virasoro algebras of Eswara Rao and Moody and myself . Another generalization is found be considering the inhomogenous term in the MF extension.
It turns out that all of Dzhumadil’daev’s generic cocycles and several of his low-dimensional ones can be obtained as limits of trivial cocycles. One constructs a family of trivial cocycles, parametrized by a continuous parameter (the conformal weight $`\lambda `$), and let $`\lambda `$ approach a critical value $`\lambda _0`$. Thus, these cocycles are non-trivial in the usual cohomological sense, but belong to the closure of the space of trivial cocycles. On the other hand, the Virasoro and MF extensions do not belong to this closure, because there is no continuous parameter that can be varied. Geometrically, they involve closed chains (one- and three-, respectively), and the closedness condition is consistent for $`\lambda =1`$ only.
Dzhumadil’daev and Ovsienko and Roger have also classified cocycles for the special case $`N=1`$. I show that some of these have higher-dimensional analogues not previously considered, although this generalization is quite unnatural and uninteresting. Moreover, in one dimension these cocycles, possibly with one exception, are limits of trivial cocycles.
Dzhumadil’daev’s classification can be used to construct interesting extensions of subalgebras of the diffeomorphism algebra. To this end, I consider the inclusion $`map(N,diff(d))diff(N+d)`$, and further $`𝔤gl(d)diff(d)`$. In fact, the algebra $`map(N,diff(d))`$ defines an interesting generalization of gauge symmetry: the replacement of a global symmetry $`𝔤`$ by a gauge symmetry $`map(N,𝔤)`$ amounts to a localization in base space, but the gauge transformations are still rigid in target space. Replacing $`𝔤`$ by $`diff(d)`$ makes transformations local in target space as well. By studying the restriction of the extensions under the above inclusions, existence of extensions for subalgebras is shown, but neither non-triviality nor exhaustion. However, non-triviality can be checked by hand, and my method tautologically exhaust all extensions that can be lifted to tensor module extensions of the algebra of diffeomorphisms in total space.
## 2 Background
### 2.1 Diffeomorphism algebra
Let $`\xi =\xi ^\mu (x)_\mu `$, $`x^N`$, $`_\mu =/x^\mu `$, be a vector field, with commutator $`[\xi ,\eta ]\xi ^\mu _\mu \eta ^\nu _\nu \eta ^\nu _\nu \xi ^\mu _\mu `$. Greek indices $`\mu ,\nu =1,2,\mathrm{},N`$ label the spacetime coordinates and the summation convention is used on all kinds of indices. The diffeomorphism algebra $`diff(N)`$ is generated by Lie derivatives $`_\xi `$. Dzhumadil’daev denotes this algebra by $`W_N`$ in honour of Witt. In the literature, it is also known as the algebra of vector fields and is denoted by $`Vect(N)`$ or $`Vect(^N)`$.
An extensions of $`diff(N)`$ is given by a bilinear cocycle $`c(\xi ,\eta )`$:
$`[_\xi ,_\eta ]=_{[\xi ,\eta ]}+c(\xi ,\eta ).`$ (2.1)
For convenience, some formulas are also displayed in a Fourier basis. With $`L_\mu (m)=_\xi `$ for $`\xi =i\mathrm{exp}(im_\rho x^\rho )_\mu `$, $`m^N`$, (2.1) is replaced by
$`[L_\mu (m),L_\nu (n)]=n_\mu L_\mu (m+n)m_\nu L_\nu (m+n)+c_{\mu \nu }(m,n).`$ (2.2)
We say that an extension is local if it has the form
$`c_{\mu \nu }(m,n)=pol_{\mu \nu }^a(m,n)A_a(m+n),`$ (2.3)
where $`pol_{\mu \nu }^a(m,n)=pol_{\nu \mu }^a(n,m)`$ is a polynomial and $`A_a`$ is some operator.
Let $`T_\nu ^\mu `$ form a basis for $`gl(N)`$, with brackets
$`[T_\nu ^\mu ,T_\tau ^\sigma ]=\delta _\nu ^\sigma T_\tau ^\mu \delta _\tau ^\mu T_\mu ^\sigma .`$ (2.4)
Then
$`_\xi =\xi ^\mu (x)_\mu +_\nu \xi ^\mu (x)T_\mu ^\nu `$ (2.5)
satisfies (2.1) with zero cocycle. Analogously, if $`T_\nu ^\mu (m)`$ form a basis for an extension of $`map(N,gl(N))`$, with brackets
$`[T_\nu ^\mu (m),T_\tau ^\sigma (n)]`$ $`=`$ $`\delta _\nu ^\sigma T_\tau ^\mu (m+n)\delta _\tau ^\mu T_\nu ^\sigma (m+n)+k_{\nu \tau }^{\mu \sigma }(m,n),`$
$`[L_\mu (m),T_\tau ^\sigma (n)]`$ $`=`$ $`n_\mu T_\tau ^\sigma (m+n),`$
then
$`L_\mu ^{}(m)=L_\mu (m)+m_\nu T_\mu ^\nu (m)`$ (2.7)
satisfies an extension of $`diff(N)`$ with cocycle
$`c_{\mu \nu }(m,n)=m_\sigma n_\tau k_{\mu \nu }^{\sigma \tau }(m,n).`$ (2.8)
A tensor module is the carrying space of the $`diff(N)`$ representation obtained by substituting a $`gl(N)`$ representation into (2.5). A tensor of type $`(p,q;\lambda )`$ ($`p`$ contravariant and $`q`$ covariant indices and conformal weight $`\lambda `$) is described by the equivalent formulas
$`[_\xi ,\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(x)]=\xi ^\mu (x)_\mu \mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(x)\lambda _\mu \xi ^\mu (x)\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(x)`$
$`+{\displaystyle \underset{i=1}{\overset{p}{}}}_\mu \xi ^{\sigma _i}(x)\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\mu ..\sigma _p}(x){\displaystyle \underset{j=1}{\overset{q}{}}}_{\tau _j}\xi ^\mu (x)\mathrm{\Phi }_{\tau _1..\mu ..\tau _q}^{\sigma _1..\sigma _p}(x),`$
$`[_\xi ,\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(\varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q})]=\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(\xi ^\mu _\mu \varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q}+(1\lambda )_\mu \xi ^\mu \varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q}`$
$`+{\displaystyle \underset{i=1}{\overset{p}{}}}_{\sigma _i}\xi ^\mu \varphi _{\sigma _1..\mu ..\sigma _p}^{\tau _1..\tau _q}{\displaystyle \underset{j=1}{\overset{q}{}}}_\mu \xi ^{\tau _j}\varphi _{\sigma _1..\sigma _p}^{\tau _1..\mu ..\tau _q}),`$ (2.9)
$`[L_\mu (m),\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(n)]=(n_\mu +(1\lambda )m_\mu )\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(m+n)+`$
$`+{\displaystyle \underset{i=1}{\overset{p}{}}}\delta _\mu ^{\sigma _i}m_\rho \mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\rho ..\sigma _p}(m+n){\displaystyle \underset{j=1}{\overset{q}{}}}m_{\tau _j}\mathrm{\Phi }_{\tau _1..\mu ..\tau _q}^{\sigma _1..\sigma _p}(m+n),`$
where $`\varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q}`$ is an arbitrary function on $`^N`$ and
$`\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(\varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q})={\displaystyle d^Nx\varphi _{\sigma _1..\sigma _p}^{\tau _1..\tau _q}(x)\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(x)}.`$ (2.10)
For brevity, we shall often write the rhs of (2.9) simply as $`(p,q;\lambda )`$. Further, we abbreviate the action on objects which contain additional terms as
$`[L_\mu (m),\mathrm{\Phi }_{\tau _1..\tau _q}^{\sigma _1..\sigma _p}(n)]=(p,q;\lambda )+\text{more},`$ (2.11)
etc. Tensor modules contain irreducible submodules consisting of symmetric, anti-symmetric, and traceless tensors, labelled by the irreps of $`gl(N)`$. As is standard in physics, (anti-)symmetrization of indices is denoted by parentheses (brackets), and vertical bars inhibit the operation. Thus, $`\varphi ^{(\mu |\nu |\rho )}=\varphi ^{\mu \nu \rho }+\varphi ^{\rho \nu \mu }`$ and $`\varphi _{[\mu \nu ]}=\varphi _{\mu \nu }\varphi _{\nu \mu }`$.
However, not all modules are tensor modules. The following cases will be considered below:
1. A totally skew tensor $`\omega _{\sigma _1..\sigma _p}=\omega _{[\sigma _1..\sigma _p]}`$ of type $`(0,p;0)`$, i.e. a $`p`$-form, contains a submodule consisting of closed $`p`$-forms, which amounts to the conditions
$`_{[\nu }\omega _{\sigma _1..\sigma _p]}(x)0,\omega _{\sigma _1..\sigma _p}(_\nu \varphi ^{[\nu \sigma _1..\sigma _p]})0,m_{[\nu }\omega _{\sigma _1..\sigma _p]}(m)0.`$ (2.12)
2. Dually, a totally skew tensor $`S^{\sigma _1..\sigma _p}=S^{[\sigma _1..\sigma _p]}`$ of type $`(p,0;1)`$, can be identified as a $`p`$-chain. Closed $`p`$-chains satisfy the conditions
$`_{\sigma _1}S^{\sigma _1..\sigma _p}(x)0,S^{\sigma _1..\sigma _p}(_{\sigma _1}\varphi _{\sigma _2..\sigma _p})0,m_{\sigma _1}S^{\sigma _1..\sigma _p}(m)0.`$ (2.13)
3. The connection $`\mathrm{\Gamma }_{\sigma \tau }^\rho `$ transforms as a tensor field of type $`(1,2;0)`$, apart from an additional term
$`[_\xi ,\mathrm{\Gamma }_{\sigma \tau }^\rho (x)]`$ $`=`$ $`(1,2;0)+_\sigma _\tau \xi ^\rho ,`$
$`[_\xi ,\mathrm{\Gamma }_{\sigma \tau }^\rho (\varphi _\rho ^{\sigma \tau })]`$ $`=`$ $`(1,2;0)+{\displaystyle d^Nx_\sigma _\tau \xi ^\rho (x)\varphi _\rho ^{\sigma \tau }(x)},`$ (2.14)
$`[L_\mu (m),\mathrm{\Gamma }_{\sigma \tau }^\rho (n)]`$ $`=`$ $`(1,2;0)+m_\sigma m_\tau \delta _\mu ^\rho \delta (m+n),`$
where $`(1,2;0)`$ denote regular terms as in (2.11).
### 2.2 Divergence-free vector fields
The algebra of divergence-free (or special) vector fields $`svect(N)diff(N)`$ is generated by $`_\xi `$ such that $`_\mu \xi ^\mu =0`$, or equivalently $`L_\mu (m)`$ such that $`m_\mu =0`$. Tensor modules are given by (2.9), except that the conformal weight $`\lambda `$ is irrelevant.
### 2.3 Hamiltonian vector fields
The algebra $`Ham(N)diff(N)`$ ($`N`$ even) consists of vector fields $`\xi `$ that leave the two-form $`ϵ_{\mu \nu }dx^\mu dx^\nu `$ invariant, where $`ϵ_{\mu \nu }=ϵ_{\nu \mu }`$ is the symplectic form, whose inverse $`ϵ^{\mu \nu }`$ satisfies $`ϵ^{\mu \rho }ϵ_{\rho \nu }=ϵ_{\nu \rho }ϵ^{\rho \mu }=\delta _\nu ^\mu `$. Such Hamiltonian vector fields are of the form $`\xi =ϵ^{\mu \nu }_\mu f_\nu `$. Dzhumadil’daev denotes $`Ham(N)`$ by $`H_n`$, where $`N=2n`$. Any extension of $`Ham(N)`$ takes the form
$`[H_f,H_g]=H_{\{f,g\}}+c_H(f,g),`$ (2.15)
where $`\{f,g\}=ϵ^{\mu \nu }_\mu f_\nu g`$ is the Poisson bracket. Alternatively, in the Fourier basis
$`[H(m),H(n)]=(m\times n)H(m+n)+c_H(m,n),`$ (2.16)
where $`m\times nϵ^{\mu \nu }m_\mu n_\nu `$. Since $`ϵ^{\mu \nu }_\mu _\nu f0`$, any Hamiltonian vector field is divergence free. The converse is also true in two dimensions, but when $`N4`$ there are divergence-free vector fields that are not Hamiltonian. One checks that $`[H_f,ϵ^{\mu \nu }]0`$, so $`ϵ^{\mu \nu }`$ and $`ϵ_{\mu \nu }`$ can be used to raise and lower indices. A typical tensor module is hence of the form
$`[H_f,\mathrm{\Phi }^\sigma (\varphi _\sigma )`$ $`=`$ $`\mathrm{\Phi }^\sigma (\{f,\varphi _\sigma \}+ϵ^{\mu \nu }_\mu _\sigma f\varphi _\nu )`$
$`[H(m),\mathrm{\Phi }^\sigma (n)]`$ $`=`$ $`n_\mu \mathrm{\Phi }^\sigma (m+n)+ϵ^{\mu \sigma }m_\mu m_\nu \mathrm{\Phi }^\nu (m+n).`$
### 2.4 Contact vector fields
The contact algebra $`K(N)diff(N)`$ ($`N`$ odd) consists of vector fields which leave the one-form $`\alpha =dx^0+ϵ_{ij}x^idx^j`$ invariant, where $`ϵ_{ij}`$ is the symplectic form in one dimension less. Here greek indices $`\mu ,\nu =0,1,2,\mathrm{},N`$ run over all $`N+1`$ indices but latin indices $`i,j=1,2,\mathrm{},N`$ exclude the time index $`0`$. Contact vector fields are of the form
$`K_f=\mathrm{\Delta }f_0+_0fx^i_i+ϵ^{ij}_jf_i,`$ (2.18)
where $`\mathrm{\Delta }f=2fx^i_if`$. Any extension of $`K(N+1)`$ has the form
$`[K_f,K_g]`$ $`=`$ $`K_{[f,g]_K}+c_K(f,g),`$ (2.19)
where the contact bracket reads
$`[f,g]_K=_0f\mathrm{\Delta }g_0g\mathrm{\Delta }f+\{f,g\}`$ (2.20)
and $`\{f,g\}=ϵ^{ij}_if_jg`$ is the Poisson bracket in $`N`$ dimensions. Due to the explicit appearence of $`x^i`$ in the definition of $`\mathrm{\Delta }`$, it is inconvenient to describe this algebra in a Fourier basis. Dzhumadil’daev denotes $`K(N)`$ by $`K_n`$, where $`N=2n+1`$.
### 2.5 Gauge algebra
Let $`𝔤`$ be a finite-dimensional Lie algebra with basis $`J^a`$ (hermitian if $`𝔤`$ is compact and semisimple), structure constants $`f^{ab}_c`$ and brackets $`[J^a,J^b]=if^{ab}{}_{c}{}^{}J_{}^{c}`$. Our notation is similar to or , chapter 13. We always assume that $`𝔤`$ has a Killing metric proportional to $`\delta ^{ab}`$. Further assume that there is a priviledged vector $`\delta ^a\text{tr}J^a`$, such that $`f^{ab}{}_{c}{}^{}\delta _{}^{c}0`$. Of course, $`\delta ^a=0`$ if $`𝔤`$ is semisimple, but it may be non-zero if $`𝔤`$ contains abelian factors. The primary example is $`𝔤=gl(N)`$, where $`\text{tr}(T_\nu ^\mu )\delta _\nu ^\mu `$.
Let $`map(N,𝔤)`$ be the algebra of maps from $`^N`$ to $`𝔤`$, also known as the gauge or current algebra. We denote its generators by $`𝒥_X`$, where $`X=X_a(x)J^a`$, $`x^N`$, is a $`𝔤`$-valued function, and define $`[X,Y]=if^{ab}{}_{c}{}^{}X_{a}^{}Y_bJ^c`$. Alternatively, we use a Fourier basis with generators $`𝒥^a(m)`$, $`m^N`$. Any extension of $`map(N,𝔤)`$ has the form
$`[𝒥_X,𝒥_Y]`$ $`=`$ $`𝒥_{[X,Y]}+c(X,Y),`$
$`[_\xi ,𝒥_X]`$ $`=`$ $`𝒥_{\xi ^\mu _\mu X}+c(\xi ,X),`$
$`[𝒥^a(m),𝒥^b(n)]`$ $`=`$ $`if^{ab}{}_{c}{}^{}𝒥_{}^{c}(m+n)+c^{ab}(m,n),`$
$`[L_\mu (m),𝒥^a(n)]`$ $`=`$ $`n_\mu 𝒥^a(m+n)+c_\mu ^a(m,n).`$
The analogue of tensor modules are functions with values in some $`𝔤`$ module $`M`$. To the three formulas in (2.9) correspond
$`[𝒥_X,\mathrm{\Phi }^i(x)]`$ $`=`$ $`X_a\sigma _j^{ia}\mathrm{\Phi }^j(x),`$
$`[𝒥_X,\mathrm{\Phi }^i(\varphi _i)]`$ $`=`$ $`\mathrm{\Phi }^i(X_a\sigma _i^{ja}\varphi _j),`$
$`[𝒥^a(m),\mathrm{\Phi }^i(n)]`$ $`=`$ $`\sigma _j^{ia}\mathrm{\Phi }^j(m+n),`$
where $`\sigma ^a=(\sigma _j^{ia})`$ are the representation matrices acting on $`M`$. Moreover, the intertwining action of diffeomorphisms is given by (2.9).
Among non-tensor modules we cite the connection, which is a central extension of the adjoint.
$`[𝒥_X,A_\nu ^b(x)]`$ $`=`$ $`if^{ab}{}_{c}{}^{}X_{a}^{}(x)A_\nu ^c(x)+\delta ^{ab}_\nu X_a(x),`$
$`[𝒥_X,A_\nu ^b(\varphi _b^\nu )]`$ $`=`$ $`A_\nu ^b(if^{ac}{}_{b}{}^{}X_{a}^{}\varphi _c^\nu )+{\displaystyle d^Nx\delta ^{ab}_\nu X_a(x)\varphi _b^\nu (x)},`$
$`[𝒥^a(m),A_\nu ^b(n)]`$ $`=`$ $`if^{ab}{}_{c}{}^{}A_{\nu }^{c}(m+n)+m_\nu \delta ^{ab}.`$
The best known extension of $`map(N,𝔤)`$, $`N3`$, is the MF extension:
$`[𝒥^a(m),𝒥^b(n)]`$ $`=`$ $`if^{ab}{}_{c}{}^{}𝒥_{}^{c}(m+n)+m_\mu n_\nu ^{ab\mu \nu }(m+n),`$
$`[𝒥^a(m),^{bc\mu \nu }(n)]`$ $`=`$ $`if^{ab}{}_{d}{}^{}_{}^{dc\mu \nu }(m+n)+`$
$`+if^{ac}{}_{d}{}^{}_{}^{bd\mu \nu }(m+n)+d^{abc}m_\rho S_3^{\mu \nu \rho }(m+n),`$
and all other brackets vanish. Here, $`S_3^{\mu \nu \rho }`$ is a closed three-chain (2.13) and $`d^{abc}=\text{tr}J^{(a}J^bJ^{c)}`$ are totally symmetric. In particular, in three dimensions we can write $`^{ab\mu \nu }(m)=ϵ^{\mu \nu \rho }d^{abc}A_\rho ^c(m)`$, $`S_3^{\mu \nu \rho }(m)=ϵ^{\mu \nu \rho }\delta (m)`$, so $`A_\rho ^c(m)`$ transforms as a connection. It was found in that the three-chain term constitutes an obstruction against the construction of Fock modules.
## 3 Dzhumadil’daev’s classification: $`diff(N)`$ cocycles
Dzhumadil’daev classified extensions of $`diff(N)`$ by tensor modules , and found 17 inequivalent ones. A tensor density can be viewed as a function with values in a $`gl(N)`$ module. Since $`gl(N)sl(N)gl(1)`$, $`gl(N)`$ irreps are labelled by an $`sl(N)`$ highest weight and a conformal weight $`\lambda `$. To an $`sl(N)`$ highest weight $`\mathrm{}_1\pi _1+\mathrm{}_2\pi _2+\mathrm{}+\mathrm{}_{N1}\pi _{N1}`$, $`\pi _i`$ being the fundamental weights, we can associate a partition $`\{\lambda _1,\lambda _2,\mathrm{},\lambda _{N1}\}`$, where $`\lambda _i=\mathrm{}_i+\mathrm{}_{i+1}+\mathrm{}+\mathrm{}_{N+1}=_{j=i}^{N1}\mathrm{}_j`$. This can be visualized as a Young tableaux with $`\lambda _i`$ boxes in the $`i`$:th row. A contravariant vector corresponds to a tableaux with a single box, i.e. to the root $`\pi _1`$. A covariant vector can be identified with a tableaux with a single column with $`N1`$ boxes, i.e. the root $`\pi _{N1}`$.
Let $`ϵ_{\mu _1..\mu _N}`$ be the totally antisymmetric symbol, given by $`ϵ_{\mu _1..\mu _N}=+1`$ if $`\mu _1..\mu _N`$ is a even permutation of $`12..N`$, $`=1`$ if it is an odd permutation, and $`=0`$ if two indices are equal. It can be regarded as a constant tensor field of type $`(0,N;1)`$, since this makes the transformation law $`[_\xi ,ϵ_{\mu _1..\mu _N}(x)]=0`$ consistent. Alternatively, we may view it as a constant tensor field $`ϵ^{\mu _1..\mu _N}`$ of type $`(N,0;1)`$. The existence of this symbol establishes the standard isomorphism between $`p`$ lower indices and $`Np`$ upper indices, e.g. $`p`$-forms and $`(Np)`$-chains:
$`A_{[\mu _1..\mu _p]}=ϵ_{\mu _1..\mu _p\nu _1..\nu _{Np}}A^{[\nu _1..\nu _{Np}]}.`$ (3.25)
In particular, a covariant vector field (of weight $`\lambda `$) is equivalent to a skew tensor field with $`N1`$ contravariant indices and weight $`\lambda +1`$.
Dzhumadil’daev’s classification is encoded in , Table 1. The following three tables describe the corresponding modules, both in his notation and tensor calculus notation.
1. For every $`NN_0`$:
$`\begin{array}{ccccccc}& N_0\hfill & \text{HW}\hfill & \lambda \hfill & \text{Type}\hfill & \text{Tensor}\hfill & \\ \psi _1^W\hfill & 1\hfill & \pi _1\hfill & 1\hfill & (1,0;1)\hfill & S^\rho \hfill & \\ \psi _2^W\hfill & 2\hfill & 2\pi _1+\pi _{N1}\hfill & 2\hfill & (2,1;1)\hfill & K_\mu ^{(\sigma \tau )}\hfill & \\ \psi _{3,4}^W\hfill & 2\hfill & \pi _2\hfill & 1\hfill & (2,0;1)\hfill & F^{[\mu \nu ]}\hfill & \\ \psi _{5,6}^W\hfill & 3\hfill & \pi _1+\pi _2+\pi _{N1}\hfill & 2\hfill & (3,1;1)\hfill & E_\mu ^{\rho (\sigma \tau )}:E_\mu ^{(\rho \sigma \tau )}=0\hfill & \\ \psi _7^W\hfill & 2\hfill & 3\pi _1+\pi _{N1}\hfill & 2\hfill & (3,1;1)\hfill & D_\mu ^{(\rho \sigma \tau )}\hfill & \\ \psi _8^W\hfill & 3\hfill & 2\pi _1+\pi _2+2\pi _{N1}\hfill & 3\hfill & (4,2;1)\hfill & B_{(\mu \nu )}^{[(\lambda \rho )(\sigma \tau )]}\hfill & \\ \psi _9^W\hfill & 3\hfill & 4\pi _1+\pi _{N2}\hfill & 2\hfill & (4,2;1)\hfill & A_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}\hfill & \\ \psi _{10}^W\hfill & 4\hfill & 2\pi _2+\pi _{N2}\hfill & 2\hfill & (4,2;1)\hfill & C_{[\mu \nu ]}^{((\lambda \rho )(\sigma \tau ))}:C_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}=0\hfill & \end{array}`$ (3.35)
The symmetry conditions can alternatively be written as
$`A_{\mu \nu }^{\lambda \rho \sigma \tau }`$ $`=`$ $`A_{\mu \nu }^{\rho \lambda \sigma \tau }=A_{\mu \nu }^{\sigma \rho \lambda \tau }=A_{\mu \nu }^{\tau \rho \sigma \lambda }=A_{\nu \mu }^{\lambda \rho \sigma \tau },`$
$`B_{\mu \nu }^{\lambda \rho \sigma \tau }`$ $`=`$ $`B_{\mu \nu }^{\rho \lambda \sigma \tau }=B_{\mu \nu }^{\sigma \tau \lambda \rho }=B_{\nu \mu }^{\lambda \rho \sigma \tau },`$
$`C_{\mu \nu }^{\lambda \rho \sigma \tau }`$ $`=`$ $`C_{\mu \nu }^{\rho \lambda \sigma \tau }=C_{\mu \nu }^{\sigma \rho \lambda \tau }=C_{\mu \nu }^{\sigma \tau \lambda \rho }=C_{\nu \mu }^{\lambda \rho \sigma \tau },`$
$`D^{\rho \sigma \tau }`$ $`=`$ $`D^{\rho \tau \sigma }=D^{\sigma \rho \tau },`$
$`E^{\rho \sigma \tau }`$ $`=`$ $`E^{\rho \tau \sigma }=E^{\sigma \rho \tau },`$
$`F^{\mu \nu }`$ $`=`$ $`F^{\nu \mu },`$
$`K_\mu ^{\sigma \tau }`$ $`=`$ $`K_\mu ^{\tau \sigma },`$
in addition to total tracelessness.
2. In addition for $`N=2`$:
$`\begin{array}{cccccc}\text{Name}\hfill & \text{HW}\hfill & \lambda \hfill & \text{Partition}\hfill & \text{Type}\hfill & \text{Tensor}\hfill \\ \psi _{11}^W,\psi _{12}^W\hfill & \pi _1\hfill & 0\hfill & \{1\}\hfill & (1,0;0)\hfill & S^\rho \hfill \\ \psi _{13}^W\hfill & 5\pi _1\hfill & 2\hfill & \{5\}\hfill & (5,0;2)\hfill & S^{(\rho _1..\rho _5)}\hfill \\ \psi _{14}^W\hfill & 7\pi _1\hfill & 3\hfill & \{7\}\hfill & (7,0;3)\hfill & S^{(\rho _1..\rho _7)}\hfill \end{array}`$ (3.41)
3. In addition for $`N=1`$, there are three cocycles with $`\lambda =1`$, $`4`$, and $`6`$, respectively.
Of course, knowledge of the relevant module is not enough to uniquely describe the cocycle. Dzhumadil’daev has also given formulas for the cocycles, but it is in fact quite easy to reconstruct them from scratch. By writing down manifestly non-trivial cocycles valued in the right modules, we are guaranteed to obtain expressions that are equivalent to Dzhumadil’daev’s, without having to decipher his notation. This is the subject of the rest of this section.
### 3.1 $`\psi _1^W`$
$`\psi _1^W`$ corresponds to the partition $`\{1,0,\mathrm{},0\}`$, i.e. a tensor density $`S^\rho `$ of type $`(1,0;1)`$. This extension was first described in :
$`c(\xi ,\eta )`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu _\nu \eta ^\nu _\mu \xi ^\mu _\rho _\nu \eta ^\nu ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\mu n_\nu (m_\rho n_\rho )S^\rho (m+n).`$
Such a tensor can be identified with a one-chain, which is reducible according to (2.13). It was noted in that (LABEL:Sext) still defines a cocycle when restricted to the submodule of closed one-chains. In one dimension, $`\psi _1^W`$ is related to the Virasoro algebra. We have
$`[L_m,L_n]`$ $`=`$ $`(nm)L_{m+n}+(nm)mnS_{m+n},`$
$`[L_m,S_n]`$ $`=`$ $`(n+m)S_{m+n}.`$
The closedness condition, $`mS_m=0`$, has the unique solution $`S_m=c/24\delta _m`$. Substituting this into (LABEL:Svir) yields the Virasoro algebra with central charge $`c`$.
Let $`X`$ be a function on $`^N`$ and $`E_X`$ a tensor density of type $`(0,0;1)`$. If
$`[_\xi ,E_X]=E_{\xi X},[E_X,E_Y]=S^\rho (X_\rho Y_\rho XY),`$ (3.44)
then $`_\xi ^{}=_\xi +E_{_\mu \xi ^\mu }`$ satisfies $`diff(N)`$ with cocycle $`\psi _1^W`$.
### 3.2 $`\psi _2^W`$
$`\psi _2^W`$ corresponds to the partition $`\{3,1,\mathrm{},1\}`$, i.e. a traceless tensor density $`K_\mu ^{(\sigma \tau )}`$ of type $`(2,1;1)`$:
$`c(\xi ,\eta )`$ $`=`$ $`K_\nu ^{(\sigma \tau )}(_\mu \xi ^\mu _\sigma _\tau \eta ^\nu )K_\mu ^{(\sigma \tau )}(_\sigma _\tau \xi ^\mu _\nu \eta ^\nu ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\mu n_\sigma n_\tau K_\nu ^{(\sigma \tau )}(m+n)n_\nu m_\sigma m_\tau K_\mu ^{(\sigma \tau )}(m+n).`$
Change the weight to some $`\lambda 1`$, i.e. $`K_\nu ^{(\sigma \tau )}`$ is of type $`(2,1;\lambda )`$, and redefine the $`diff(N)`$ generators by
$`L_\mu (m)L_\mu (m)+am_\sigma m_\tau K_\mu ^{(\sigma \tau )}(m).`$ (3.46)
The new generators satisfy $`diff(N)`$ with the extension $`a(1\lambda )c_{\mu \nu }(m,n)`$, which thus is trivial. If we now fix $`a=(1\lambda )^1`$ and let $`\lambda 1`$, $`\psi _2^W`$ is recovered. Tracelessness does not play a role here; setting $`K_\mu ^{(\sigma \tau )}=\delta _\mu ^{(\sigma }S^{\tau )}`$, we see that the trace is of type $`\psi _1^W`$.
### 3.3 $`\psi _3^W`$$`\psi _{10}^W`$
These eight extensions all follow from the following reducible extension
$`c(\xi ,\eta )`$ $`=`$ $`R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}(_\lambda _\rho \xi ^\mu _\sigma _\tau \eta ^\nu ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\lambda m_\rho n_\sigma n_\tau R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}(m+n),`$
where $`R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}`$ is a tensor of type $`(4,2;1)`$, and
$`R_{\nu \mu }^{(\sigma \tau )(\lambda \rho )}=R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}.`$ (3.48)
Such a tensor can be decomposed into irreducible submodules as follows.
$`R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}`$ $`=`$ $`A_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}+B_{(\mu \nu )}^{[(\lambda \rho )(\sigma \tau )]}+C_{[\mu \nu ]}^{((\lambda \rho )(\sigma \tau ))}+`$
$`\delta _\mu ^{(\lambda }G_\nu ^{\rho )(\sigma \tau )}\delta _\nu ^{(\sigma }G_\mu ^{\tau )(\lambda \rho )}+\delta _\nu ^{(\lambda }G_\mu ^{\rho )(\sigma \tau )}\delta _\mu ^{(\sigma }G_\nu ^{\tau )(\lambda \rho )},`$
where
$`G_\nu ^{\rho (\sigma \tau )}=D_\nu ^{(\rho \sigma \tau )}+E_\nu ^{\rho (\sigma \tau )}+\delta _\nu ^{(\sigma }F^{\tau )\rho }+\delta _\nu ^{(\sigma }H^{\tau )\rho }+\delta _\nu ^\rho H^{\sigma \tau },`$
$`C_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}0,E_\nu ^{(\rho \sigma \tau )}0,`$ (3.50)
$`F^{\rho \sigma }=F^{[\rho \sigma ]},H^{\rho \sigma }=H^{(\rho \sigma )}.`$
Substitution of (3.3)–(3.50) into (LABEL:Rext) yields
$$\begin{array}{cc}\text{Cocycle}\hfill & c(\xi ,\eta )\hfill \\ \text{Partition}\hfill & c_{\mu \nu }(m,n)\hfill \\ & \\ \psi _3^W\hfill & F^{[\rho \tau ]}(_\rho _\mu \xi ^\mu _\tau _\nu \eta ^\nu )\hfill \\ \{1,1,0,\mathrm{},0\}\hfill & m_\rho m_\mu n_\tau n_\nu F^{[\rho \tau ]}(m+n)\hfill \\ & \\ \psi _4^W\hfill & F^{[\rho \tau ]}(_\rho _\nu \xi ^\mu _\tau _\mu \eta ^\nu )\hfill \\ \{1,1,0,\mathrm{},0\}\hfill & m_\rho m_\nu n_\tau n_\mu F^{[\rho \tau ]}(m+n)\hfill \\ & \\ \psi _5^W\hfill & E_\nu ^{\rho (\sigma \tau )}(_\rho _\mu \xi ^\mu _\sigma _\tau \eta ^\nu )\xi \eta \hfill \\ \{3,2,1,\mathrm{},1\}\hfill & m_\rho m_\mu n_\sigma n_\tau E_\nu ^{\rho (\sigma \tau )}(m+n)mn\hfill \\ & \\ \psi _6^W\hfill & E_\mu ^{\rho (\sigma \tau )}(_\rho _\nu \xi ^\mu _\sigma _\tau \eta ^\nu )\xi \eta \hfill \\ \{3,2,1,\mathrm{},1\}\hfill & m_\rho m_\nu n_\sigma n_\tau E_\mu ^{\rho (\sigma \tau )}(m+n)mn\hfill \\ & \\ \psi _7^W\hfill & D_\nu ^{(\rho \sigma \tau )}(_\rho _\mu \xi ^\mu _\sigma _\tau \eta ^\nu )\xi \eta \hfill \\ \{4,1,\mathrm{},1\}\hfill & m_\rho m_\mu n_\sigma n_\tau D_\nu ^{(\rho \sigma \tau )}(m+n)mn\hfill \\ & \\ \psi _8^W\hfill & B_{(\mu \nu )}^{[(\lambda \rho )(\sigma \tau )]}(_\lambda _\rho \xi ^\mu _\sigma _\tau \eta ^\nu )\hfill \\ \{5,3,2,\mathrm{},2\}\hfill & m_\lambda m_\rho n_\sigma n_\tau B_{(\mu \nu )}^{[(\lambda \rho )(\sigma \tau )]}(m+n)\hfill \\ & \\ \psi _9^W\hfill & A_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}(_\lambda _\rho \xi ^\mu _\sigma _\tau \eta ^\nu )\hfill \\ \{5,1,\mathrm{},1,0\}\hfill & m_\lambda m_\rho n_\sigma n_\tau A_{[\mu \nu ]}^{(\lambda \rho \sigma \tau )}(m+n)\hfill \\ & \\ \psi _{10}^W\hfill & C_{[\mu \nu ]}^{((\lambda \rho )(\sigma \tau ))}(_\lambda _\rho \xi ^\mu _\sigma _\tau \eta ^\nu )\hfill \\ \{3,3,1,\mathrm{},1,0\}\hfill & m_\lambda m_\rho n_\sigma n_\tau C_{[\mu \nu ]}^{((\lambda \rho )(\sigma \tau ))}(m+n)\hfill \end{array}$$
Splitting $`R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}`$ as in (3.3) and (3.50) suggests that there should be additional cocycles
$`m_\rho m_\nu n_\sigma n_\tau D_\mu ^{(\rho \sigma \tau )}(m+n)mn,`$
(3.51)
$`m_\mu m_\nu n_\sigma n_\tau H^{(\sigma \tau )}(m+n)mn,`$
but these cocycles can be removed by the redefinitions
$`L_\mu (m)`$ $``$ $`L_\mu (m)+m_\rho m_\sigma m_\tau D_\mu ^{(\rho \sigma \tau )}(m),`$
$`L_\mu (m)`$ $``$ $`L_\mu (m)+m_\mu m_\sigma m_\tau H^{(\sigma \tau )}(m),`$
respectively.
The extension (LABEL:Rext) can be understood as a MF term (2.5) for $`gl(N)`$ with $`S_3^{\mu \nu \rho }=0`$. This MF algebra is the extension of $`map(N,gl(N))`$ (LABEL:glNext) with
$`k_{\nu \tau }^{\mu \sigma }(m,n)=m_\rho n_\lambda R_{\nu \tau }^{\mu \rho \sigma \lambda }(m+n),`$ (3.53)
where $`R_{\tau \nu }^{\mu \rho \sigma \lambda }`$ transforms as a tensor of type $`(4,2)`$ under $`gl(N)`$. (LABEL:Rext) now follows immediately from (2.8). Moreover, since $`m_\mu m_\rho m_{(\mu }m_{\rho )}`$, only the part with symmetries (3.48) is relevant.
There is another way to arrive at the extension (LABEL:Rext). An analogous construction was carried out by in the case of arbitrary gauge algebras $`map(N,𝔤)`$. Let $`P_\mu ^{\nu \rho }`$ be a tensor field of type $`(2,1;1)`$. Then
$`_\xi ^{}=_\xi +P_\mu ^{\nu \rho }(_\nu _\rho \xi ^\mu ),L_\mu ^{}(m)=L_\mu (m)+m_\nu m_\rho P_\mu ^{\nu \rho }(m),`$ (3.54)
satisfies an extension of $`diff(N)`$ given by
$`c(\xi ,\eta )`$ $`=`$ $`[P_\mu ^{\lambda \rho }(_\lambda _\rho \xi ^\mu ),P_\nu ^{\sigma \tau }(_\sigma _\tau \eta ^\nu )],`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\lambda m_\rho n_\sigma n_\tau [P_\mu ^{\lambda \rho }(m),P_\nu ^{\sigma \tau }(n)].`$
This is of the form (LABEL:Rext), if we impose the condition that the extension be local in the sense of (2.3). In particular, the symmetry condition (3.48) holds automatically.
### 3.4 $`N=2`$: $`\psi _{11}^W`$$`\psi _{14}^W`$
In two dimensions, the symplectic form $`ϵ_{\mu \nu }`$ coincides with the anti-symmetric symbol, and hence it commutes with all vector fields, not just the Hamiltonian ones. This makes it possible to construct further cocycles:
$$\begin{array}{cc}\text{Cocycle}\hfill & c(\xi ,\eta )\hfill \\ & c_{\mu \nu }(m,n)\hfill \\ & \\ \psi _{11}^W\hfill & S^\rho (_\rho (ϵ^{\sigma \tau }_\sigma _\nu \xi ^\mu _\tau _\mu \eta ^\nu ϵ_{\mu \nu }ϵ^{\sigma \tau }ϵ^{\rho \lambda }_\sigma _\rho \xi ^\mu _\tau _\lambda \eta ^\nu ))\hfill \\ & (m_\nu n_\mu (m\times n)ϵ_{\mu \nu }(m\times n)^2)(m_\rho +n_\rho )S^\rho (m+n)\hfill \\ & \\ \psi _{12}^W\hfill & S^\rho (_\rho (ϵ^{\sigma \tau }_\sigma _\mu \xi ^\mu _\tau _\nu \eta ^\nu ))\hfill \\ & m_\mu n_\nu (m\times n)(m_\rho +n_\rho )S^\rho (m+n)\hfill \\ & \\ \psi _{13}^W\hfill & S^{(\kappa \lambda \rho \sigma \tau )}(3ϵ_{\mu \kappa }ϵ_{\nu \lambda }ϵ^{\alpha \beta }_\alpha _\rho _\sigma \xi ^\mu _\beta _\tau \eta ^\nu 4ϵ_{\mu \nu }_\kappa _\rho _\sigma \xi ^\mu _\lambda _\tau \eta ^\nu )\hfill \\ & \xi \eta \hfill \\ & (3(m\times n)ϵ_{\mu \kappa }ϵ_{\nu \lambda }m_\rho m_\sigma n_\tau 4ϵ_{\mu \nu }m_\kappa m_\rho m_\sigma n_\lambda n_\tau )\hfill \\ & S^{(\kappa \lambda \rho \sigma \tau )}(m+n)mn\hfill \\ & \\ \psi _{14}^W\hfill & S^{(\alpha \beta \kappa \lambda \rho \sigma \tau )}(ϵ_{\mu \alpha }ϵ_{\nu \beta }_\kappa _\rho _\sigma \xi ^\mu _\lambda _\tau \eta ^\nu )\xi \eta \hfill \\ & ϵ_{\mu \alpha }ϵ_{\nu \beta }m_\kappa m_\rho m_\sigma n_\lambda n_\tau S^{(\alpha \beta \kappa \lambda \rho \sigma \tau )}mn\hfill \end{array}$$
As in subsection 2.3, $`ϵ^{12}=ϵ^{21}=ϵ_{12}=ϵ_{21}=1`$ and $`m\times n=m_1n_2m_2n_1=ϵ^{\mu \nu }m_\mu n_\nu `$.
The Jacobi identities, in the Fourier basis, were verified numerically on a computer. To this end, it was useful to write the cocycles as
$`c(\xi ,\eta )`$ $`=`$ $`R(\xi ,\xi ,\xi ,\eta ,\eta )R(\eta ,\eta ,\eta ,\xi ,\xi ),`$ (3.56)
$`c_{\mu \nu }(m,n)`$ $`=`$ $`R_{\mu \nu }(m,m,m,n,n|m+n)R_{\nu \mu }(n,n,n,m,m|m+n),`$
where e.g.
$`R(m,r,s,n,t|u)=m_\kappa r_\lambda s_\rho n_\sigma t_\tau R_{\mu \nu }^{(\kappa \lambda \rho )(\sigma \tau )}(u),`$ (3.57)
and this operator carries conformal weight $`1`$. The Jacobi identities now lead to the conditions
$`n_\mu R_{\nu \sigma }(m,m,n,s,s|m+n+s)+s_\nu R_{\sigma \mu }(n,n,s,m,m|m+n+s)`$
$`+m_\sigma R_{\mu \nu }(s,s,m,n,n|m+n+s)s_\mu R_{\sigma \nu }(m,m,s,n,n|m+n+s)`$ (3.58)
$`m_\nu R_{\mu \sigma }(n,n,m,s,s|m+n+s)n_\sigma R_{\nu \mu }(s,s,n,m,m|m+n+s)`$ $`=`$ $`0.`$
Note how the epsilons conspire to yield the correct assignments of conformal weights: $`ϵ^{\mu \nu }`$ and $`ϵ_{\mu \nu }`$ carry weight $`+1`$ and $`1`$, respectively. In particular, $`S^\rho `$ has weight zero and is therefore not a one-chain, so it is not possible to write $`S^\rho (_\rho F)=\mathrm{\Phi }(F)`$ for $`\mathrm{\Phi }`$ a tensor density.
### 3.5 $`N=1`$
In one dimension vectors have only one component, so we can use the simplified notation $`L_m=L_1(m)`$. A density with weight $`\lambda `$ (often called a primary field) transforms as $`[L_m,A_n]=(n+(1\lambda )m)A_{m+n}`$. Dzhumadil’daev describes the cocycles as follows.
$`\begin{array}{cc}\lambda \hfill & \text{Cocycle}\hfill \\ 1\hfill & L_0L_21\hfill \\ 4\hfill & L_2L_31\hfill \\ 6\hfill & L_2L_51,L_3L_43\hfill \end{array}`$ (3.63)
It is not easy to guess the explicit forms of the cocycles from this description, but fortunately they were given in . This list was later rediscovered by Ovsienko and Roger , and the super generalization has recently be worked out by Marcel . I follow his naming scheme for the cocycles.
$`\begin{array}{ccc}& \lambda & c(m,n)\hfill \\ \gamma _1\hfill & 1& (mn)A_{m+n}\hfill \\ \gamma _2=\psi _1^W\hfill & 0& (m^2nmn^2)A_{m+n}\hfill \\ \gamma _3\hfill & 0& (m^2n^2)A_{m+n}\hfill \\ \gamma _4\hfill & 1& (m^3nmn^3)A_{m+n}\hfill \\ \gamma _5\hfill & 1& (m^3n^3)A_{m+n}\hfill \\ \gamma _6\hfill & 4& (m^3n^4n^3m^4)A_{m+n}\hfill \\ \gamma _7\hfill & 6& (2m^3n^69m^4n^5+9n^4m^52n^3m^6)A_{m+n}\hfill \end{array}`$ (3.72)
For some reason, $`\gamma _1`$, $`\gamma _3`$ and $`\gamma _5`$ are not included in Dzhumadil’daev’s 1996 classification, although they are present in his 1992 paper. Moreover, we have the Virasoro cocycle with values in the trivial module.
Generalize primary fields to translated primary fields:
$`[L_m,A_n]=(n+r+(1\lambda )m)A_{m+n}.`$ (3.73)
Now consider the redefinition
$`L_mL_m^{}=L_m+am^pA_m,`$ (3.74)
where $`a`$ is a parameter. This redefinition gives rise to a trivial cocycle, except when $`\lambda =\lambda _0`$ and $`r=0`$, where $`\lambda _0`$ is the weight in the table above. In this critical case, (3.74) gives rise to no cocycle at all. Now set $`a=1/(\lambda \lambda _0)`$, $`r=0`$, and take the limit $`\lambda \lambda _0`$. This limiting procedure yields the cocycle $`c_\lambda (m,n)`$. Or set $`a=1/r`$, $`\lambda =\lambda _0`$, and take the limit $`r0`$, giving cocycle $`c_r(m,n)`$. The result is
$$\begin{array}{cccc}p& \lambda _0& c_\lambda (m,n)& (m,n)\\ 0& 1& \gamma _1& \\ 1& any& & \gamma _1\\ 2& 0& \gamma _2& \gamma _3\\ 3& 1& \gamma _4& \gamma _5\end{array}$$
In this way, the cocycles $`\gamma _1\gamma _5`$ arise as limits of trivial cocycles. To “explain” $`\gamma _6`$, assume that in the $`p=3`$ case,
$$[A_m,A_n]=(mn)B_{m+n},$$
(“locality”) implying that $`B`$ transforms with $`\lambda =4`$. The same assumption can be made also when $`p=0,1,2`$, but this gives nothing new.
## 4 Dzhumadil’daev’s classification: subalgebra cocycles
### 4.1 Divergence-free vector fields
The cocycles are obtained by restriction from $`diff(N)`$. Since $`_\mu \xi ^\mu =0`$, cocycles $`\psi _1^W`$, $`\psi _2^W`$, $`\psi _3^W`$, $`\psi _6^W`$ and $`\psi _7^W`$ vanish, and the remaining extensions for $`N3`$ are denoted in , Table 2, by
$$\begin{array}{cccccc}svect(N)& \psi _1^S& \psi _2^S& \psi _3^S& \psi _4^S& \psi _5^S\\ & & & & & \\ diff(N)& \psi _4^W& \psi _5^W& \psi _8^W& \psi _9^W& \psi _{10}^W\end{array}$$
The treatment of the special two-dimensional cocycles is deferred to the next subsection, because $`svect(2)Ham(2)`$
### 4.2 Hamiltonian vector fields
$`Ham(N)`$ has nontrivial cocycles in the following modules:
$$\begin{array}{cccccc}\text{Name}\hfill & N\hfill & \text{HW}\hfill & \text{Tensor}\hfill & & \\ \psi _1^H\text{(Moyal)}\hfill & N2\hfill & 0\hfill & \mathrm{\Phi }\hfill & & \\ \psi _2^H\hfill & N4\hfill & \pi _2\hfill & A^{[\mu \nu ]}\hfill & & \\ \psi _3^H\hfill & N4\hfill & 2\pi _2\hfill & B^{[(\mu \rho )(\nu \sigma )]}=\stackrel{~}{B}^{([\mu \nu ][\rho \sigma ])}\hfill & & \\ \psi _4^H\hfill & N4\hfill & 3\pi _2\hfill & C^{[(\mu \rho \lambda )(\nu \sigma \tau )]}=\stackrel{~}{C}^{([\mu \nu ][\rho \sigma ][\lambda \tau ])}\hfill & & \\ \psi _5^H\hfill & N4\hfill & 4\pi _1+\pi _2\hfill & D^{(\rho \lambda \sigma \tau )[\mu \nu ]}\hfill & & \\ \psi _6^H\hfill & N2\hfill & \pi _1\hfill & S^\rho \hfill & & \\ \psi _7^H\hfill & N=2\hfill & 7\pi _1\hfill & S^{(\mu \nu \kappa \lambda \rho \sigma \tau )}\hfill & & \\ \psi _8^H\hfill & N=2\hfill & 2\pi _1\hfill & S^{(\mu \nu )}\hfill & & \end{array}$$
The tensors are demanded to be totally traceless, in the following sense.
$`ϵ_{\mu \nu }A^{[\mu \nu ]}=ϵ_{\mu \nu }B^{[(\mu \rho )(\nu \sigma )]}=ϵ_{\mu \nu }C^{[(\mu \rho \lambda )(\nu \sigma \tau )]}=ϵ_{\mu \nu }D^{(\rho \lambda \sigma \tau )[\mu \nu ]}=0.`$ (4.75)
Explicitly, the cocycles are given by (, Table 4)
$$\begin{array}{cc}\text{Cocycle}\hfill & c_H(f,g)\hfill \\ N\hfill & c_H(m,n)\hfill \\ & \\ \psi _1^H\hfill & \mathrm{\Phi }(ϵ^{\mu \nu }ϵ^{\rho \sigma }ϵ^{\lambda \tau }_\mu _\rho _\lambda f_\nu _\sigma _\tau g)\hfill \\ N2\hfill & (m\times n)^3\mathrm{\Phi }(m+n)\hfill \\ & \\ \psi _2^H\hfill & A^{[\mu \nu ]}(ϵ^{\rho \sigma }ϵ^{\lambda \tau }_\mu _\rho _\lambda f_\nu _\sigma _\tau g)\hfill \\ N2\hfill & (m\times n)^2m_\mu n_\nu A^{[\mu \nu ]}(m+n)\hfill \\ & \\ \psi _3^H\hfill & B^{[(\mu \rho )(\nu \sigma )]}(ϵ^{\lambda \tau }_\mu _\rho _\lambda f_\nu _\sigma _\tau g)\hfill \\ N2\hfill & (m\times n)m_\mu m_\rho n_\nu n_\sigma B^{[(\mu \rho )(\nu \sigma )]}(m+n)\hfill \\ & \\ \psi _4^H\hfill & C^{[(\mu \rho \lambda )(\nu \sigma \tau )]}(_\mu _\rho _\lambda f_\nu _\sigma _\tau g)\hfill \\ N2\hfill & m_\mu m_\rho m_\lambda n_\nu n_\sigma n_\tau C^{[(\mu \rho \lambda )(\nu \sigma \tau )]}(m+n)\hfill \\ & \\ \psi _5^H\hfill & D^{(\rho \lambda \sigma \tau )[\mu \nu ]}(_\mu _\rho _\lambda f_\nu _\sigma _\tau g)\hfill \\ N2\hfill & m_\mu m_\rho m_\lambda n_\nu n_\sigma n_\tau D^{(\rho \lambda \sigma \tau )[\mu \nu ]}(m+n)\hfill \\ & \\ \psi _6^H\hfill & S^\rho (ϵ^{\mu \nu }ϵ^{\kappa \sigma }ϵ^{\lambda \tau }_\rho (_\mu _\kappa _\lambda f_\nu _\sigma _\tau g))\hfill \\ N2\hfill & (m\times n)^3(m_\rho +n_\rho )S^\rho (m+n)\hfill \\ & \\ \psi _7^H\hfill & S^{\mu \nu \kappa \lambda \rho \sigma \tau }(_\mu _\kappa _\rho _\sigma f_\nu _\lambda _\tau g)fg\hfill \\ N=2\hfill & m_\mu m_\kappa m_\rho m_\sigma n_\nu n_\lambda n_\tau S^{\mu \nu \kappa \lambda \rho \sigma \tau }(m+n)mn,\hfill \\ & \\ \psi _8^H\hfill & ϵ^{\kappa \lambda }ϵ^{\rho \pi }ϵ^{\sigma \tau }S^{\mu \nu }(7_\mu _\kappa _\rho _\sigma f_\nu _\lambda _\pi _\tau g+\hfill \\ & +3(_\mu _\nu _\kappa _\rho _\sigma f_\lambda _\pi _\tau g+_\kappa _\rho _\sigma f_\mu _\nu _\lambda _\pi _\tau g))\hfill \\ N=2\hfill & (7m_\mu n_\nu +3(m_\mu m_\nu +n_\mu n_\nu ))(m\times n)^3S^{\mu \nu }(m+n)\hfill \end{array}$$
$`\psi _1^H\psi _5^H`$ arise from restriction of the traceless cocycles of $`\psi _3^W\psi _{10}^W`$, i.e. $`\psi _1^S\psi _5^S`$. The modules appear different because we can use the symplectic form to eliminate all lower indices. Explicitly, we have
$`R_{\mu \nu }^{(\lambda \rho )(\sigma \tau )}`$ $`=`$ $`ϵ_{\mu \alpha }ϵ_{\nu \beta }R^{\alpha (\lambda \rho )\beta (\sigma \tau )},`$
$`R^{\beta (\sigma \tau )\alpha (\lambda \rho )}`$ $`=`$ $`R^{\alpha (\lambda \rho )\beta (\sigma \tau )}.`$
In this way, each traceless module in $`\psi _3^W\psi _{10}^W`$ can be expanded as a direct sum of the modules in $`\psi _1^H\psi _5^H`$. E.g., the field in $`\psi _4^W`$ can be written as $`F^{[\mu \nu ]}=ϵ^{\mu \nu }\mathrm{\Phi }+A^{[\mu \nu ]}`$ with $`ϵ_{\mu \nu }A^{[\mu \nu ]}=0`$, and thus we obtain $`\psi _1^H`$ and $`\psi _2^H`$. The conformal weight is irrelevant for Hamiltonian vector fields, wherefore we can consistently substitute $`\mathrm{\Phi }(\varphi )S^\rho (_\rho \varphi )`$ in $`\psi _1^H`$; this gives $`\psi _6^H`$. $`\psi _6^H`$ also follows by restriction from $`\psi _{11}^W`$ in two dimensions, but exists in all dimensions. For the special two-dimensional cocycles, $`\psi _7^H`$ is the restriction of $`\psi _{14}^W`$, whereas $`\psi _{12}^W`$ is a divergence which restricts to zero. Finally, it was checked numerically that $`\psi _8^H`$ is a cocycle. It may be related to $`\psi _{13}^W`$.
As is well known, the Moyal cocycle $`\psi _1^W`$ can be integrated to a full-fledged deformation of $`Ham(N)`$. Consider the Moyal (or sine) algebra, which is the Lie algebra with brackets (in Fourier basis)
$`[H(m),H(n)]={\displaystyle \frac{1}{\mathrm{}}}sin(\mathrm{}(m\times n))H(m+n).`$ (4.77)
The Moyal cocycle appears at the lowest non-trivial order in $`\mathrm{}`$.
$`[H(m),H(n)]`$ $`=`$ $`(m\times n)H(m+n)+(m\times n)^3\mathrm{\Phi }(m+n),`$
$`[H(m),\mathrm{\Phi }(n)]`$ $`=`$ $`(m\times n)\mathrm{\Phi }(m+n)`$ (4.78)
$`[\mathrm{\Phi }(m),\mathrm{\Phi }(n)]`$ $`=`$ $`0,`$
where $`\mathrm{\Phi }(m)=(\mathrm{}^2/6)H(m)`$.
### 4.3 Contact vector fields
Dzhumadil’daev lists the $`K(N)`$ cocycles in his Table 5. His results for the relevant modules and thee cocycles are
$$\begin{array}{ccccc}\text{Cocycle}\hfill & HW\hfill & N\hfill & & c_K\hfill \\ \psi _1^K\hfill & 0\hfill & N3\hfill & & \widehat{\psi }_1^H\hfill \\ \psi _2^K\hfill & \pi _2\hfill & N5\hfill & & \widehat{\psi }_2^H\hfill \\ \psi _3^K\hfill & \pi _2\hfill & N5\hfill & & \psi _4^W\hfill \\ \psi _4^K\hfill & 2\pi _2\hfill & N5\hfill & & \widehat{\psi }_3^H\hfill \\ \psi _5^K\hfill & 3\pi _2\hfill & N5\hfill & & \widehat{\psi }_4^H\hfill \\ \psi _6^K\hfill & 4\pi _1+\pi _2\hfill & N5\hfill & & \widehat{\psi }_5^H\hfill \\ \psi _7^K\hfill & 2\pi _1+\pi _2\hfill & N5\hfill & & \\ \psi _8^K\hfill & 4\pi _1\hfill & N3\hfill & & \\ \psi _9^K\hfill & \pi _1\hfill & N=3\hfill & & \widehat{\psi }_6^H=\psi _{11}^W\hfill \\ \psi _{10}^K\hfill & \pi _1\hfill & N=3\hfill & & \psi _{12}^W\hfill \\ \psi _{11}^K\hfill & 3\pi _1\hfill & N=3\hfill & & \\ \psi _{12}^K\hfill & 5\pi _1\hfill & N=3\hfill & & \psi _{13}^W\hfill \\ \psi _{13}^K\hfill & 7\pi _1\hfill & N=3\hfill & & \psi _{14}^W\hfill \end{array}$$
$`\psi _7^K`$ and $`\psi _8^K`$ arise by restriction from $`\psi _6^W\psi _7^W`$. An important point is that contact vector fields have non-zero divergence, $`\text{div}K_f_0f`$. Here $`\widehat{\psi }_n^H`$ means that the restriction to the Hamiltonian subalgebra is $`\psi _n^H`$, and $`\psi _n^W`$ that the cocycle is obtained by restriction from to unrestricted diffeomorphism algebra. $`\psi _{11}^K`$ is a special cocycle which seems to be unique to the contact algebra; I have not verified its existence.
However, I fail to understand Dzhumadil’daev’s results on two points.
1. In view of the results in the previous subsection, the eight cocycles $`\psi _1^K\psi _8^K`$ arise by restriction from the eight MF cocycles $`\psi _3^W\psi _{10}^W`$. However, moving all indices upstairs as in (LABEL:Rup) requires the existence of an invariant and invertible two-form $`ϵ_{\mu \nu }`$. This exists for the Hamiltonian algebra but not, as far as I understand, for the contact algebra.
2. $`K(N)`$ contains a $`diff(1)`$ subalgebra obtained by requiring that the function $`f(x^0)`$ depends on $`x^0`$ only. In this case, (2.18) becomes
$`K_f=2f(x^0)_0+f^{}(x^0)x^i_i.`$ (4.79)
These operators generate $`diff(1)`$ and $`K_f`$ is recognized as the expression for a primary field. Upon the restriction $`diff(N)K(N)diff(1)`$, the cocycle $`\psi _1^W`$ for $`diff(N)`$ becomes $`\psi _1^W`$ for $`diff(1)`$. Hence there must exist a nontrivial $`K(N)`$ cocycle with coefficients in $`\pi _1`$, also for $`N5`$.
Nevertheless, almost all cocycles follow by restriction from $`diff(N)`$.
## 5 Beyond tensor modules
### 5.1 Higher-dimensional Virasoro algebras
We start from the tensor extensions $`\psi _3^W`$ and $`\psi _4^W`$, which involve the skew tensor field $`F^{\rho \sigma }`$ of type $`(2,0;1)`$, i.e. a two-chain. However, the extensions have the form
$`\begin{array}{ccc}c(\xi ,\eta )& & c_{\mu \nu }(m,n)\\ F^{\rho \tau }(_\tau (_\mu \xi ^\mu _\rho _\nu \eta ^\nu )),& & m_\mu n_\rho n_\nu (m_\tau +n_\tau )F^{\rho \tau }(m+n),\\ F^{\rho \tau }(_\tau (_\nu \xi ^\mu _\rho _\mu \eta ^\nu )),& & m_\nu n_\rho n_\mu (m_\tau +n_\tau )F^{\rho \tau }(m+n),\end{array}`$ (5.83)
respectively. By (2.13), we can now rewrite the extensions as
$`\begin{array}{ccc}c(\xi ,\eta )& & c_{\mu \nu }(m,n)\\ S^\rho (_\mu \xi ^\mu _\rho _\nu \eta ^\nu ),& & m_\mu n_\rho n_\nu S^\rho (m+n),\\ S^\rho (_\nu \xi ^\mu _\rho _\mu \eta ^\nu ),& & m_\nu n_\rho n_\mu S^\rho (m+n),\end{array}`$ (5.87)
where $`S^\rho `$ is the exact one-chain defined by
$`S^\rho (\varphi _\rho )=F^{\rho \tau }(_\tau \varphi _\rho ),S^\rho (m)=m_\tau F^{\rho \tau }(m).`$ (5.88)
In particular, exact one-chains are also closed, and it turns out that this is enough to satisfy the cocycle condition. Thus, (5.87) defines cocycles, to be denoted by $`\overline{\psi }_3^W`$ and $`\overline{\psi }_4^W`$, provided that
$`S^\rho (_\rho \varphi )0,m_\rho S^\rho (m)0`$ (5.89)
holds identically. $`\overline{\psi }_4^W`$ is the Eswara Rao-Moody cocycle , and $`\overline{\psi }_3^W`$ was first described by myself . Contrary to $`\psi _3^W`$ and $`\psi _4^W`$, these cocycles are defined for all $`N`$ including $`N=1`$, and in one dimension both reduce to the Virasoro cocycle.
There was some confusion in regarding these cocycles. The reason that they are not included in Dzhumadil’daev’s list is that they do not involve tensor modules, but rather submodules thereof. $`\overline{\psi }_3^W`$ and $`\overline{\psi }_4^W`$ arise naturally in toroidal Lie algebras.
### 5.2 Mickelsson-Faddeev
Let $`d_{\kappa \nu \gamma }^{\rho \tau \beta }`$ be totally symmetric under interchange of the pairs $`(\rho ,\kappa )`$, $`(\tau ,\nu )`$, $`(\beta ,\gamma )`$. Such structure constants can be defined in terms of Kronecker deltas, but the interesting case in $`N3`$ dimensions is
$`d_{\kappa \nu \gamma }^{\rho \tau \beta }=ϵ^{\rho \tau \beta \mu _1..\mu _{N3}}ϵ_{\kappa \nu \gamma \mu _1..\mu _{N3}}`$ (5.90)
Then we can add an inhomogeneous term to the transformation law for the $`(4,2;1)`$-type tensor field in (LABEL:Rext).
$`[L_\mu (m),R_{\kappa \nu }^{(\lambda \rho )(\sigma \tau )}(n)]=(4,2;1)+`$
(5.91)
$`+d_{\kappa \nu \mu }^{\rho \tau \alpha }m_\alpha m_\beta S_3^{\lambda \sigma \beta }(m+n)+\mathrm{symm}(\lambda \rho ,\sigma \tau ),`$
where $`\mathrm{symm}(\lambda \rho ,\sigma \tau )`$ stands for the three extra terms needed to give the rhs the appropriate symmetries. This follows immediately by specializing (2.5) to $`𝔤=gl(N)`$ and (2.7). In three dimensions,
$`R_{\kappa \nu }^{(\lambda \rho )(\sigma \tau )}(n)=ϵ^{\lambda \sigma \alpha }d_{\kappa \nu \gamma }^{\rho \tau \beta }\mathrm{\Gamma }_{\alpha \beta }^\gamma (n)+\mathrm{symm}(\lambda \rho ,\sigma \tau ),`$ (5.92)
where $`\mathrm{\Gamma }_{\sigma \tau }^\nu `$ is the connection (2.14). The additional term in (5.91) is new. It can not be embedded in a larger algebra using (3.54), because that would violate the Jacobi identities .
### 5.3 Dzhumadil’daev-Ovsienko-Roger cocycles in higher dimensions
In this subsection I describe higher-dimensional generalizations of the cocycles $`\gamma _1`$$`\gamma _3`$ of (3.72). Since such generalizations contain non-trivial extensions of $`diff(1)`$, these new cocycles are also non-trivial. First tensor modules (2.5) must be generalized to translated tensor modules with dead indices. This concept is best illustrated by an example. If
$`[L_\mu (m),\mathrm{\Phi }_{\sigma \tau }^{\nu \rho }(n)]`$ $`=`$ $`(n_\mu +\lambda m_\mu +r_\mu )\mathrm{\Phi }_{\sigma \tau }^{\nu \rho }(m+n)+`$
$`+\delta _\mu ^\nu m_\lambda \mathrm{\Phi }_{\sigma \tau }^{\lambda \rho }(m+n)m_\sigma \mathrm{\Phi }_{\mu \tau }^{\nu \rho }(m+n),`$
we say that $`\mathrm{\Phi }_{\sigma \tau }^{\nu \rho }(mr)`$ is of type $`(1,1;1)`$ with one dead upper index ($`\rho `$) and one dead lower index ($`\tau `$). The remaining indices are, of course, alive.
1. Consider the redefinition
$`L_\mu (m)`$ $``$ $`L_\mu (m)+aA_\mu (m),`$
$`[L_\mu (m),A_\nu (n)]`$ $`=`$ $`(n_\mu +(1\lambda )m_\mu +r_\mu )A_\nu (m+n).`$
Thus, $`A_\nu (mr)`$ is of type $`(0,0;\lambda )`$ with a dead lower index. The limit $`\lambda 1`$, $`r_\mu =0`$, $`a(1\lambda )=1`$, gives rise to the cocycle
$`c(\xi ,\eta )`$ $`=`$ $`A_\rho (_\mu \xi ^\mu \eta ^\rho _\nu \eta ^\nu \xi ^\rho ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\mu A_\nu (m+n)n_\nu A_\mu (m+n),`$
which is a higher-dimensional generalization of $`\gamma _1`$. The limit $`r_\mu 0`$, $`\lambda =1`$, $`ar_\mu =e_\mu `$, yields
$`c(\xi ,\eta )`$ $`=`$ $`A_\rho (e_\mu \xi ^\mu \eta ^\rho e_\nu \eta ^\nu \xi ^\rho ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`e_\mu A_\nu (m+n)e_\nu A_\mu (m+n),`$
This cocycle vanishes when $`N=1`$.
2. Consider the redefinition $`L_\mu (m)L_\mu (m)+am_\nu B_\mu ^\nu (m)`$, where either $`B_\mu ^\nu (mr)`$ is of type $`(1,0;\lambda )`$ with a dead lower index, or it is of type $`(0,1;\lambda )`$ with a dead upper index. The limit $`\lambda 1`$, $`r_\mu =0`$, $`a(1\lambda )=1`$, gives rise to the cocycle
$`c(\xi ,\eta )`$ $`=`$ $`B_\sigma ^\rho (_\mu \xi ^\mu _\rho \eta ^\sigma _\nu \eta ^\nu _\rho \xi ^\sigma ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`m_\mu n_\rho B_\nu ^\rho (m+n)n_\nu m_\rho B_\mu ^\rho (m+n).`$
This cocycle vanishes when $`N=1`$. The limit $`r_\mu 0`$, $`\lambda =1`$, $`ar_\mu =e_\mu `$, yields a trivial cocycle.
3. Consider the redefinition $`L_\mu (m)L_\mu (m)+am_\rho m_\sigma K_\mu ^{\rho \sigma }(m)`$, where $`K_\mu ^{\rho \sigma }(mr)`$ is of type $`(2,1;\lambda )`$, symmetric and $`\rho `$ and $`\sigma `$, and all indices are alive. As described above, the limit $`\lambda 1`$, $`r_\mu =0`$, $`a(1\lambda )=1`$, gives rise to the cocycles $`\psi _1^W`$ and $`\psi _2^W`$, for the trace and traceless parts, respectively. These are higher-dimensional generalizations of $`\gamma _2`$. The limit $`r_\mu 0`$, $`\lambda =1`$, $`ar_\mu =e_\mu `$, yields
$`c(\xi ,\eta )`$ $`=`$ $`K_\rho ^{\sigma \tau }(_\sigma _\tau \xi ^\mu e_\nu \eta ^\nu _\sigma _\tau \eta ^\nu e_\mu \xi ^\mu ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`e_\nu m_\sigma m_\tau K_\mu ^{\sigma \tau }(m+n)e_\mu n_\sigma n_\tau K_\nu ^{\sigma \tau }(m+n),`$
which is a higher-dimensional generalization of $`\gamma _3`$. With $`K_\rho ^{\sigma \tau }=\delta _\rho ^{(\tau }S^{\sigma )}`$, we obtain
$`c(\xi ,\eta )`$ $`=`$ $`S^\sigma (_\sigma _\mu \xi ^\mu e_\nu \eta ^\nu _\sigma _\nu \eta ^\nu e_\mu \xi ^\mu ),`$
$`c_{\mu \nu }(m,n)`$ $`=`$ $`(e_\nu m_\sigma m_\mu e_\mu n_\sigma n_\nu )S^\sigma (m+n),`$
where $`S^\rho `$ is of type $`(1,0;1)`$.
4. The natural way to generalize $`\gamma _4`$ and $`\gamma _5`$ would be to redefine $`L_\mu (m)L_\mu (m)+am_\rho m_\sigma m_\tau D_\mu ^{\rho \sigma \tau }(m)`$, where $`D_\mu ^{\rho \sigma \tau }(mr)`$ is of type $`(3,1;\lambda )`$ and totally symmetric. However, as noted in (LABEL:3redef), this gives rise to a trivial cocycle even when $`\lambda =1`$ and $`r_\mu =0`$, except in one dimension. Hence I suspect that $`\gamma _4`$ and $`\gamma _5`$ have no $`N>1`$ counterparts.
### 5.4 Anisotropic extensions
In I constructed two complicated cocycles satisfied by the representations introduced by Eswara-Rao and Moody . It turned out that they could be obtained from the DRO algebra defined below (section 9), by imposing the second-class constraint $`L_f0`$, $`q^0(t)t`$. The former conditions can be viewed as a first class constraint and the latter as a gauge condition. Other cocycles can be found by replacing the gauge condition, as long as the constraints together are second class, i.e. the Poisson bracket matrix is invertible.
## 6 Extensions of $`diff(N)map(N,diff(d))`$
Replace $`N`$ by $`N+d`$ everywhere in the previous sections. The total space $`^{N+d}`$ have coordinates $`z^A=(x^\mu ,y^i)`$, where greek indices $`\mu ,\nu ,\rho ,\sigma ,\tau =1,\mathrm{},N`$ label horizontal (base space) directions, latin indices $`i,j,k,\mathrm{}=1,\mathrm{},d`$ label vertical (target space) directions, and capitals $`A=(\mu ,i)`$, etc. label directions in total space. This induces splits $`_A/z^A=(_\mu ,_i)(/x^\mu ,/y^i)`$, $`\mathrm{\Xi }^A(z)=(\xi ^\mu (x),X^i(x,y))`$, $`_\mathrm{\Xi }=(_\xi ,𝒥_X)`$, etc. What makes this split a proper embedding is that the horizontal components of the vector fields $`\xi ^\mu (x)`$ are taken to be independent of the vertical coordinates $`y^i`$, so $`_i\xi ^\mu =0`$.
An extension of $`diff(N)map(N,diff(d))`$ has the form
$`[_\xi ,_\eta ]`$ $`=`$ $`_{[\xi ,\eta ]}+c(\xi ,\eta ),`$
$`[_\xi ,𝒥_X]`$ $`=`$ $`𝒥_{\xi ^\mu _\mu X}+c(\xi ,X),`$
$`[𝒥_X,𝒥_Y]`$ $`=`$ $`𝒥_{[X,Y]}+c(X,Y).`$
Tensor densities are described by (2.5), $`_\mathrm{\Xi }=\mathrm{\Xi }^A_A+_B\mathrm{\Xi }^AT_A^B`$, where $`T_B^A`$ satisfy $`gl(N+d)`$. Hence
$`_\xi `$ $`=`$ $`\xi ^\mu _\mu +_\nu \xi ^\mu T_\mu ^\nu ,`$
$`𝒥_X`$ $`=`$ $`X^i_i+_jX^iT_i^j+_\mu X^iT_i^\mu .`$
Since the $`T_\nu ^i`$ component never enters any formulas, its value is unimportant and may be set to zero. We can then perform a similarity transformation $`T_B^AT_B^A=\stackrel{~}{S}_C^AT_D^CS_B^D`$, with
$`S_B^A=\left(\begin{array}{cc}\delta _\nu ^\mu & 0\\ 0& \epsilon \delta _j^i\end{array}\right),\stackrel{~}{S}_B^A=\left(\begin{array}{cc}\delta _\nu ^\mu & 0\\ 0& \epsilon ^1\delta _j^i\end{array}\right),T_B^A=\left(\begin{array}{cc}T_\nu ^\mu & \epsilon T_j^\mu \\ 0& T_j^i\end{array}\right).`$ (6.102)
This amounts to multiplying the last term in (LABEL:mtensor) by $`\epsilon `$. For convenience, the transformation laws for a tensor field of type $`(1,1;1)`$ in base space and $`(1,1)`$ in target space is given explicitly; the general case follows readily.
$`[_\xi ,\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(\varphi _{\sigma k}^\tau \mathrm{})]`$ $`=`$ $`\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(\xi ^\mu _\mu \varphi _{\sigma k}^\tau \mathrm{}+_\sigma \xi ^\mu \varphi _{\mu k}^\tau \mathrm{}_\nu \xi ^\tau \varphi _{\sigma k}^\nu \mathrm{}),`$ (6.103)
$`[𝒥_X,\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(\varphi _{\sigma k}^\tau \mathrm{})]`$ $`=`$ $`\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(X^i_i\varphi _{\sigma k}^\tau \mathrm{}+_kX^i\varphi _{\sigma i}^\tau \mathrm{}_jX^{\mathrm{}}\varphi _{\sigma k}^{\tau j}+`$
$`+\epsilon _\sigma X^i\varphi _{ik}^\tau \mathrm{}),`$
where $`\varphi _{\sigma k}^\tau \mathrm{}(x,y)`$ is an arbitrary function on total space. Of course, a similarity transformation does not bring anything essentially new, but we can set $`\epsilon =0`$ in (LABEL:targtens), corresponding to a singular matrix $`S_B^A`$. Base space and target space indices then decouple which makes the transformation laws particularly simple.
The similarity transformation amounts to a rescaling of $`_\sigma X^i`$ by $`\epsilon `$ without affecting other components of $`_B\mathrm{\Xi }^A`$. This is equivalent to rescaling $`X^i`$ by $`\epsilon `$ and $`_j`$ by $`\epsilon ^1`$. $`_j\xi ^\mu `$ would also rescale by $`\epsilon ^1`$, but this is no problem since it vanishes anyway. What is a problem is that $`_j_kX^i`$ also rescales by $`\epsilon ^1`$. Since all cocycles contain such terms, we can in fact not put $`\epsilon =0`$, but it will become possible in the next section. For the remainder of this section, we set $`\epsilon =1`$.
The restrictions of the generic extensions are as follows.
$`\overline{\psi }_3^W`$:
$`c(\xi ,X)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu _iX^i),`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho _iX^i_jY^j)+S^k(_k_iX^i_jY^j).`$
$`\overline{\psi }_4^W`$:
$`c(\xi ,X)`$ $`=`$ $`0,`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho _jX^i_iY^j)+S^k(_k_jX^i_iY^j).`$
Here, $`S^C(\varphi _C)=S^\rho (\varphi _\rho )+S^k(\varphi _k)`$ is a tensor density in total space of type $`(1,0;1)`$, satisfying the auxiliary condition $`S^C(_C\varphi )0`$. Explicitly, the transformation laws are given by
$`[_\xi ,S^\rho (\varphi _\rho )]`$ $`=`$ $`S^\rho (\xi ^\mu _\mu \varphi _\rho +_\rho \xi ^\mu \varphi _\mu ),`$
$`[_\xi ,S^k(\varphi _k)]`$ $`=`$ $`S^k(\xi ^\mu _\mu \varphi _k),`$
$`[𝒥_X,S^\rho (\varphi _\rho )]`$ $`=`$ $`S^\rho (X^i_i\varphi _\rho ),`$
$`[𝒥_X,S^k(\varphi _k)]`$ $`=`$ $`S^k(X^i_i\varphi _k+_kX^i\varphi _i)+S^\rho (_\rho X^i\varphi _i),`$
where $`S^\rho (_\rho \varphi )+S^k(_k\varphi )0`$.
$`\psi _1^W`$:
$`c(\xi ,X)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu _iX^i_\mu \xi ^\mu _\rho _iX^i)S^k(_\mu \xi ^\mu _k_iX^i),`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho _iX^i_jY^j_iX^i_\rho _jY^j)`$
$`+S^k(_k_iX^i_jY^j_iX^i_k_jY^j).`$
where $`S^C`$ is as above but the closedness condition is no longer necessary.
$`\psi _2^W`$:
$`c(\xi ,X)`$ $`=`$ $`K_i^{(AB)}(_\mu \xi ^\mu _A_BX^i)K_\mu ^{(\sigma \tau )}(_\sigma _\tau \xi ^\mu _iX^i),`$
$`c(X,Y)`$ $`=`$ $`K_j^{(AB)}(_iX^i_A_BY^j)K_i^{(AB)}(_A_BX^i_jY^j),`$
where $`K_C^{(AB)}`$ is a tensor field of type $`(2,1;1)`$.
$`\psi _3^W`$$`\psi _{10}^W`$:
$`c(\xi ,X)`$ $`=`$ $`R_{\mu i}^{(\lambda \rho )(CD)}(_\lambda _\rho \xi ^\mu _C_DX^i),`$
$`c(X,Y)`$ $`=`$ $`R_{ij}^{(AB)(CD)}(_A_BX^i_C_DY^j),`$
where $`R_{EF}^{(AB)(CD)}`$ is a tensor field of type $`(4,2;1)`$.
## 7 Extensions of $`diff(N)map(N,gl(d))`$
Now consider the subalgebra $`gl(d)diff(d)`$, with vertical vector fields $`X=X^i(x,y)_i=X_j^i(x)y^j_i`$. In the previous section, we substitute $`_jX^i=X_j^i`$, $`_j_kX^i=0`$. The algebra formally takes the same form (LABEL:LxX), but now $`[X,Y]=(X_k^iY_j^kX_j^kY_k^i)y^j_i`$. Tensor fields are decomposed into components which are homogeneous in $`y^i`$, e.g.,
$`\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(\varphi _{\sigma k}^\tau \mathrm{})={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\mathrm{\Phi }_\tau \mathrm{}^{\sigma k|m_1..m_n}(\varphi _{\sigma k|m_1..m_n}^\tau \mathrm{}),`$ (7.111)
where $`\varphi _{\sigma k|m_1..m_n}^\tau \mathrm{}(x)`$ is a function independent of the vertical coordinate $`y^i`$ and
$`\mathrm{\Phi }_\tau \mathrm{}^{\sigma k|m_1..m_n}()\mathrm{\Phi }_\tau \mathrm{}^{\sigma k}(y^{m_1}\mathrm{}y^{m_n}).`$ (7.112)
The base space transformation law (6.103) is unchanged, whereas (LABEL:targtens) is replaced by
$`[𝒥_X,\mathrm{\Phi }_\tau \mathrm{}^{\sigma k|m_1..m_n}(\varphi _{\sigma k|m_1..m_n}^\tau \mathrm{})]=\mathrm{\Phi }_\tau \mathrm{}^{\sigma k|m_1..m_n}(X_k^i\varphi _{\sigma i|m_1..m_n}^\tau \mathrm{}`$ (7.113)
$`X_j^{\mathrm{}}\varphi _{\sigma k|m_1..m_n}^{\tau j}+{\displaystyle \underset{r=1}{\overset{n}{}}}X_{m_r}^i\varphi _{\sigma k|m_1..i..m_n}^\tau \mathrm{})+\epsilon \mathrm{\Phi }^{\sigma k|m_1..m_nj}_\tau \mathrm{}(_\sigma X_j^i\varphi _{ik|m_1..m_n}^\tau \mathrm{}).`$
In this section we can set $`\epsilon =0`$, since the dangerous term $`_j_kX^i=0`$ anyway. Because $`X_j^i=_jX^i`$, this amounts to rescalings of $`y^i`$ by $`\epsilon `$ and of $`_j`$ by $`\epsilon ^1`$, and hence $`\mathrm{\Phi }_\tau \mathrm{}^{\sigma k|m_1..m_n}()`$ must be multiplied by $`\epsilon ^n`$.
With any value of $`\epsilon `$, transformation laws are readily read off from the index structure. In particular, target space indices transform in the same way independent of if they appear to the left or to the right of a vertical bar.
$`\overline{\psi }_3^W`$:
$`c(\xi ,X)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu X_i^i),`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho X_i^iY_j^j).`$
$`\overline{\psi }_4^W`$:
$`c(\xi ,X)`$ $`=`$ $`0,`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho X_j^iY_i^j).`$
Note that only the horizontal component of the one-form $`S^C(\varphi _C)`$ appears, and that its argument is independent of $`y^i`$. Therefore, we can limit our attention to $`S^\rho (\varphi _\rho )`$, where $`\varphi _\rho (x)`$ is independent of the vertical coordinates and $`\varphi _i=0`$. The transformation laws read
$`[_\xi ,S^\rho (\varphi _\rho )]`$ $`=`$ $`S^\rho (\xi ^\mu _\mu \varphi _\rho +_\rho \xi ^\mu \varphi _\mu ),`$
$`[𝒥_X,S^\rho (\varphi _\rho )]`$ $`=`$ $`0,`$
where $`S^\rho (_\rho \varphi )0`$.
$`\psi _1^W`$:
$`c(\xi ,X)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu X_i^i_\mu \xi ^\mu _\rho X_i^i),`$
$`c(X,Y)`$ $`=`$ $`S^\rho (_\rho X_i^iY_j^jX_i^i_\rho Y_j^j).`$
where $`S^\rho `$ is as above but the closedness condition is no longer necessary.
$`\psi _2^W`$:
$`c(\xi ,X)`$ $`=`$ $`\epsilon K_i^{(\sigma \tau )|j}(_\mu \xi ^\mu _\sigma _\tau X_j^i)+2K_i^{(\sigma j)}(_\mu \xi ^\mu _\sigma X_j^i)`$ (7.118)
$`K_\mu ^{(\sigma \tau )}(_\sigma _\tau \xi ^\mu X_i^i),`$
$`c(X,Y)`$ $`=`$ $`\epsilon K_j^{(\sigma \tau )|k}(X_i^i_\sigma _\tau Y_k^j)+2K_j^{(\sigma k)}(X_i^i_\sigma Y_k^j)XY,`$
where $`K_i^{(\sigma \tau )|j}()=K_i^{(\sigma \tau )}(y^j)`$. The two cocycles that survive when $`\epsilon =0`$ are independent.
$`\psi _3^W`$$`\psi _{10}^W`$:
$`c(\xi ,X)`$ $`=`$ $`\epsilon R_{\mu i}^{(\lambda \rho )(\sigma \tau )|j}(_\lambda _\rho \xi ^\mu _\sigma _\tau X_j^i)+2R_{\mu i}^{(\lambda \rho )(\sigma j)}(_\lambda _\rho \xi ^\mu _\sigma X_j^i),`$
$`c(X,Y)`$ $`=`$ $`\epsilon ^2R_{ij}^{(\lambda \rho )(\sigma \tau )|k\mathrm{}}(_\lambda _\rho X_k^i_\sigma _\tau Y_{\mathrm{}}^j)+`$
$`+2\epsilon (R_{ij}^{(k\rho )(\sigma \tau )|\mathrm{}}(_\rho X_k^i_\sigma _\tau Y_{\mathrm{}}^j)+R_{ij}^{(\lambda \rho )(\mathrm{}\tau )|k}(_\lambda _\rho X_k^i_\tau Y_{\mathrm{}}^j))+`$
$`+4R_{ij}^{(k\rho )(\mathrm{}\tau )}(_\rho X_k^i_\tau Y_{\mathrm{}}^j),`$
where
$`R_{\mu i}^{(\lambda \rho )(\sigma \tau )|j}()`$ $`=`$ $`R_{\mu i}^{(\lambda \rho )(\sigma \tau )}(y^j),`$
$`R_{ij}^{(\lambda \rho )(\sigma \tau )|k\mathrm{}}()`$ $`=`$ $`R_{\mu i}^{(\lambda \rho )(\sigma \tau )}(y^ky^{\mathrm{}}),`$
$`R_{ij}^{(k\rho )(\sigma \tau )|\mathrm{}}()`$ $`=`$ $`R_{ij}^{(k\rho )(\sigma \tau )}(y^{\mathrm{}}).`$
The two cocycles that survive when $`\epsilon =0`$ are independent, and the the last term in (7) is recognized as the MF cocycle (2.5) for $`map(N,gl(d))`$.
## 8 Extensions of $`diff(N)map(N,𝔤)`$
Assume that the finite-dimensional Lie algebra $`𝔤`$ has a $`d`$-dimensional representation with matrices $`\sigma ^a=(\sigma _j^{ia})`$. In the previous section, we substitute $`X_j^i=X_a\sigma _j^{ia}`$. Set $`\text{tr}\sigma ^a=\sigma _i^{ia}=z_M\delta ^a`$, where either $`\delta ^cf^{ab}{}_{c}{}^{}=0`$ or $`z_M=0`$, and $`\text{tr}\sigma ^a\sigma ^b=\sigma _j^{ia}\sigma _i^{jb}=y_M\delta ^{ab}`$. Now $`[X,Y]_c=if^{ab}{}_{c}{}^{}X_{a}^{}Y_b`$, $`X_i^i=z_M\delta ^aX_a`$ and $`X_j^iY_i^j=y_M\delta ^{ab}X_aY_b`$. Tensor fields are given by
$`_\xi `$ $`=`$ $`\xi ^\mu _\mu +_\nu \xi ^\mu T_\mu ^\nu ,`$
$`𝒥_X`$ $`=`$ $`X_a\sigma _j^{ia}y^j_i+X_a\sigma _j^{ia}T_i^j+\epsilon _\mu X_a\sigma _j^{ia}T_i^\mu .`$
$`\overline{\psi }_3^W`$:
$`c(\xi ,X)`$ $`=`$ $`z_M\delta ^aS^\rho (_\rho _\mu \xi ^\mu X_a),`$
$`c(X,Y)`$ $`=`$ $`z_M^2\delta ^a\delta ^bS^\rho (_\rho X_aY_b).`$
$`\overline{\psi }_4^W`$:
$`c(\xi ,X)`$ $`=`$ $`0,`$
$`c(X,Y)`$ $`=`$ $`y_M\delta ^{ab}S^\rho (_\rho X_aY_b).`$
In particular, in one dimension we get
$`[L_m,J_n^a]`$ $`=`$ $`nJ_{m+n}^a+z_M\delta ^am^2\delta _{m+n},`$
$`[J_m^a,J_n^b]`$ $`=`$ $`if^{ab}{}_{c}{}^{}J_{m+n}^{c}+z_M^2\delta ^a\delta ^bm\delta _{m+n}+y_M\delta ^{ab}m\delta _{m+n}.`$
The last term is recognized as the Kac-Moody cocycle. The other two are not so well known, because they vanish for $`𝔤`$ semisimple. However, all three cocycles are non-trivial.
$`\psi _1^W`$:
$`c(\xi ,X)`$ $`=`$ $`z_M\delta ^aS^\rho (_\rho _\mu \xi ^\mu X_a_\mu \xi ^\mu _\rho X_a),`$
$`c(X,Y)`$ $`=`$ $`z_M^2\delta ^a\delta ^bS^\rho (_\rho X_aY_bX_a_\rho Y_b).`$
$`\psi _2^W`$:
$`c(\xi ,X)`$ $`=`$ $`\epsilon \sigma _j^{ia}K_i^{(\sigma \tau )|j}(_\mu \xi ^\mu _\sigma _\tau X_a)+2\sigma _j^{ia}K_i^{(\sigma j)}(_\mu \xi ^\mu _\sigma X_a)`$ (8.126)
$`z_M\delta ^aK_\mu ^{(\sigma \tau )}(_\sigma _\tau \xi ^\mu X_a),`$
$`c(X,Y)`$ $`=`$ $`\epsilon z_M\delta ^a\sigma _k^{jb}K_j^{(\sigma \tau )|k}(X_a_\sigma _\tau Y_b)+`$
$`+2\delta ^a\sigma _k^{jb}K_j^{(\sigma k)}(X_a_\sigma Y_b)XY.`$
The two cocycles that survive when $`\epsilon =0`$ are independent.
$`\psi _3^W`$$`\psi _{10}^W`$:
$`c(\xi ,X)`$ $`=`$ $`\epsilon \sigma _j^{ia}R_{\mu i}^{(\lambda \rho )(\sigma \tau )|j}(_\lambda _\rho \xi ^\mu _\sigma _\tau X_a)+2\sigma _j^{ia}R_{\mu i}^{(\lambda \rho )(\sigma j)}(_\lambda _\rho \xi ^\mu _\sigma X_a),`$
$`c(X,Y)`$ $`=`$ $`\epsilon ^2\sigma _k^{ia}\sigma _{\mathrm{}}^{jb}R_{ij}^{(\lambda \rho )(\sigma \tau )|k\mathrm{}}(_\lambda _\rho X_a_\sigma _\tau Y_b)+`$
$`+2\epsilon \sigma _k^{ia}\sigma _{\mathrm{}}^{jb}(R_{ij}^{(k\rho )(\sigma \tau )|\mathrm{}}(_\rho X_a_\sigma _\tau Y_b)+`$
$`+R_{ij}^{(\lambda \rho )(\mathrm{}\tau )|k}(_\lambda _\rho X_a_\tau Y_b))+`$
$`+4\sigma _k^{ia}\sigma _{\mathrm{}}^{jb}R_{ij}^{(k\rho )(\mathrm{}\tau )}(_\rho X_a_\tau Y_b).`$
The two cocycles that survive when $`\epsilon =0`$ are independent, and the latter is recognized as the MF cocycle (2.5) for $`map(N,𝔤)`$.
## 9 Extensions of $`diff(N)diff(1)`$: DRO algebra
The DRO (Diffeomorphism, Reparametrization, Observer) algebra $`DRO(N)`$ was introduced in as an extension of $`diff(N)diff(1)`$ by the observer’s trajectory $`q^\mu (t)`$. The reason for giving this algebra a special name is its importance for Fock representations of $`diff(N)`$. Expand all fields in a Taylor series around $`q^\mu (t)`$, where $`tS^1`$. The Taylor coefficients, or jets, are
$`\mathrm{\Phi }_{,𝐦}(t)=_𝐦\mathrm{\Phi }(q(t))_1^{m_1}\mathrm{}_N^{m_N}\mathrm{\Phi }(q(t)).`$ (9.128)
where $`𝐦=(m_1,\mathrm{},m_N)`$ is a multi-index. Note that the jets depend on $`t`$ although the field $`\mathrm{\Phi }(x)`$ does not, since this dependence enters through the expansion point. In I took the space of $`p`$-jets $`\mathrm{\Phi }_{,𝐦}(t)`$, with $`|𝐦|=_{\mu =1}^Nm_\mu p`$, as the starting point for the Fock construction. This leads to consistent results because the jet space consists of finitely many functions of a single variable $`t`$. The full DRO algebra acts naturally on the jets; the additional $`diff(1)`$ factor describes reparametrizations of the observer’s trajectory.
Any extension of $`diff(N)diff(1)`$ has the form
$`[_\xi ,_\eta ]`$ $`=`$ $`_{[\xi ,\eta ]}+c(\xi ,\eta ),`$
$`[_\xi ,L_f]`$ $`=`$ $`c(\xi ,f),`$
$`[L_f,L_g]`$ $`=`$ $`L_{[f,g]}+c(f,g),`$
where $`f=f(t)d/dt`$ is a vector field on the circle and $`[f,g]=(f\dot{g}g\dot{f})d/dt`$.
We embed $`diff(N)diff(1)diff(N+1)`$ in the natural way: set $`z^A(z^\mu ,z^0)=(x^\mu ,t)`$, $`_A=(_\mu ,d/dt)`$, $`\mathrm{\Xi }^A(z)=(\xi ^\mu (x),f(t))`$, $`_\mathrm{\Xi }=(_\xi ,L_f)`$. Tensor densities restrict to
$`_\xi `$ $`=`$ $`\xi ^\mu _\mu +_\nu \xi ^\mu T_\mu ^\nu ,`$
$`L_f`$ $`=`$ $`f{\displaystyle \frac{d}{dt}}+\dot{f}T_0^0,`$
where $`T_0^0`$ was called the causal weight in . Thus, both the $`T_\mu ^0`$ and $`T_0^\mu `$ components of the $`gl(N+1)`$ generator $`T_B^A`$ decouple.
$`\overline{\psi }_3^W`$:
$`c(\xi ,f)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu \dot{f}),`$
$`c(f,g)`$ $`=`$ $`S^0(\ddot{f}\dot{g}).`$
$`\overline{\psi }_4^W`$:
$`c(\xi ,f)`$ $`=`$ $`0,`$
$`c(f,g)`$ $`=`$ $`S^0(\ddot{f}\dot{g}).`$
With $`\varphi (x)`$ independent of $`t`$ and $`f(t)`$ independent of $`x^\mu `$, closedness implies
$`S^\rho (_\rho \varphi f)+S^0(\varphi \dot{f})0.`$ (9.133)
In particular, $`S^0(\dot{f})0`$, so $`S^0(f)𝑑tf(t)`$ and $`c(f,g)`$ is the Virasoro cocycle in both cases.
Thus $`DRO(N)`$ has four independent Virasoro-like cocycles, namely the terms proportional to $`c_1`$, $`c_2`$, $`c_3`$ and $`c_4`$ in the notation of . In the notation of the present paper, $`c_1=\psi _4^W`$, $`c_2=\psi _3^W`$, $`c_3=c(\xi ,f)`$ from (LABEL:DRO3), and $`c_4=c(f,g)`$ from (LABEL:DRO3) or (LABEL:DRO4). As described in subsection 5.4, we can eliminate reparametrizations by a second class constraint, trading the last two cocycles for anisotropic cocycles of $`diff(N)`$.
$`\psi _1^W`$:
$`c(\xi ,f)`$ $`=`$ $`S^\rho (_\rho _\mu \xi ^\mu \dot{f})S^0(_\mu \xi ^\mu \ddot{f}),`$
$`c(f,g)`$ $`=`$ $`S^0(\ddot{f}\dot{g}\dot{f}\ddot{g}).`$
where (9.133) no longer holds. The second formula is the Virasoro generalization (LABEL:Svir).
$`\psi _2^W`$:
$`c(\xi ,f)`$ $`=`$ $`K_0^{00}(_\mu \xi ^\mu \ddot{f})K_\mu ^{(\sigma \tau )}(_\sigma _\tau \xi ^\mu \dot{f}),`$
$`c(f,g)`$ $`=`$ $`K_0^{00}(\dot{f}\ddot{g}\ddot{f}\dot{g}).`$
$`\psi _3^W`$$`\psi _{10}^W`$:
$`c(\xi ,f)`$ $`=`$ $`R_\mu ^{(\lambda \rho )}(_\lambda _\rho \xi ^\mu \ddot{f}),`$
$`c(f,g)`$ $`=`$ $`0.`$
where $`R_\mu ^{(\lambda \rho )}`$ is a tensor field of type $`(3,1;1)`$.
## 10 Conclusion
In this paper I have reviewed Dzhumadil’daev’s exhaustive classification of tensor extensions of $`diff(N)`$ and subalgebras, extended it beyond tensor modules, and studied the chain of restrictions down to $`map(N,𝔤)`$. The method proves existence for the cocycles of the subalgebras, but it neither proves non-triviality nor exhaustion. However, since the extensions obtained in the last step are in fact recognized as non-trivial (Kac-Moody, MF, etc.), the entire chain is non-trivial. Moreover, I tautologically exhaust the class of subalgebra cocycles with values in tensor modules, which can be lifted to the diffeomorphism algebra in total space.
The construction of projective Fock modules of $`diff(N)`$ was initiated in and further developped in . By restriction, this gives Fock modules of subalgebras, of the type described in and in the papers just cited. Berman and Billig constructed another type of module, postulating the two cocycles $`\overline{\psi }_3^W`$ and $`\overline{\psi }_4^W`$ from the outset. It seems likely that a deep generalization of their modules exists, if one starts from the four inequivelent Virasoro-like extensions of $`DRO(N)`$ instead. The $`diff(1)`$ factor should then provide the necessary extra-grading. Finally, Fock modules for extensions of current algebras that are similar to, but different from, the Mickelsson-Faddeev algebra have recently been constructed .
This work can be extended in several directions. One can consider subalgebras of $`diff(N)`$ such as algebras of divergence-free, Hamiltonian or contact vector fields, or superize by letting some coordinates become fermionic. I expect no essential difficulties here, except that I am not aware of any classification of extensions of superdiffeomorphism algebras.
## Acknowledgments
I am grateful A. Dzhumadil’daev for explaining his results to me, in particular the special two-dimensional cocycles.
|
warning/0002/astro-ph0002333.html
|
ar5iv
|
text
|
# A Close-Separation Double Quasar Lensed by a Gas-Rich Galaxy 1footnote 11footnote 1Based on observations with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract No. NAS5-26555.
## 1 Introduction
The study of gravitationally lensed quasars has become a powerful tool for addressing a number of astrophysical questions. In particular, concentrating on studying the lensing objects themselves provides a sample of distant galaxies selected by mass rather than by light (Kochanek et al. 1999). Because the component separations scale with the square root of the mass of the lens, sampling the low end of the lens mass function becomes difficult from the ground, particularly in the optical/IR, for separations $`1\mathrm{}`$. This observational bias leads to a preponderance of massive spheroidal galaxies in the present sample of lenses and at least partly accounts for the relative lack of known close-separation lenses, which are predicted to exist by theoretical models of the lensing phenomenon (e.g. Maoz & Rix 1993; Rix et al. 1994; Jain et al. 1999). There are currently only seven systems with separations $`0\stackrel{}{\mathrm{.}}9`$ out of 43 confirmed lensed quasars listed by Kochanek et al. (1998).
We are now well into a Cycle 8 snapshot survey of up to 300 targets, aimed specifically at finding close-separation lensed quasars using the imaging capabilities of STIS (Kimble et al. 1997; Woodgate et al. 1998) on board the Hubble Space Telescope. The probability that a quasar is lensed increases with redshift and apparent magnitude (Turner, Ostriker, & Gott 1984); the snapshot survey targets bright, high redshifts quasars selected using estimates for the probability of lensing by Kochanek (1998). The results of the full snapshot survey will appear in time (Gregg et al. 2000, in prep.); here we report the discovery of a close-separation gravitationally lensed quasar from among the first 80 snapshot targets.
## 2 Observations
### 2.1 Discovery
The quasar HE 0512$``$3329 was originally identified in the Hamburg/ESO survey for bright QSO’s (Wisotzki et al. 1996). With $`B`$ = 17.0 and z = 1.569 (Reimers, Köhler, & Wisotzki 1996), HE 0512$``$3329 had an a priori probability of $`1.3\%`$ of being lensed, fairly typical for the targets in our snapshot survey. The STIS snapshot sequence, obtained on 1999 August 26, consists of $`3\times 40`$s CR-split exposures in the clear 50CCD (CL) aperture and one additional 80s CR-split exposure in the longpass F28$`\times `$50LP (LP) filter. The effective wavelength and full width half maximum of the CL band are 6167.6Å and 4410Å and for the LP band are 7333Å and 2721Å. The STIS images reveal two point sources with a separation of 0$`\stackrel{}{\mathrm{.}}`$644. The difference in brightness between the two components A and B is $`\mathrm{\Delta }CL=0.35`$ and $`\mathrm{\Delta }LP=0.49`$; this small difference is characteristic of the more highly magnified lensed systems, which our selection technique is designed to favor.
The lensing hypothesis was strengthened by examining the discovery spectrum of HE 0512$``$3329 which shows a rather typical quasar energy distribution having emission lines of C IV 1549 and C III\] 1909 (see Figure 1 of Reimers et al. 1996). If one of the two components were a garden variety Galactic star, the strongest stellar absorption lines would be easily identifiable (see below, §2.3). A binary quasar is a possible alternative explanation (Kochanek, Falco, & Muñoz 1999), however, the discovery spectrum exhibits a strong absorption feature consistent with Mg II at an intervening redshift of 0.93, and a few weaker absorption lines of Fe II at the same redshift. These low-ionization absorption features suggested that the duplicity is due to a lensing object at this redshift.
### 2.2 Follow-up Spectroscopy at Keck Observatory
In early 2000 January, we obtained a 9Å resolution spectrum of HE 0512$``$3329 using the Low Resolution Imaging Spectrograph (LRIS, Oke et al. 1995) at Keck Observatory. The slit was oriented at the position angle of the two quasar images on the sky, 17°. The seeing was 1″, insufficient to resolve the components. From this 300s exposure (Figure 1), we obtain a redshift of $`1.565\pm 0.001`$ based on Gaussian fits to the C IV and C III\] emission peaks. The S/N of this new spectrum is 50 to 100 over most of its wavelength range and confirms the presence of the strong intervening absorption, clearly resolving the Mg II 2796.4, 2803.5 doublet and detecting the associated Mg I 2853 line. Also seen is a rich absorption system of Fe II 2260.8, 2344.2, 2374.5, 2382.8, 2586.7, and 2600.2, and Ca I 4227.9 belonging to the same intervening system; Ca II 3933 and 3969 fall in the atmospheric A band. The mean of the absorption redshifts is z = $`0.9313\pm 0.0005`$. The profile of the Mg II emission line is asymmetric, which can be attributed to absorption by Fe I 3721.0 or intervening Mg II local to the quasar. There is another weak intervening Mg II absorption feature at z=1.1346.
Component B is 70% the brightness of A. The lack of any discernible stellar absorption features in the Keck spectrum (Figure 1) argues strongly against component B being a foreground star. The RMS noise in the spectrum is at the level of 0.15Å equivalent width. The strongest features in late type stars have equivalent widths of a few Ångstroms and would be detected easily in the Keck spectrum; for comparison, the equivalent width of the intervening Mg I 2853Å feature is 1.4Å. The only possible stellar contaminant is a completely featureless O-type subdwarf or white dwarf and such stars are extremely rare. If this were the case, however, the spectrum of HE 0512$``$3329 would be much bluer, unless either the QSO or the putative star has a large amount of intrinsic reddening.
### 2.3 Photometry
We have done point-spread function (PSF) fitting photometry on the STIS images using IRAF/DAOPHOT. From observations of an unlensed quasar in our program, we obtain aperture corrections of $`0.222`$ and $`0.303`$ for the CL and LP bands, to go from the fitted 3 pixel radius to 0$`\stackrel{}{\mathrm{.}}`$5. To this we add an additional $`0.1`$ magnitudes to correct to the “true” magnitude in an infinite aperture, as is standard practice with WFPC2 (Holtzman et al. 1996). The resulting calibrated STIS “STmagnitudes” and errors are listed in Table 1.
The $`LP`$ bandpass is completely contained within the $`CL`$. Because they are similar in shape, the $`LP`$ flux can be scaled by the relative throughputs and subtracted from the $`CL`$, producing an effective “shortpass” measurement (Gregg & Minniti 1997; Gardner et al. 2000) extending from 5500Å to 2000Å, with effective wavelength of 4424Å and FWHM of 2569Å. The $`SPLP`$ difference provides some wide-band color information (Table 1).
In 1999 December, we obtained VRI photometry of HE 0512$``$3329 using the Mosaic II CCD imager at the Blanco 4m telescope at Cerro Tololo Inter-American Observatory<sup>2</sup><sup>2</sup>2Cerro Tololo Inter-American Observatory, NOAO, is operated by the Association of Universities for Research in Astronomy, Inc. (AURA), under cooperative agreement with the National Science Foundation.. Although the seeing was $`0\stackrel{}{\mathrm{.}}80\stackrel{}{\mathrm{.}}9`$ and the two components are not cleanly resolved, point-spread function (PSF) fitting using IRAF/DAOPHOT successfully separated them, yielding positions in excellent agreement with the HST images and photometry consistent with the STIS results. The separations obtained in V, R, and I are 0$`\stackrel{}{\mathrm{.}}`$654, 0$`\stackrel{}{\mathrm{.}}`$646, and 0$`\stackrel{}{\mathrm{.}}`$643, respectively, compared to 0$`\stackrel{}{\mathrm{.}}`$644 obtained from the centroids of the components in the STIS CL images.
No photometric standards were taken at CTIO, so we have calibrated the CTIO photometry using zeropoints determined by convolving the Keck spectrophotometry with Cousins VRI passbands. This procedure is itself calibrated using a model for the spectrum of Vega (Kurucz 1992) for which we adopt $`B=V=R=I=0.0`$. Slit losses limit the absolute accuracy, but, fortuitously, the spectroscopy was obtained when the position angle of HE 0512$``$3329 was only 23° from the parallactic angle. Because the effective slit width was somewhat greater than the atmospheric dispersion between the red and blue extremes of the spectrum (Filippenko 1982), the colors obtained from the composite spectrum are reasonably accurate and can be used to establish the relative zeropoints of the VRI photometry. Also, we have determined a zeropoint transformation between the effective STIS SP bandpass and Johnson B using the mean quasar spectrum (Brotherton et al. 2000) from the FIRST Bright Quasar Survey (FBQS; White et al. 2000), redshifted to z=1.565. The resulting colors of HE 0512$``$3329 are $`BV=0.68,VR=0.46,VI=0.82`$ for component A, and $`BV=0.32,VR=0.37,VI=0.69`$ for component B (Table 1). Schlegel, Finkbeiner, & Davis (1998) estimate a Galactic extinction of $`A_B=0.104`$ for this line of sight; Burstein & Heiles (1982) give a much lower value of 0.010. The numbers in Table 1 have not been corrected for Galactic extinction.
The broad band colors of the two components clinch the case for HE 0512$``$3329 being a lensed quasar. By comparison with the Bruzual et al. stellar library, the $`BV`$ color of component B indicates a spectral type of F0, yet V-R and V-I are consistent with a much cooler object, about F9/G0. Our simulations show that any star in this spectral range with the relative brightness of component B would contribute easily detectable absorption features, at $`10\sigma `$ level or greater, to the composite spectrum at the indicated locations in Figure 1; Ca II H and K and the Balmer lines would be particularly conspicuous. The broad band colors of component B are, in fact, more consistent with a slightly reddened quasar than a star.
For future reference for monitoring variability of the lens components, we list in Table 2 the instrumental magnitude differences, $`m_im_\mathrm{A}`$, for nine field stars with $`V16`$ to 18 in the vicinity of HE 0512$``$3329. The astrometry has been derived from the digitized sky survey and has an offset of $`\mathrm{\Delta }R.A.=+0\stackrel{}{\mathrm{.}}55`$ and $`\mathrm{\Delta }Dec.=+0\stackrel{}{\mathrm{.}}77`$ relative to the STIS images, but these positions are sufficient to unambiguously identify the comparison stars.
## 3 The Nature of the Lensing Object
The presence of the strong intervening Mg II absorption and the many associated low ionization lines are evidence that the lensing object contains a damped Ly$`\alpha `$ absorption (DLA) system (Boisse et al. 1998). A DLA system at a redshift $`<1`$ is most likely to be the hydrogen-rich disk of a spiral galaxy. Dust in the galaxy may produce differential reddening in the two components of HE 0512$``$3329.
### 3.1 Possible Detection of a Third Object
To explore for the lensing galaxy and possible additional quasar images, the STIS $`CL`$ images were combined using the DRIZZLE package (Fruchter & Hook 1998) in IRAF/STSDAS. A sampling rate of 0.5 times the original image scale and a PIXFRAC value of 0.6 were used. The subpixel image shifts were determined using the IRAF task XREGISTER. The final combined $`CL`$ image is shown in the left panel of Figure 2. The distance between centroids of the two images is 0$`\stackrel{}{\mathrm{.}}`$644; at the probable lens redshift of 0.9313, this separation is only $`4`$ kpc, adopting $`\mathrm{H}_{}`$= 70 $`\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$and $`\mathrm{q}_{}`$= 0.5. The mass associated with an Einstein ring of this scale is $`3\times 10^{10}`$ M.
The PSF removal was done using the SCLEAN task in IRAF/STSDAS. For the PSF itself, we used the theoretical STIS PSF from the Tiny Tim package and also the STIS PSF generated by the Hubble Deep Field South project (Gardner et al. 2000). They yield very similar results. The residual image is shown in the right hand panel of Figure 2. There is an excess of counts just above and to the right of component A (white arrow in Figure 2). The RMS in the background-subtracted image is $`\pm 1.3`$ counts while the peak in the excess region is 8.7. The total flux is $`5.9`$ magnitudes fainter than component A, giving it $`CL=23.8`$. Its FWHM is roughly twice that of a point source, consistent with being nonstellar. We tentatively identify this object as the nucleus of the lensing galaxy. For our adopted cosmology, this object has $`M_V19`$, roughly the nucleus of a roughly $`L^{}`$ galaxy with a bulge-to-disk ratio of $`3`$ and, for the above quoted Einstein ring mass, a M/L ratio of $`20`$. At the intervening redshift of 0.9313, the the two lines of sight to the quasar pass 1.6 (A) and 2.7 (B) kpc from the position of the third object.
There is excess light of lower surface brightness between the two quasar images and immediately below component A as well as to the right of component B in Figure 2. Although the third object is not detected with confidence in the LP image, the low surface brightness light distribution is qualitatively reproduced in the redder passband. No such excess light is seen when the same analysis is applied to an image of an unlensed quasar from our snapshot program. Deeper images are needed to confirm the reality of this low surface brightness fuzz and, if real, determine whether it is due to the lens, the host galaxy, or another object.
### 3.2 Reddening Analysis
The STIS and ground-based photometry are consistent in showing that component A is redder than B. Going from red to blue, the magnitude difference between the two images decreases, becoming equal within the errors in the concocted STIS $`SP`$ and transformed $`B`$ bands (Figure 3). Observed in the ultraviolet, component B will be the brighter. This trend can be attributed to differential reddening, with extinction along the line of sight to component A being greater. Color differences between the quasar images can also be arise from microlensing by stars in the lensing galaxy, producing differential magnification of the quasar continuum (Wambsganss & Paczyński 1991) along the two lines of sight. This effect has been observed in at least one quasar, HE 1104$``$1805 (Wisotzki et al. 1993). The following reddening analysis is valid only if microlensing is negligible in HE 0512$``$3329.
To quantify the amount of differential reddening, we have fitted an extinction model to the photometry results, following the procedure of Falco et al. (1999). In this approach, it is assumed that the quasar is not variable, that the lensing magnification is not wavelength dependent, and that the extinction law does not vary with position in the lensing galaxy and is well-approximated by a typical Galactic extinction curve with $`R_V=A_V/E(BV)=3.1`$. Correcting a sign error in equation 3 of Falco et al., we have
$$\chi ^2=\underset{j=1}{\overset{N_\lambda }{}}\underset{i=1}{\overset{N_c}{}}\frac{\{m_i(\lambda _j)m_{}(\lambda _j)+2.5\mathrm{log}(M_i)E_iR[\lambda _j/(1+z)]\}^2}{\sigma _{ij}^2}$$
(1)
where $`m_i`$ are the observed magnitudes of each of the $`N_c`$ components in the $`N_\lambda `$ photometric bands with effective wavelength $`\lambda _j`$, $`m_{}`$ is the unlensed magnitude of the quasar, $`M_i`$ is magnification of each component, $`E_i`$ is the extinction of each component, z is the redshift of the lens, and $`\sigma _{ij}`$ are the photometric errors. The summations are over the 4 bandpasses, $`B,V,R,`$ and $`I`$, and two components, A and B.
Relative magnifications and extinctions can be found by minimizing $`\chi ^2`$ while holding one magnification fixed at unity and one extinction at 0. As component B is bluer and fainter, we fix its parameters at these values. The extinction law has been parametrized using the equations of Cardelli, Clayton, & Mathis (1989). For this analysis, we first corrected the photometry listed in Table 1 for Galactic extinction of $`A_B=0.104`$ from Schlegel et al. (1998).
The fit for the relative extinction results in the estimate of $`A_V=0.34`$ for component A, in excess of the extinction at component B; the unextincted, wavelength independent relative magnification of A is 2.45 times that of B. The $`\chi ^2`$ of this fit is 0.66. For comparison, a fit with both extinctions held to zero yields a relative magnification of 1.35, roughly consistent with the brightness difference between the two components in $`V`$ or $`R`$. The $`\chi ^2`$ for this fit is 67, as might be expected given the varying magnitude difference between the two components as a function of wavelength, which renders an achromatic magnification model a poor explanation of the brightness variation with wavelength.
The separate extinctions to each component of HE 0512$``$3326 can be estimated by assuming that the unlensed quasar spectrum has typical colors. After correcting for Galactic reddening using the Schlegel et al. (1998) value, the difference in $`BV`$ between the composite spectrum of HE 0512$``$3326 (Figure 1) and the FBQS mean spectrum is 0.31. With $`R_V=3.1`$, this is equivalent to $`A_V=0.97`$. Knowing the $`V`$ magnitudes and relative extinction, the separate extinctions can be computed as $`A_V^\mathrm{A}=1.10`$ and $`A_V^\mathrm{B}=0.76`$, excluding any grey component. Given the multitude of assumptions and the bootstrapping from the spectrophotometry, these numbers must be considered provisional, but they do suggest that the extinction to each component is comparable and that both lines of sight intercept the same or similar absorption systems. Spectroscopy of the two components separately would allow a detailed study of the extinction curve in the disk of the lensing galaxy and would also determine whether microlensing could be contributing to the pattern of color differences.
## 4 Conclusion
The presently available data leave little doubt that HE 0512$``$3329 is gravitationally lensed. The spectroscopic evidence strongly suggests that the lens is a spiral galaxy. Spatially resolved spectroscopy of the two images of HE 0512$``$3329 is needed to confirm its nature; however, the presence of strong low-ionization lines in the composite spectrum indicates that at least one of the lines of sight is sure to pass through a damped Ly$`\alpha `$ system in the disk of the lens. Ultraviolet spectroscopy of the A and B components can further be used to derive the extinction curve in the disk of the lens as well as abundances of heavy elements. This lensed quasar has been found among the first 80 targets of an HST Cycle 8 snapshot program designed to search for such small separation systems. The program was renewed for up to 300 additional snapshots in Cycle 9. If close-separation targets are found with this frequency for the duration of the survey, the lensing statistics for small separation systems will be boosted by a significant factor.
Mark Lacy is thanked for helpful discussions. The referee is credited with constructive comments which improved this paper. We are grateful to Sune Toft for calling our attention to an error in our original calculation of the differential reddening. Support for this work was provided by NASA through grant number GO-8202 from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS5-26555. We also acknowledge support from NSF grant AST-98-02791. This work was performed under the auspices of the U.S. Department of Energy by University of California Lawrence Livermore National Laboratory under contract No. W-7405-Eng-48. J.N.W. thanks the Fannie and John Hertz Foundation for financial support.
|
warning/0002/gr-qc0002063.html
|
ar5iv
|
text
|
# The Dynamics of Curved Gravitating Walls
## I Introduction
Over the past few decades, topological defects have become a familiar class of objects in many areas of physics. In cosmology, defects are believed to arise generically during phase transitions in the early Universe, and have been notably invoked to account for the anisotropies which seeded cosmological structures . Despite the discouraging discovery that the power spectrum of global strings does not agree with observations of the COBE satellite , recent claims of a nongaussian component in the microwave background (see also ), and the improved agreement of the spectrum predictions for models of cosmic strings with a cosmological constant (whose best fit coincides with the value of $`\mathrm{\Lambda }`$ determined from type Ia supernovae ) suggest that it may be still too early to discard them as the source of cosmic structure.
Domain walls are defects that arise when the phase transition occurs by the breakdown of a discrete symmetry. They correspond to solitons in $`1+1`$ dimensions which are extended in two spatial dimensions to form a wall structure. Because static wall solutions depend only on one coordinate (the distance from the wall’s core), they can often be found analytically in the absence of gravity, and perturbatively analytically in its presence. In a cosmological context, it was soon realised that the existence of domain walls with $`\eta \mathrm{\hspace{0.17em}1}`$ MeV must be ruled out, because a network of such defects would rapidly evolve to dominate the energy of the Universe. Nevertheless, domain walls remain intrinsically interesting objects to study, for instance for their properties as hypersurfaces; in cosmology, domain walls have been proposed as a realization of our universe in higher dimensions , and are currently being explored as a possible resolution of the hierarchy problem .
When considering ‘defects’, there are two main aspects to understand: their gravitational (or other particle) interactions and their dynamics. The gravity of domain walls is an interesting and rather more subtle topic than it might seem at first sight. Indeed, unlike all other defects (with the exception of global strings ), the wall’s metric is not in general static , but admits a de Sitter-like expansion in its plane. Moreover, observers experience a repulsion from the wall, and there is a cosmological horizon at a finite proper distance from the defect’s core. This horizon is a consequence of the choice of coordinates, and in a different set of coordinates the wall has the appearance of a bubble which contracts in from infinite radius to some minimum radius, then re-expands, undergoing uniform acceleration from the origin. The ‘horizon’ is then simply the lightcone of the origin in these coordinates, and is somewhat similar to the horizon of Rindler spacetime. These results were originally obtained for infinitesimally thin walls, using the hypersurface formalism developed by Israel , but can be shown to be robust as an approximate description of a thick domain wall by a perturbative expansion in the thickness of the wall , or within the context of a fully nonlinear treatment of a scalar field coupled to gravity .
The crucial physical difference, then, of the self-gravitating domain wall spacetime, is the presence of the cosmological horizon, which introduces a second length scale into the system. Ordinarily, a defect possesses one length scale, its thickness $`w`$, however, the distance to the event horizon of the domain wall gives another length scale, $`u_\mathrm{h}`$, which can be compared to $`w`$. These lengths are given in terms of the coupling constants of the theory and, as taking a thin wall limit turns out to be a very artificial construction in terms of these underlying parameters, it becomes pertinent to examine both the gravity and dynamics for a thick domain wall.
The dynamics of topological defects are typically extremely nonlinear, and their study is usually carried out in the so-called Nambu approximation, where the full-field theory action of the model in which the defects arise is replaced by an approximation based on the degrees of freedom of the defect’s core. At first sight, this looks like a perfectly reasonable approximation, since—even for finite-sized defects, such as string loops and wall bubbles—the thickness of the defect is typically many orders of magnitude smaller than its size. It must be noted however, that the Nambu action is only a leading order approximation of the real action, and is obtained in the limit $`\alpha 0`$, where $`\alpha Kw`$ is proportional to the defect’s typical curvature and thickness . However, it is often the case that the small-scale structure of defects (when $`\alpha `$ can be significant) is of particular importance when considering their impact. It seems that quite generically these are the points at which the defects lose most of their energy (by Higgs, gravitational or possibly other types of radiation, depending on the type of defect) and the properties of the small-scale therefore have a direct impact on the defects’ lifespan, which in turn dictates the possible cosmological implications they can have.
The first attempts to derive effective actions for walls naturally neglected gravity, in the sense that neither curvature of the background spacetime nor the self-gravity of the wall were considered. The effect of background spacetime curvature was considered (using the effective action method discussed below) in , and the self-gravity of the wall for a very special trajectory was considered in ; however, the motion of a fully self-gravitating thick domain wall (where the curvature of spacetime is that induced by the wall) has not to date been considered. In this paper, we address this problem, namely, using the $`\lambda \mathrm{\Phi }^4`$ field theory, we examine the dynamics of a thick, self-gravitating kink domain wall solution.
There are essentially two methods which have been used to obtain the effective motion of a thick defect, both of which involve in some way an expansion of the fields around a well-known solution (such as the hyperbolic tangent kink for a $`\lambda \mathrm{\Phi }^4`$ wall). The first approach consists of replacing the solution in the action and integrating out perpendicular to the defect, which yields an effective action based on the defect’s core. This method has been employed for instance in . The second approach consists of examining the field equations perturbatively in a relevant parameter (or parameters) such as $`\alpha `$, with the equations of motion to a particular order arising as an integrability condition (see for instance ). Here we use this latter method, adapted to take the wall’s self-gravity into account. The reason for this choice is in fact related to the inclusion of gravity in the problem; the motion of the core of the defect interacts with massless degrees of freedom in the bulk (the graviton) and hence a correct application of the effective action method is less transparent.
The layout of the paper is as follows. In the next section we present the model, then briefly introduce the Gauss–Codazzi formalism and derive the Einstein equations in the corresponding “$`3+1`$” notation. We end this section by reviewing our method in the case of a flat background spacetime. In section III we solve the field equations for the case $`\alpha >ϵ`$ (where $`ϵ`$ characterizes the gravitational interaction of the scalar field). In section IV we discuss the particular case of a collapsing spherical wall, and we conclude in the last section.
## II The Gauss–Codazzi Formalism
Our starting point is the usual Goldstone matter Lagrangian,
$$=\left(_a\mathrm{\Phi }\right)^2U(\mathrm{\Phi }),$$
(1)
where $`\mathrm{\Phi }`$ is a real Higgs scalar field and $`U(\mathrm{\Phi })`$ is a symmetry-breaking potential which we take to be
$$U(\mathrm{\Phi })=\lambda \left(\mathrm{\Phi }^2\eta ^2\right)^2=\lambda \eta ^4V\left(\frac{\mathrm{\Phi }}{\eta }\right).$$
(2)
This model admits domain wall solutions, where the Higgs field tends to different vacua $`\mathrm{\Phi }=\pm \eta `$ at, say, $`u=\pm \mathrm{}`$ for a flat wall. This implies the existence of a surface for which $`\mathrm{\Phi }=0`$, and this surface defines the defect’s core.
As in , we scale out the dimensionful parameters from the Lagrangian by defining
$$X=\mathrm{\Phi }/\eta ϵ=8\pi G\eta ^2;$$
(3)
$`X`$ now tends to $`\pm 1`$ at the vacua. The scalar and Einstein equations are
$`\mathrm{}X+{\displaystyle \frac{2}{w^2}}X\left(X^21\right)`$ $`=`$ $`0`$ (5)
$`_{ab}`$ $`=`$ $`ϵ\left[2X_{,a}X_{,b}{\displaystyle \frac{1}{w^2}}g_{ab}\left(X^21\right)^2\right],`$ (6)
where $`_{ab}`$ is the spacetime Ricci tensor, and $`w`$ is proportional to the inverse mass of the Higgs after the symmetry breaking, and therefore characterizes the width of topological defects within the theory. Note that according to (6) $`ϵ`$ characterizes the gravitational interaction of the Higgs field.
As mentioned in the introduction, our method for finding the effective dynamics of domain walls requires the expansion of the quantities appearing in the full equations of motion in powers of some small parameter. This can be achieved by splitting these quantities in their components parallel and perpendicular to the wall’s worldvolume $`\mathrm{\Sigma }`$, with the help of the Gauss–Codazzi formalism . Before we start, a remark on our notation: although we shall generally use lowercase Latin indices $`a,b,\mathrm{}`$, we may emphasize the parallel character of some index by using uppercase Latin letters $`A,B,\mathrm{}`$ For instance, the coordinates parallel to the defect will be called $`\sigma ^A`$. The proper distance from the wall will be denoted by $`u`$.
A domain wall’s core (defined by the location of $`X0`$ in the above model) is a three-dimensional surface in four-dimensional spacetime, and consequently it admits a (spacelike) unit normal field denoted by $`n^a`$. This normal field can be regularly extended off the worldvolume by imposing $`n^a_an_b=0`$, so that each surface of constant $`u`$ has a normal field $`n^a`$, a first fundamental tensor $`h_{ab}`$ and a second fundamental tensor $`\widehat{K}_{ab}`$, the latter two being defined by
$$h_{ab}=g_{ab}+n_an_b\widehat{K}_{ab}=h^c{}_{a}{}^{}_{c}^{}n_b.$$
(7)
$`h_{ab}`$ is the projection tensor onto the worldvolume (with our choice of coordinates, its parallel components are equal to the intrinsic metric’s, and its perpendicular components vanish). $`K_{ab}`$ is the extrinsic curvature of $`\mathrm{\Sigma }`$, which lies tangentially to the worldsheet and describes how the wall curves away from a hyperplane in spacetime. The metric of the wall spacetime is therefore written in this coordinate system as
$$ds^2=h_{AB}d\sigma ^Ad\sigma ^Bdu^2.$$
(8)
The Gauss and Codazzi integrability conditions for the hypersurface generated by the wall’s core are,
$$\widehat{R}_{ab}=_{cd}h_a^ch_b^d+\widehat{K}_{ac}\widehat{K}_b^c\widehat{K}\widehat{K}_{ab}+_{acbd}n^cn^d$$
(9)
$$\widehat{D}_c\widehat{K}_a^c\widehat{D}_a\widehat{K}=\widehat{R}_{de}n^eh_a^d,$$
(10)
where $`\widehat{D}_a=h_a^c_c`$. Using these integrability conditions, the equations of motion (3) can be rewritten
$`{\displaystyle \frac{h_{ab}}{u}}`$ $`=`$ $`2\widehat{K}_{ab}`$ (12)
$`{\displaystyle \frac{\widehat{K}_{ab}}{u}}`$ $`=`$ $`ϵ(X^21)^2h_{ab}+\left[2\widehat{K}_{ac}\widehat{K}_b^c\widehat{K}\widehat{K}_{ab}\widehat{R}_{ab}\right]+2ϵ\widehat{D}_aX\widehat{D}_bX`$ (13)
$`{\displaystyle \frac{^2X}{u^2}}2X(X^21)`$ $`=`$ $`\widehat{K}{\displaystyle \frac{X}{u}}+\widehat{D}_a\widehat{D}^aX`$ (14)
$`{\displaystyle \frac{\widehat{K}}{u}}`$ $`=`$ $`\widehat{K}_{ab}^2ϵ\left[2X^2+(X^21)^2\right]`$ (15)
$`\widehat{D}_c\widehat{K}_a^c\widehat{D}_a\widehat{K}`$ $`=`$ $`2ϵ{\displaystyle \frac{X}{u}}\widehat{D}_aX`$ (16)
$`\widehat{R}`$ $`=`$ $`\widehat{K}_{ab}^2\widehat{K}^2+2ϵ\left[X^2(X^21)^2+\widehat{D}_aX\widehat{D}^aX\right].`$ (17)
Here we have (without loss of generality) set $`w=1`$, which amounts to using wall rather than Planck units. The mean curvature $`\widehat{K}`$ is the trace of $`\widehat{K}_{ab}`$ and $`\widehat{R}_{ab}`$ is the worldvolume Ricci tensor. Note that the equations for $`\widehat{K}_{ab}`$, $`\widehat{K}`$ and the derivatives of $`\widehat{K}`$ in the wall correspond respectively to the “$`AB`$,” “$`uu`$” and non-diagonal Einstein equations. The final equation results from the trace of (9) and turns out to be related to the integrability condition which gives the wall equation of motion.
We now need to identify the two key parameters in these equations, as well as the dependence of the variables on these parameters. Clearly the gravitational parameter appears explicitly in the equations, however, the parameter $`\alpha `$ characterising the motion of the wall is only implicit in the equations. The first step to identifying this parameter is to quantify what one means by ‘motion’ of the wall; this is encoded in the components of the extrinsic curvature on the wall core itself, since this tells us that the wall is curved in the ambient spacetime. We therefore set
$$\alpha |\widehat{K}_b^a(u=0)|.$$
(18)
Note that this is not a fundamental parameter of the theory, in that it is not given in terms of any coupling constants or masses, but simply represents the physical motion of the wall and enables the effect of that motion to be correctly considered. The extrinsic curvature $`\widehat{K}_{ab}`$, therefore, has two main contributions: that of the motion of the wall core and that of the gravity of the wall. This can be estimated by considering the case of the plane-symmetric wall , for which $`\alpha =0`$, and
$$\widehat{K}_{ab}=A^{}A\mathrm{diag}(1,\mathrm{e}^{2ct},\mathrm{e}^{2ct}),$$
(19)
where $`A(u)=1+ϵA_1(u)+O(ϵ^2)`$, $`c=2ϵ/3+O(ϵ^2)`$. Clearly then, the components of $`\widehat{K}_b^a`$ are O$`(ϵ)`$.
To summarise: in order to describe the motion of a domain wall in a curved spacetime we naturally have two parameters; one characterising the motion of the wall itself, $`\alpha `$, and one the curvature of the ambient spacetime, $`ϵ`$. The basic procedure for determining the equation of motion of the wall is to solve (10) order by order in these parameters, investigating any constraints arising on the extrinsic curvature at each step. It is perhaps worthwhile briefly reviewing this process for $`ϵ=0`$, since the methodology is very similar when gravity is included.
We begin by rescaling the extrinsic curvature and parallel coordinates via
$`\widehat{K}_{ab}`$ $`=`$ $`\alpha K_{ab}`$ (21)
$`\sigma ^A`$ $`=`$ $`x^A/\alpha .`$ (22)
Since we are in flat space the Gauss identity (9) simplifies to,
$$R_{ab}=K_{ac}K_b^cKK_{ab};$$
(23)
hence
$$K_{ab}=K_{ab}|_0+\alpha uK_a^c|_0K_{bc}|_0$$
(24)
(where “$`|_0`$” indicates that a quantity is evaluated at the wall core) is actually an implicit exact solution to the $`K`$-equation. To order $`\alpha `$, we see that
$$K=K|_0\alpha uK_{ab}^2|_0,$$
(25)
and we can examine the $`X`$-equation (14) by setting $`X=X+\alpha X__1`$, where $`X_0=\mathrm{tanh}u`$, finding
$$𝒟X__1X__1^{\prime \prime }2X__1(3X__0^21)=K|_0X__0^{}.$$
(26)
Ordinarily, we might expect to be able to write the solution $`X__1`$ in terms of the basis of eigenfunctions of the operator $`𝒟`$, however, we cannot do this directly, since $`X__0^{}`$ is in fact the zero mode of $`𝒟`$. We are therefore forced to either deduce that $`K|_0=0`$, or, we can take the approach of reference and remove the requirement that the equations of motion be regular at the wall. Since we are looking for freely moving wall trajectories, we will take the former approach, which can be summed up as an ‘integrability requirement’: multiplying both sides of (26) with $`X_0^{}`$, and integrating over IR implies
$$\left(X_1^{}X_0^{}X_0^{\prime \prime }X_1\right)_{\mathrm{}}^{\mathrm{}}=_{\mathrm{}}^{\mathrm{}}K_0|_0(X_0^{})^2du.$$
(27)
In order for $`X_1`$ to have the appropriate asymptotic behavior for large $`u`$ it follows that
$$K_0|_0=0(X__1=0).$$
(28)
Equation (28) is of course the Nambu equation $`K|_0=0`$, which simply means that the core of the defect, situated at $`u=0`$, follows the Nambu dynamics to zeroth order.
We can then repeat this process, expanding order by order to get
$`h_{ab}`$ $`=`$ $`h_{ab}|_0+2\alpha uK_{ab}|_0+\alpha ^2u^2K_{ac}|_0K_b^c|_0`$ (30)
$`K`$ $`=`$ $`K|_0\alpha uK_{ab}^2|_0+\alpha ^2u^2K_b^a|_0K_c^b|_0K_a^c|_0`$ (31)
$`X`$ $`=`$ $`X_0+\alpha X_1+\alpha ^2X_2,`$ (32)
where $`K|_0=O(\alpha ^2)`$, and
$$X_2=\mathrm{sech}^2u_0^u\mathrm{cosh}^4u_{\mathrm{}}^u(uK_{ab}^2|_0)\mathrm{sech}^4u.$$
(33)
This is sufficient to obtain the leading corrections to the Nambu action via the integrability constraints to third order. To third order (14) gives
$$[X_3^{}X_0^{}X_3X_0^{\prime \prime }]^{}=K_2X_0^2=(K_2|_0+u^2K_b^a|_0K_c^b|_0K_a^c|_0)X_0^2,$$
(34)
hence
$$K_2|_0=\frac{f_2(\mathrm{})}{f_0(\mathrm{})}K_b^a|_0K_c^b|_0K_a^c|_0=\left(\frac{\pi ^2}{6}1\right)K_b^a|_0K_c^b|_0K_a^c|_0,$$
(35)
where $`f_n(u)=_0^u𝑑uu^n\mathrm{sech}^4u`$.
Note that this process of using the integrability condition to derive a constraint on the extrinsic curvature uses the $`X`$-equation to order O($`\alpha ^{n+1}`$) for a constraint on $`K`$ to order O($`\alpha ^n`$), and that this constraint only involves the even part of $`K`$, since any odd parts integrate to zero. Moreover, the $`K`$-equation (15) shows that the even part of $`K`$ to order O($`\alpha ^n`$) depends on the odd part of $`K_{ab}^2`$ to order O($`\alpha ^{n1}`$). Keeping this observation in mind prevents the unnecessary calculation of corrections to the geometric parameters.
## III The Motion of a Wall with Gravity
In order to include gravity, we will make the initial assumption that gravity is subdominant to the motion of the wall, i.e. $`ϵ<\alpha `$. Of course this need not always be the case, however, the derivation of the wall equations for $`ϵ>\alpha `$ is almost identical to $`ϵ<\alpha `$, and an expansion for general $`ϵ`$ and $`\alpha `$ is so notationally cumbersome that we choose to present the analysis in this particular case for brevity and clarity.
First of all, note that the Ricci curvature of the wall is at least of order O($`\alpha ^2`$), as can be seen from (9), since the flat space Ricci curvature is given in terms of products of the extrinsic curvature, and the self-gravitating wall has $`R_{ab}=O(ϵ^2)`$. We may therefore set
$$\widehat{R}_{ab}=\alpha ^2R_{ab}.$$
(36)
Rescaling the extrinsic curvature and parallel coordinates as in (II), and defining
$$\delta =\frac{ϵ}{\alpha },$$
(37)
the equations of motion in the presence of gravity (here characterised by $`\delta `$) become
$`h_{ab}^{}`$ $`=`$ $`2\alpha K_{ab}`$ (39)
$`K_{ab}^{}`$ $`=`$ $`\delta (X^21)^2h_{ab}+\alpha \left(2K_{ac}K_{bd}h^{cd}R_{ab}KK_{ab}\right)+2\delta \alpha ^2D_aXD_bX`$ (40)
$`X^{\prime \prime }`$ $`=`$ $`2X\left(X^21\right)\alpha KX^{}+\alpha ^2D_aD^aX`$ (41)
$`K^{}`$ $`=`$ $`\delta \left[2X^{}{}_{}{}^{2}+(X^21)^2\right]\alpha K_{ab}^2`$ (42)
$`D_cK^c{}_{a}{}^{}D_aK`$ $`=`$ $`2\delta X^{}D_aX`$ (43)
$`\alpha R`$ $`=`$ $`\alpha K_{ab}^2\alpha K^2+2\delta \left[X^2(X^21)^2+\alpha ^2D_aXD^aX\right].`$ (44)
To solve these equations, we expand all quantities with respect to $`\alpha `$:
$$\begin{array}{ccccccccc}X\hfill & =& X_0\hfill & +& \alpha X_1\hfill & +& \alpha ^2X_2\hfill & +& \mathrm{}\hfill \\ h_{ab}\hfill & =& h_{0ab}\hfill & +& \alpha h_{1ab}\hfill & +& \alpha ^2h_{2ab}\hfill & +& \mathrm{}\hfill \\ K_{ab}\hfill & =& K_{0ab}\hfill & +& \alpha K_{1ab}\hfill & +& \alpha ^2K_{2ab}\hfill & +& \mathrm{}\hfill \\ R_{ab}\hfill & =& R_{0ab}\hfill & +& \alpha R_{1ab}\hfill & +& \alpha ^2R_{2ab}\hfill & +& \mathrm{},\hfill \end{array}$$
(45)
which also implies similar series for the traces $`K`$ and $`R`$ with, for instance,
$$K_2=K_{2ab}h_0^{ab}K_{1ab}h_1^{ab}K_{0ab}h_2^{ab}+K_{0ab}h_{1c}^ah_1^{bc},$$
(46)
where all indices are raised using $`h_{0ab}`$. Note that we do not expand in a double series with $`\delta `$, since the presence of the $`\alpha `$ terms in the RHS of (39,c,d) means that at any particular order in $`\alpha `$ the series expansion in $`\delta `$ terminates, as we can see from (III) below. We can now solve equations (37) order by order.
To zeroth order in $`\alpha `$ we obtain
$`h_{0ab}`$ $`=`$ $`h_{0ab}|_0`$ (48)
$`K_{0ab}`$ $`=`$ $`K_{0ab}|_0\delta f_0(u)h_{0ab}|_0`$ (49)
$`K_0`$ $`=`$ $`K_0|_03\delta f_0(u)`$ (50)
$`X_0`$ $`=`$ $`\mathrm{tanh}u,`$ (51)
and (44) is identically satisfied for $`X_0=\mathrm{tanh}u`$. We define
$`f_n(u)`$ $`=`$ $`{\displaystyle _0^u}𝑑uu^nV(X_0)`$ (53)
$`F_n(u)`$ $`=`$ $`{\displaystyle _0^u}𝑑uf_n(u).`$ (54)
Since $`V(X)`$ is an even function $`f_n`$ is odd (respectively, even) for $`n`$ even (respectively, odd). As a result $`F_n`$ is even (respectively, odd) for $`n`$ even (respectively, odd). By considering the derivative of (44) we can deduce that
$$R_0=R_0|_0R_{{}_{0}{}^{}ab}=R_{{}_{0}{}^{}ab}|_0;$$
(55)
however, to find the actual value of $`R_0|_0`$ we need to go to first order for $`X`$ (see equation (66) below).
To first order, we immediately obtain
$$h_{{}_{1}{}^{}ab}=h_{{}_{1}{}^{}ab}|_0+2uK_{{}_{0}{}^{}ab}|_02\delta F_0h_{{}_{0}{}^{}ab}|_0$$
(56)
and
$$X__1^{\prime \prime }2X__1(3X__0^21)=X__0^{}[K__0|_03\delta f_0].$$
(57)
The integrability requirement then constrains $`K__0|_0=0`$, and $`X_1`$ is found to be
$$X_1=3\delta X_0^{}_0^u\frac{1}{X_0^{}^2}_{\mathrm{}}^uf_0X_0^{}=\frac{\delta }{6}\frac{3u+\mathrm{tanh}u}{\mathrm{cosh}^2u}.$$
(58)
Note that $`X_1`$ is a correction due to the presence of gravity. Then
$`K_1`$ $`=`$ $`K_1|_0uK_{0ab}^2|_04\delta X_0^{}X_1\delta ^2G_0`$ (60)
$`K_{1ab}`$ $`=`$ $`K_{1ab}|_0\delta f_0h_{1ab}|_0+2uK_{0ac}|_0K_0{}_{}{}^{c}{}_{b}{}^{}|_{0}^{}\delta (2f_1+F_0)K_{0ab}|_0`$ (62)
$`+\delta G_1(u)h_{0ab}uR_{0ab}|_0,`$
where the functions $`G_0`$ and $`G_1`$ are defined by
$`G_0(u)`$ $`=`$ $`{\displaystyle 𝑑u\left(4\mathrm{tanh}u\mathrm{sech}^2uX_1+3f_0^2\right)}`$ (64)
$`G_1(u)`$ $`=`$ $`2f_0F_0+{\displaystyle 𝑑u\left(4\mathrm{tanh}u\mathrm{sech}^2uX_13f_0^2\right)},`$ (65)
and are both of odd parity. Finally, from (44) one obtains
$$R_0=K_{0ab}^2|_0+2\delta ^2\left[2X_0^{}X_1^{}4X_0X_1(X_0^21)6f_0^2\right]=K_{0ab}^2|_0\frac{8}{3}\delta ^2.$$
(66)
At second order in $`\alpha `$, the equation for $`X_2`$ is
$`𝒟X_2`$ $`=`$ $`6X_0X_1^2K_1X_0^{}K_0X_1^{}`$ (67)
$`=`$ $`K_1|_0X_0^{}+uK_{0ab}^2|_0X_0^{}+6X_0X_1^2+4\delta X_1X_0^{}{}_{}{}^{2}+\delta ^2G_0X_0^{}+3\delta X_1^{}f_0.`$ (68)
Since all but the first term on the RHS have odd parity, the integrability requirement once again constrains $`K_1|_0=0`$, and we can solve for $`X_2`$ giving
$$X_2=\mathrm{sech}^2u_0^u\mathrm{cosh}^4u_{\mathrm{}}^u(uK_{ab}^2|_0)\mathrm{sech}^4u𝑑u,$$
(69)
which is an odd function with respect to $`u`$. Therefore, there are no corrections to the Nambu equation at first order just as in the case of flat spacetime. We do anticipate however that such corrections will appear at second order, and so proceed to calculate $`K_2`$. However, as we commented earlier, the constraints due to the integrability requirement only pertain to the even part of $`K_2`$, and this in turn depends on the odd part of $`K_2^{}`$:
$`K_2^{}`$ $`=`$ $`2K_0^{ab}K_{1ab}+2K_{0ab}K_{0c}{}_{}{}^{b}h_{1}^{ac}`$ (71)
$`2\delta [2X_0^{}X_2^{}+X_1^{}{}_{}{}^{2}+2X_0X_2(X_0^21)+X_1^2(3X_0^21)]`$
$`=`$ $`2uK_{0ab}|_0R_0^{ab}|_0+[\text{even terms}].`$ (72)
Therefore,
$$K_2|_{\mathrm{even}}=K_2|_0+u^2K_{0ab}|_0R_0^{ab}|_0.$$
(73)
Now, examining the $`X`$-equation at order O$`\left(\alpha ^3\right)`$ we find
$$𝒟X_3=2X_1^3+12X_0X_1X_2K_2X_0^{}K_1X_1^{}K_0X_2^{}$$
(74)
and therefore the integrability requirement yields the constraint
$$K_2|_0=\frac{f_2(\mathrm{})}{f_0(\mathrm{})}K_{0ab}|_0R_0^{ab}|_0=\left(\frac{\pi ^2}{6}1\right)K_{0ab}|_0R_0^{ab}|_0.$$
(75)
which gives us the first perturbation of the Nambu equation.
To summarize, the first corrections to the Nambu equations of motion (reversing the rescalings performed) appear at second order in $`\alpha `$ and are:
$$\widehat{K}|_0=\left(\frac{\pi ^2}{6}1\right)\widehat{K}_{ab}|_0\widehat{R}^{ab}|_0.$$
(76)
Although we obtained this result assuming $`ϵ<\alpha `$, it is in fact quite general, since in a similar calculation for $`ϵ>\alpha `$ rescaling with respect to $`ϵ`$ gives exactly the same result. Clearly from (9) this correction has the correct flat space limit (35), and it would seem that the inclusion of gravity simply modifies the second order correction to the equations of motion, rather than causing a completely new correction to appear. Indeed the three dimensional Ricci curvature can be seen in (9) to relate to the extrinsic and background geometry of our spacetime.
A totally geodesic ($`K_{ab}|_0=0`$) solution such as the plane-symmetric wall always trivially satisfies (76). What is of real interest, however, is the existence of non-totally geodesic solutions verifying the perturbative equations of motion (76). A better understanding of the Nambu correction term will involve the computation of the three dimensional curvature tensor $`R_{0ab}|_0`$. This task is undertaken in the next section for the specific example of a collapsing spherical domain wall.
## IV The Collapse of a Spherical Domain Wall
In this section we apply the general equations of motion (76) to the case of a collapsing spherical domain wall. This is perhaps the simplest non-trivial example of a curved domain wall in curved space-time, i.e. where both of our perturbation parameters $`\alpha `$ and $`ϵ`$ are not zero. The spherical domain wall has been already studied in different contexts using Israel’s thin wall formalism . A first attempt to study the thick case was undertaken in but only using equations (35) valid for a flat space-time background.
Consider, in a spherical system of coordinates $`(t,r,\theta ,\varphi )`$, a non-static scalar field representing a domain wall matter coupled to a spherically symmetric metric
$$ds^2=A^2(t,r)dt^2B^2(t,r)dr^2r^2d\mathrm{\Omega }_{II}^2.$$
(77)
The field equations one has to solve are the coupled Einstein and scalar equations, which can be written in a convenient way as
$`{\displaystyle \frac{(AB)^{}}{AB^3}}`$ $`=`$ $`ϵr\left({\displaystyle \frac{\dot{X}^2}{A^2}}+{\displaystyle \frac{X^2}{B^2}}\right)`$ (79)
$`\left[\left(1{\displaystyle \frac{1}{B^2}}\right)r\right]^{}`$ $`=`$ $`ϵr^2\left({\displaystyle \frac{\dot{X}^2}{A^2}}+{\displaystyle \frac{X^2}{B^2}}+V(X)\right)`$ (80)
$`{\displaystyle \frac{\dot{B}}{B}}`$ $`=`$ $`ϵrX^{}\dot{X}`$ (81)
$`{\displaystyle \frac{A^{\prime \prime }}{AB^2}}{\displaystyle \frac{A^{}B^{}}{AB^3}}{\displaystyle \frac{\ddot{B}}{BA^2}}+{\displaystyle \frac{\dot{B}\dot{A}}{BA^3}}+{\displaystyle \frac{2A^{}}{rAB^2}}`$ $`=`$ $`2ϵ{\displaystyle \frac{\dot{X}^2}{A^2}}ϵV(X)`$ (82)
$`\mathrm{}X+2X(X^21)`$ $`=`$ $`0.`$ (83)
For small values of $`ϵ1`$, consider the field expansion,
$`X`$ $`=`$ $`X_0+ϵX_1+O(ϵ^2)`$ (85)
$`A`$ $`=`$ $`1+ϵA_1+O(ϵ^2)`$ (86)
$`B`$ $`=`$ $`1+ϵB_1+O(ϵ^2).`$ (87)
In order to solve (IV) perturbatively one has to first solve the scalar equation (83) to zeroth order in $`ϵ`$. Then integrating out the Einstein equations (79-81) we can obtain the first order $`ϵ`$ corrections for the metric $`A_1`$ and $`B_1`$. Let us first define the wall’s position and discuss some general features about the solution before solving (IV).
The location of the wall is defined by the zero of the Higgs field, and will have coordinates $`X^a=(t,R(t),\theta ,\varphi )`$. We start by making the observation that in order for the wall to be identifiably a topological kink solution, $`R(t)1`$, and without loss of generality, we can assume that $`X<0`$ (respectively, $`X>0`$) for $`r<R(t)`$ ($`r>R(t)`$). (Note that these are not the Gauss–Codazzi gauge coordinates centered on the wall’s core.)
We consider the following initial conditions which are compatible with the fact that, due to its spherical symmetry, the wall is not radiating:
$$R(t=0)=R_\mathrm{i},\dot{R}(t=0)=0.$$
(88)
The wall’s initial bending parameter is defined as $`\alpha =1/R_\mathrm{i}`$.
For an inertial observer situated outside the wall, $`r>R(t)`$, so by Birkhoff’s theorem the exterior metric is Schwarzschild:
$$ds^2=\left(1\frac{2GM}{r}\right)dt^2\left(1\frac{2GM}{r}\right)^1dr^2+r^2d\mathrm{\Omega }_{II}^2,r>R(t).$$
(89)
In the same way, inside the wall the spacetime metric is flat to a very good approximation. The wall’s mass, as measured by a distant observer, is $`M=4\pi \sigma R_\mathrm{i}^2`$ and the Schwarzchild radius is given by $`r_\mathrm{S}=2GM`$. In order for $`R_\mathrm{i}>r_\mathrm{S}`$, one must impose the relation $`\alpha >ϵ`$.
Keeping these considerations in mind, let us now proceed with solving our field equations order by order in $`ϵ`$ in the region close to the wall’s core, $`rR(t)`$. In order to do so, we must also expand the function $`R(t)`$, determining the wall’s position, in powers of $`ϵ`$,
$$R(t)=R_0(t)+ϵR_1(t)+O(ϵ^2).$$
(90)
Of course $`R_0(t)`$ will in fact be given in terms of a power series in $`\alpha `$: $`R_0(t)=_0(t)+\alpha ^2_2(t)`$ etc.
To zeroth order in $`ϵ`$, i.e. in a flat background spacetime, we can define the unit, exterior pointing, normal vector to the wall’s core at $`r=R_0(t)`$ as
$$n^a=(\dot{R_0},1,0,0)/\sqrt{1\dot{R_0}^2}.$$
(91)
The intrinsic metric and extrinsic curvature components of the wall (at $`r=R_0(t`$)) are easily calculated in turn using (7). The extrinsic curvature for $`ϵ=0`$ is then given by
$$K_b^a=\frac{\ddot{R_0}\delta _t^a\delta _b^t}{(1\dot{R_0}^2)^{3/2}}+\frac{\delta _\theta ^a\delta _b^\theta +\delta _\varphi ^a\delta _b^\varphi }{R_0\sqrt{1\dot{R_0}^2}}$$
(92)
and the equation of motion (using (35)) is
$$\ddot{R_0}=\frac{2}{R_0}(1\dot{R_0}^2)\left(\frac{\pi ^2}{6}1\right)\left[\frac{\ddot{R}_{0}^{}{}_{}{}^{3}}{(1\dot{R}_0^2)^3}+\frac{2}{R_0^3}\right]+O(\alpha ^3).$$
(93)
This can be solved iteratively, giving to leading order
$$\left(\frac{_0}{R_\mathrm{i}}\right)^4=1\dot{_0}^2.$$
(94)
which can be solved analytically and numerically, as can the O($`\alpha ^2`$) correction (see and ), giving the wall trajectory as shown in figure 1. To O($`\alpha ^2`$) we note that $`R_0`$ satisfies
$$1\dot{R}_0^2=\frac{R_0^4}{R_\mathrm{i}^4}\left[1\frac{2C}{R_\mathrm{i}^2}\left(1\frac{R_\mathrm{i}^6}{R_0^6}\right)\right],$$
(95)
where $`Cf_2(\mathrm{})/f_0(\mathrm{})=\pi ^2/61`$. From (93) and (94) we note that $`\frac{d^n}{dt^n}R_0=\mathrm{O}(\alpha ^{n1})`$.
In order to solve for the spacetime metric, we need $`X_0`$. From the work of the previous section we know that $`X_0=\mathrm{tanh}u+\chi _2`$, where $`\chi _2`$ is given from equation (33) as
$$\chi _2=\frac{6}{R_0^2(1\dot{R}_{0}^{}{}_{}{}^{2})}\mathrm{sech}^2u_0^u\mathrm{cosh}^4u_{\mathrm{}}^uu\mathrm{sech}^4u,$$
(96)
and is clearly of order O($`\alpha ^2`$). Here, $`u`$ is the proper distance from the wall. Using (91) we note that
$`t`$ $`=`$ $`t^{}+{\displaystyle \frac{u\dot{R_0}(t^{})}{\sqrt{1\dot{R_0}^2(t^{})}}}`$ (98)
$`r`$ $`=`$ $`R_0(t^{})+{\displaystyle \frac{u}{\sqrt{1\dot{R_0}^2(t^{})}}},`$ (99)
where $`(t^{},u)`$ are the coordinates of the point $`(t,r)`$ in Gaussian Normal gauge.
In order to find the metric, first note that we can directly integrate (79) and (80) implicitly, finding
$`B_1`$ $`=`$ $`{\displaystyle \frac{1}{2r}}{\displaystyle _0^r}r^2\left[\left({\displaystyle \frac{1+\dot{R}_0^2(t^{})}{1\dot{R}_0^2(t^{})}}\right)X_0^2+V(X_0)\right]𝑑r`$ (101)
$`A_1`$ $`=`$ $`{\displaystyle \frac{1}{2r}}{\displaystyle _0^r}r^2\left[\left({\displaystyle \frac{1+\dot{R}_0^2(t^{})}{1\dot{R}_0^2(t^{})}}\right)X_0^2+V(X_0)\right]𝑑r+{\displaystyle _0^r}r\left({\displaystyle \frac{1+\dot{R}_0^2(t^{})}{1\dot{R}_0^2(t^{})}}\right)X_0^2𝑑r,`$ (102)
where we have used (96) to obtain $`\dot{u}^2+u^2=[1+\dot{R}_0^2(t^{})]/[1\dot{R}_0^2(t^{})]`$. We may now substitute the form of $`X_0(u,t^{})`$ to the required order, replace the $`r`$-integral by a $`u`$-integral along a line $`t=\text{const}`$, then Taylor expand $`t^{}`$ around $`u=0`$ to the required order in $`\alpha `$. For example, $`B_1`$ gives the mass function via $`B^2=12GM(r)/rGM(r)=ϵrB_1`$. Computing $`B_1`$ from the above expression to O($`\alpha ^2`$) yields
$$GM(r)=\frac{ϵR_0^2}{\sqrt{1\dot{R}_0^2}}\stackrel{~}{f}_0(u)+2ϵ\frac{R_\mathrm{i}^4}{R_0^3}\stackrel{~}{f}_1(u)+ϵR_\mathrm{i}^2\mathrm{sech}^2u\chi _2(u)+ϵ\frac{R_\mathrm{i}^6}{R_0^6}\left[\stackrel{~}{f}_2(u)+6\dot{R}_0^2u\stackrel{~}{f}_1(u)\right]$$
(103)
where we have put
$$\stackrel{~}{f}_n(u)=_{\mathrm{}}^u𝑑uu^nV(X_0).$$
(104)
This gives the ADM mass (using 95) as
$$GM_{\mathrm{ADM}}=\underset{r\mathrm{}}{lim}ϵrB_1=\frac{ϵR_0^2}{\sqrt{1\dot{R}_0^2}}\stackrel{~}{f}_0(\mathrm{})+ϵ\frac{R_\mathrm{i}^6}{R_0^6}\stackrel{~}{f}_2(\mathrm{})=\frac{4}{3}ϵ\left(R_\mathrm{i}^2+\frac{\pi ^2}{6}1\right),$$
(105)
which is indeed constant, and agrees to leading order with the expected result. Figure 2 shows the evolution of the $`g_{rr}`$ metric component up to the formation of a black hole.
The gravitational Nambu equation for $`ϵ0`$ at $`u=0`$ is given by
$$\ddot{R}+\frac{2}{R}nA^2+(A^{}A+\dot{R}\dot{B}B)n+\dot{R}^2\left(\frac{B^{}}{B}\frac{A^{}}{A}\right)+\dot{R}\left(\frac{\dot{B}}{B}\frac{\dot{A}}{A}\right)=O(\alpha ^2ϵ),$$
(106)
where $`n=B^2\dot{R}^2A^2`$. Note that the gravitational correction to $`A`$ and $`B`$ is O($`\delta `$) not O($`ϵ`$), where $`\delta =ϵ/\alpha `$ was defined in (37). This means that gravitational corrections to the wall motion potentially could appear at O($`\delta `$). Since the flat space wall equations are intially O($`\alpha `$), we could have the catastrophic situation that adding gravity swamps the wall motion, and has a superdominant effect. In fact, this turns out not to be the case. The computation of the $`A`$ and $`B`$ contributions to O($`\delta `$) in (106) shows that they cancel. Instead, the leading order correction appears at O($`\delta \alpha =ϵ`$), which is therefore subdominant to the flat space motion, and is plotted in figure 3
$$\ddot{R}=\frac{2}{R}(1\dot{R}^2)+2ϵ(1\dot{R}^2)^{3/2}+\frac{2ϵ(4\mathrm{ln}21)\dot{R}^2}{3R}(16\dot{R}^2+3\dot{R}^4)$$
(107)
Note that we have included the O($`\alpha ϵ`$) correction, since it is the same order as the finite width correction, however, this has not been used in the computation of the corrected Nambu trajectory in figure 3. We should stress that this correction does not mean that the wall trajectory is no longer Nambu, it simply gives the right gravitational alteration to the trajectory to allow the wall to remain a minimal surface in the now curved spacetime. What is important however, is that the correction is subdominant to the flat space motion and is a simple ‘nudge’; there is no evidence that adding gravity gives any catastrophic effect which could force a trajectory to be totally geodesic.
What is of real interest to us however, at least in the context of our general equation (76), is the correction to the Nambu motion. Indeed, having a particular example, we can explicitly calculate $`R_{ab}`$ from Gauss’s equation (9). Since we have not rescaled our quantities here, we can check that the correction term is indeed of the right order, and get an idea of the physical implication that the correction induces (at least in this particular example). We obtain,
$$\widehat{K}_0=\frac{\pi ^26}{\left(1\dot{R_0}^2\right)^{3/2}}\left(\frac{1}{R_0^3}+\frac{ϵ}{R_0}\frac{\dot{R_0}^2(1+\dot{R_0}^2)}{\left(1\dot{R_0}^2\right)^2}\right).$$
(108)
The first term is a correction term of order O($`\alpha ^3`$) due to the bending of the wall, already present in flat space-time as predicted in . The second correction term is due to self-gravity appearing at O($`\alpha ϵ`$), as we were indeed expecting (for a spherical wall $`\alpha >ϵ`$). Note that the finite width gravitational correction can dominate the extrinsic curvature correction, depending on the relative magnitudes of $`ϵ`$ and $`\alpha ^2`$. Finally from the overall positive sign we can deduce that the corrections to the Nambu motion induce a slowdown of the wall’s collapse.
Using (107) and (108) we can write the full equation of motion (i.e. including the dominant deviation to Nambu motion) for a collapsing spherical wall up to and including order $`\alpha ϵ`$,
$`\ddot{R}=`$ $``$ $`{\displaystyle \frac{2}{R}}(1\dot{R}^2)+2ϵ(1\dot{R}^2)^{3/2}+{\displaystyle \frac{2ϵ(4\mathrm{ln}21)\dot{R}^2}{3R}}(16\dot{R}^2+3\dot{R}^4)`$ (109)
$`+`$ $`(\pi ^26){\displaystyle \frac{ϵ}{R}}{\displaystyle \frac{\dot{R}^2(1+\dot{R}^2)}{\left(1\dot{R}^2\right)^2}}.`$ (110)
Before leaving the collapsing wall, it is worthwhile comparing our thick wall calculation with the results of Ipser and Sikivie, , obtained for the collapsing thin wall. In order to obtain the Israel thin wall approximation one should reintroduce the width parameter $`w`$ which has been set equal to unity throughout this analysis. Then $`\alpha =w/R`$ (where R is a typical radius of curvature of the wall) is a dimensionless parameter. The Israel limit is then obtained by letting $`\alpha `$ and $`ϵ`$ tend to zero, while keeping their quotient $`\delta `$ fixed. This amounts to keeping the horizon distance from the wall finite i.e., keeping the wall self-gravitating (as one should in this formalism). This also gives us the correct limit without involving the normal coordinate $`u`$ in the limiting procedure. Taking this limit in (110) gives
$$\ddot{R}=\frac{2}{R}(1\dot{R}^2)+2\delta (1\dot{R}^2)^{3/2}$$
(111)
which is obviously only correct to O($`\delta `$). The translation to the calculation of Ipser and Sikivie is not direct, since our coordinates correspond to the interior coordinates of their bubble, however, once the correct correspondence is made, we do indeed find precise agreement to order O($`\delta `$)$`=`$O($`G\sigma `$).
## V Conclusions
We have obtained a general equation of motion (76) describing a moving wall in curved spacetime. This was achieved by analytically solving the Einstein and scalar matter field equations order by order with respect to two parameters: the wall’s bending parameter, $`\alpha `$, and the gravitational strength parameter, $`ϵ`$, expressing the curving of spacetime. We then considered a particular example, the collapsing spherical domain wall, which is perhaps the simplest non-trivial example with both parameters different from zero. In the context of this example we found that the corrections to the flat space Nambu motion tended to slow down the wall’s collapse.
Throughout this paper we have considered a thick wall with scalar (Higgs) matter. This was done in order to examine the problem in its analytic (with respect to the spacetime metric) and most general context and also in order to pick up finite width gravitational corrections. As we discussed at the end of the previous section, in order to obtain the Israel thin wall approximation one takes the limit $`\alpha 0`$ and $`ϵ0`$ keeping $`\delta `$ fixed. Although the spherical wall was only explored to O($`\delta `$), the arguments in section III show in all generality that the corrections to the Nambu equation of motion $`K=0`$ are finite width corrections, and hence disappear in the thin wall limit, giving simply gravitational corrections to the flat space trajectory.
A totally geodesic ($`K_{ab}|_0=0`$) solution such as the plane-symmetric wall always trivially satisfies (76). With respect to the physical motion of the wall a totally geodesic solution is trivial and in particular a totally geodesic wall will not emit gravitational waves. What is of real interest physically, is the existence of non-totally geodesic solutions such as the spherical domain wall satisfying (76). There have been claims however (see for the related example of a cosmic string, , and references therein), that the presence of gravity in general constrains a defect’s core to be totally geodesic. For a domain wall in particular it has been claimed that the presence of gravity induces the wall to lose its dynamical degree of freedom and not to radiate . We should stress that throughout our treatment this constraint has not appeared and gravity affects the motion of the defect in a very natural way in the sense that corrections to the Nambu motion appear at the same order as in the case of a flat background spacetime. We suspect that the reason for this discrepancy is that in rather specific asymptotic boundary conditions have been placed on the spacetime, namely that it asymptote the static planar domain wall solution. If, however, a wall is oscillating, we expect that its effective equation of state will change, analogous to that of the wiggly cosmic string , which will change the spacetime metric even asymptotically. Mathematically, this can be seen via the divergent behaviour of the metric perturbations in the Gaussian Normal gauge due to proper motion of the wall, and was discussed in the context of higher dimensional domain walls in . Following , we expect that the equation of state of the wall will have the form $`\sigma T^2=\sigma _0^3`$, or, that perturbatively the effect on the energy-momentum tensor of the wall will be to increase the energy by $`\delta \sigma `$, and decrease the tension by $`\delta \sigma /2`$. Note that this perturbation is tracefree, and localised on the wall. In fact if we regard our domain wall as a $`2+1`$-dimensional universe, this energy momentum would be that appropriate to a radiation dominated cosmology. Such systems have been explored in a different context in, for example, .
As a final point it should be clear that although we have considered a four-dimensional spacetime our equations of motion are valid for any $`(n2)`$-brane of an $`n`$-dimensional spacetime. Our analysis however relies heavily on the fact that the wall is a hypersurface of the ambient spacetime i.e. that codimension is one. If we were to consider dynamics of strings for instance, the picture could in principle be quite different.
## Acknowledgements
We would like to thank Roberto Emparan, David Langlois, Joao Magueijo and Valery Rubakov for useful discussions. F.B. is supported by a FAPESP grant; C.C. is supported by PPARC and R.G. is supported by the Royal Society.
|
warning/0002/hep-ph0002295.html
|
ar5iv
|
text
|
# A note on spectator effects and quark-hadron duality in inclusive beauty decays
## I introduction
According to heavy quark expansion (HQE) , inclusive decay rates of heavy hadrons are expected to have almost same lifetimes. At the leading order of the HQE, the hadronic decay rate is, that of the free heavy quark decay, proportional to $`m^5`$, where $`m`$ is the heavy quark mass. The decay rate at the next-to-leading order (NLO) includes the terms of heavy quark motion inside the hadron and chromomagnetic mass splitting due to spin orientation of the heavy quark. Since the latter vanishes for baryons except for $`\mathrm{\Omega }_Q`$, the NLO decay rate splits up into mesonic and baryonic ones. However, the lifetimes of baryon and meson of given flavour is expected to be roughly the same. But, the experimental value of the ratio $`\tau (\mathrm{\Lambda }_b)/\tau (B)=0.78`$, much lower than the theoretical prediction of 0.9. In order to reduce the discrepancy, the contributions from the terms of $`O(1/m^3)`$ have necessarily to be included. At the third order in $`1/m`$, the decay rate becomes flavour dependent, involving both the heavy and light quark fields. The matrix element at order $`1/m^3`$ is
$$C_6(\mu )\frac{1}{2M}H|(\overline{Q}\mathrm{\Gamma }^\mu Q)(\overline{q}\mathrm{\Gamma }_\mu q)|H$$
(1)
where the coefficient functions, $`C_6`$, describe the spectator quarks processes such as Pauli interference (PI), weak annihilation (WA) and $`W`$-scattering (WS). The evaluation of the expectation values of the operators in the above equation is not straight. The difficult tast of this evaluation is nevertheless accomplished taking course to vacuum saturation approximation (in the case of meson), valence quark approximation (in the case of baryon), hadronic parameterisation and QCD sum rules . The result showed that the third order term, precisely the fourquark operators, does not adequatly contribute to enhance the decay rate of $`\mathrm{\Lambda }_b`$. In , Voloshin pointed out the relations between the decay rates of the heavy baryons. But these relations are not applicable to the cham sector for the reason mentioned below.
In this paper, we attempt to evaluate the expectation values of four-quark operators from the differences in the total decay rates of the $`b`$-hadrons. That means we assume that the heavy quark expansion, being asymptotic in nature, converges at $`O(1/m^3)`$ of the expansion. Already the next-to-leading order contribution due to terms of $`O(1/m^2)`$ is only about five percent of the leading order. Thus, it cannot be anticipated that the size of the terms at the third order in $`1/m`$ more than a few percent. On the other hand, the observation made above is applicable only to the beauty case, since
$$\frac{16\pi ^2}{m_c^3}C(\mu )O_6_{H_c}\frac{16\pi ^2}{m_b^3}C(\mu )O_6_{H_b}$$
(2)
where $`C(\mu )`$ stands for some structure involving $`c_\pm `$ and $`O_6_H`$ the dimension six four-quark operators (FQO) of hadron. Hence, in the background described, we make use of the total decay rates to obtain the expectation values of four quark operators (EVFQO) for $`b`$-hadrons. Thus, the present evaluation depends only on the heavy quark expansion and the $`SU(3)`$ flavour symmetry.
## II splitting of decay rates and expectation values of four-quark operators
The $`B`$ mesons, $`B^{}`$, $`B^0`$ and $`B_s^0`$, are triplet under $`SU(3)_f`$ flavour symmetry. Their total decay rate splits up due to its light quark flavour dependence at the third order in the HQE. The differences in the decay rates of the triplet, $`\mathrm{\Gamma }`$($`B_d^0`$) - $`\mathrm{\Gamma }`$($`B^{}`$), $`\mathrm{\Gamma }`$($`B_s^0`$) - $`\mathrm{\Gamma }`$($`B^{}`$) and $`\mathrm{\Gamma }`$($`B_s^0`$) - $`\mathrm{\Gamma }`$($`B_d^0`$), are related to the third order terms in $`1/m`$ by
$`\mathrm{\Gamma }(B_d^0)\mathrm{\Gamma }(B^{})`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2`$ (4)
$`\times \left\{Z_1{\displaystyle \frac{1}{3}}(2c_+c_{}+6)+2(c_+c_{}/2+1)\right\}O_6_{B_d^0B^{}}`$
$`\mathrm{\Gamma }(B_s^0)\mathrm{\Gamma }(B^{})`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2`$ (6)
$`\times \left\{Z_2{\displaystyle \frac{1}{3}}(2c_+c_{}+6)+2(c_+c_{}/2+1)\right\}O_6_{B_s^0B^{}}`$
$`\mathrm{\Gamma }(B_s^0)\mathrm{\Gamma }(B_d^0)`$ $`=`$ $`\mathrm{\Gamma }_0^{}(1x)^2`$ (8)
$`\times \left\{(Z_1Z_2){\displaystyle \frac{1}{3}}(2c_+c_{}+6)\right\}O_6_{B_s^0B_d^0}`$
where $`\mathrm{\Gamma }_0^{}`$ = $`2G_f^2|V_{cb}|^2m_b^2/3\pi `$, $`c_{}=c_+^2=(\alpha (m)/\alpha (m_W))^{2\gamma /\beta },\beta =112n_f/3`$ ($`n_f`$ being the number of flavour), $`\gamma =2`$, $`x`$ = $`m_c^2/m_b^2`$ and
$`Z_1`$ $`=`$ $`\left(cos^2\theta _c(1+{\displaystyle \frac{x}{2}})+sin^2\theta _c\sqrt{14x}(1x)\right)`$ (9)
$`Z_2`$ $`=`$ $`\left(sin^2\theta _c(1+{\displaystyle \frac{x}{2}})+cos^2\theta _c\sqrt{14x}(1x)\right)`$ (10)
$`O_6`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\overline{b}\mathrm{\Gamma }_\mu b)[(\overline{d}\mathrm{\Gamma }_\mu d)(\overline{u}\mathrm{\Gamma }_\mu u)]{\displaystyle \frac{1}{2}}(\overline{b}\mathrm{\Gamma }_\mu b)[(\overline{s}\mathrm{\Gamma }_\mu s)(\overline{q}\mathrm{\Gamma }_\mu q)]`$ (11)
with $`q=u,d`$. In eqns. (4-8), the $`rhs`$ contains the terms corresponding to the unsuppresed and suppressed nonleptonic decay rates and twice the semileptonic decay rates at the third order.
For the decay rates $`\mathrm{\Gamma }`$($`B^{}`$) = 0.617 $`ps^1`$, $`\mathrm{\Gamma }`$($`B^0`$) = 0.637 $`ps^1`$ and $`\mathrm{\Gamma }`$($`B_s^0`$) = 0.645 $`ps^1`$ , the EVFQO are obtained for $`B`$ meson, as an average from eqs. (4-8):
$$O_6_B=8.08\times 10^3GeV^3.$$
(12)
This is smaller than the one obtained in terms of the leptonic decay constant, $`f_B`$.
On the other hand, for the triplet baryons, $`\mathrm{\Lambda }_b`$, $`\mathrm{\Xi }^{}`$ and $`\mathrm{\Xi }^0`$, with $`\tau (\mathrm{\Lambda }_b)`$ $`<`$ $`\tau (\mathrm{\Xi }^0)\tau (\mathrm{\Xi }^{})`$, we have the relation between the difference in the total decay rates and the terms of $`O(1/m^3)`$ in the HQE, as
$$\mathrm{\Gamma }(\mathrm{\Lambda }_b)\mathrm{\Gamma }(\mathrm{\Xi }^0)=\frac{3}{8}\mathrm{\Gamma }_0^{}(c_+(2c_{}+c_+2)O_6_{\mathrm{\Lambda }_b\mathrm{\Xi }^0}$$
(13)
We obtain the EVFQO for the baryon
$$O_6_{\mathrm{\Lambda }_b\mathrm{\Xi }^0}=3.072\times 10^2GeV^3$$
(14)
where we have used the decay rates corresponding to the lifetimes 1.24 $`ps`$ and 1.39 $`ps`$ of $`\mathrm{\Lambda }_b`$ and $`\mathrm{\Xi }^0`$ respectively. The EVFQO for baryon is about 3.8 times larger than that of B. But, it is about 3.5 - 4.0, dpending on the changes in the mass and other sources of uncertainity.For these values
$$\tau (\mathrm{\Lambda }_b)/\tau (B)=0.810.84$$
(15)
Using the experimental value of $`\tau (B^{})`$ = 1.55 $`ps`$ alongwith the above theoretical value, the lifetime of $`\mathrm{\Lambda }_b`$ turns out to be
$$\tau (\mathrm{\Lambda }_b)=\frac{\mathrm{\Gamma }(\mathrm{\Lambda }_b)}{\mathrm{\Gamma }(B)}\tau (B^{})=1.261.3ps.$$
(16)
The central values would change little bit in view of the uncertainities in the parameters like the heavy quark mass and the kinetic energy paprameter.
The expectation values for baryon is quite large. In fact, Rosner found it to be 1.8 times larger than that of meson which accounts for about 25%, 45% if renormalisation group improved , of the needed enhancement in the decay rate. However, what we obtain is model independent estimate. We have only used the experimental total decay rates. The result is surprising as well as genuine, if one has to believe the experimental values used. Still, there is room for uncertainity of few percent. On the other hand, the generic structure we employed can decomposed into as found in by Neubert and Sachrajda. Using them will give an improved estimate but in that case it may become somewhat model dependent.
We will briefly take note two previous works. In , using potential model, the expectation values of four-quark operators of meson and baryons are obtained and the ratio $`\tau (\mathrm{\Lambda }_b)/\tau (B)`$ is estimated to be in the range 0.79 - 0.84. Using QCD sum rules , assuming a different duality ansatz, in the ratio is got to be 0.81. Though these methods arise on different basis, they seem to agree with the present model independent evaluation.
## III quark-hadron duality: a comment
Assuming the convergence of the heavy quark expansion is valid one as long as there is an uncertainity of few percent. This would also positively suggest that the quark-hadron duality violating oscillating component might be small. So such violation would not deter a decent determination of quantities of interest in the heay quark expansion like the lifetimes of beauty hadrons, the semileptonic branching ratio and the charm counting. For the reasons mentioned in the beginning, the assumption on convergence cannot be made for charmed case.
There are renormalon contribution from the perturbative part of the expansion . They are IR renormalons of the order $`\mathrm{\Lambda }_{QCD}`$. We, as of now, don’t have deep insight of it. They would be expected to differ for meson and baryon. Because this contribution does not correspond to any local operators of the thoery and independent of the heavy quark mass, a difference of about 50 to 100 MeV between meson and baryon would imply much significance for the quatities described by the heavy quark expansion. We cannot on obviuos terms argue that renormalons are related to the assumption of quark-hadron duality. On the other hand, it will shed light on the quantities concerned and the underlying assumption versus the heavy quark expansion.
## IV concluding remarks
We have estimated the expectation values of the four-quark operators of beauty hadron and found that they yielded the ratio of lifetimes of $`\mathrm{\Lambda }_b`$ and $`B`$ meson closer to the experimental value. We conclude by noting that the model indepent prediction will be an accurate one when other structure of the currents given in are also included.
###### Acknowledgements.
The author is grateful to Prof. H. Yamamoto, Prof. P. R. Subramanian, Dr. D. Caleb Chanthi Raj, and Mr R. Justin Joseyphus for discussions and encouragement. He acknowledges UGC for the support through its Special Assistance Programme (Phase III).
|
warning/0002/physics0002047.html
|
ar5iv
|
text
|
# Untitled Document
JAYNES-CUMMINGS MODEL WITH DEGENERATE ATOMIC LEVELS
V. A. Reshetov
Department of Physics, Tolyatti Pedagogical Institute, 13 Boulevard Korolyova, 445859 Tolyatti, Russia
## Abstract
The Jaynes-Cummings model describing the interaction of a single linearly-polarized mode of the quantized electromagnetic field with an isolated two-level atom is generalized to the case of atomic levels degenerate in the projections of the angular momenta on the quantization axis, which is a usual case in the experiments. This generalization, like the original model, obtains the explicit solution. The model is applied to calculate the dependence of the atomic level populations on the angle between the polarization of cavity field mode and that of the laser excitation pulse in the experiment with one-atom micromaser.
The Jaynes-Cummings model describes the interaction of a single linearly-polarized mode of the quantized electromagnetic field with an isolated two-level atom. The full set of states of the system atom+field is
$$|n,\alpha >=|n>|\alpha >,,n=0,1,\mathrm{},\alpha =b,c,$$
where $`n`$ is the number of photons in the field mode, while $`b`$ and $`c`$ denote the upper and lower atomic levels correspondingly. This model is applied successfully to analyse the results of the experiments with one-atom micromasers (see, e.g., ). However, the levels of an isolated atom are degenerate in the projections of the total elctronic angular momenta on the quantization axis, so that the original Jaynes-Cummings model becomes, in general, invalid.
Now, let us take into account the degeneracy of atomic levels. Then, the full set of states of the system becomes
$$|n,J_\alpha ,m_\alpha >=|n>|J_\alpha ,m_\alpha >,n=0,1,\mathrm{},m_\alpha =J_\alpha ,\mathrm{},J_\alpha ,\alpha =b,c,$$
where $`J_b`$ and $`J_c`$ are the values of the total electronic angular momenta of resonant levels, while $`m_b`$ and $`m_c`$ are their projections on the quantization axis - the cartesian axis Z, which is directed along the polarization vector of the field mode.
The Hamiltonian of the system may be written as
$$\widehat{H}=\widehat{H}_F+\widehat{H}_A+\widehat{V},$$
(1)
where
$$\widehat{H}_F=\mathrm{}\omega \widehat{a}^+\widehat{a}$$
is a free-field Hamiltonian,
$$\widehat{H}_A=\frac{1}{2}\mathrm{}\omega _0(\widehat{n}_b\widehat{n}_c)$$
is a free-atom Hamiltonian,
$$\widehat{V}=(\widehat{𝐃}\widehat{𝐄})$$
is an operator of field-atom interaction, while $`\widehat{a}^+`$ and $`\widehat{a}`$ are the operators of the creation and annihilation of photons with the frequency $`\omega `$ in the field mode,
$$\widehat{n}_\alpha =\underset{m_\alpha =J_\alpha }{\overset{J_\alpha }{}}|J_\alpha ,m_\alpha ><J_\alpha ,m_\alpha |,\alpha =b,c,$$
are the operators of total populations of resonant atomic levels $`b`$ and $`c,`$ $`\omega _0`$ is the frequency of the optically-allowed atomic transition $`J_bJ_c,`$
$$\widehat{𝐄}=𝐞\widehat{a}+𝐞^{}\widehat{a}^+,$$
$$𝐞=ı𝐥_z\sqrt{\frac{2\pi \mathrm{}\omega }{V}},$$
is the electric field intensity operator, $`V`$ and $`𝐥_z`$ being the resonator cavity volume and the unit vector of the cartesian axis Z,
$$\widehat{𝐃}=\widehat{𝐝}+\widehat{𝐝}^+,$$
$$\widehat{𝐝}=\underset{m_b,m_c}{}𝐝_{m_cm_b}^{J_cJ_b}|J_c,m_c><J_b,m_b|,$$
is the dipole moment operator of the atomic transition $`J_bJ_c,`$ which matrix elements are defined through Wigner 3j-symbols (see, e.g., ):
$$(d_q)_{m_cm_b}^{J_bJ_c}=d(1)^{J_bm_b}\left(\begin{array}{ccc}J_b& 1& J_c\\ m_b& q& m_c\end{array}\right),$$
$`d=d(J_bJ_c)`$ -being a reduced matrix element and $`d_q(q=1,0,1)`$ \- are the circular components of vector $`𝐝.`$
In the interaction representation
$$\widehat{f}_I=\mathrm{exp}\left(\frac{ı\widehat{H}_0t}{\mathrm{}}\right)\widehat{f}\mathrm{exp}\left(\frac{ı\widehat{H}_0t}{\mathrm{}}\right),$$
where
$$\widehat{H}_0=\mathrm{}\omega \left\{\widehat{a}^+\widehat{a}+\frac{1}{2}(\widehat{n}_b\widehat{n}_c)\right\},$$
the operators $`\widehat{a}`$ and $`\widehat{𝐝}`$ obtain the oscillating factors
$$\widehat{a}_I=\widehat{a}\mathrm{exp}(ı\omega t),\widehat{𝐝}_I=\widehat{𝐝}\mathrm{exp}(ı\omega t).$$
Then, in the rotating wave approximation, when the terms oscillating with double frequences are neglected, the Hamiltonian (1) becomes
$$\widehat{H}_I=\widehat{H}_0\mathrm{}\widehat{\mathrm{\Omega }},$$
where
$$\widehat{\mathrm{\Omega }}=\frac{\delta }{2}(\widehat{n}_b\widehat{n}_c)+ıg(\widehat{a}\widehat{p}^+\widehat{a}^+\widehat{p}),$$
while
$$\delta =(\omega \omega _0)$$
is the frequency detuning,
$$g=\sqrt{\frac{2\pi d^2\omega }{\mathrm{}V}}$$
and
$$\widehat{p}=\underset{m}{}\alpha _m|J_c,m><J_b,m|,$$
$$\alpha _m=(1)^{J_bm}\left(\begin{array}{ccc}J_b& 1& J_c\\ m& 0& m\end{array}\right).$$
From the equation
$$\frac{d\widehat{\sigma }}{dt}=\frac{ı}{\mathrm{}}[\widehat{\sigma },\widehat{H}]$$
for the system density matrix $`\widehat{\sigma }`$ follows the equation
$$\frac{d\widehat{\rho }}{dt}=ı[\widehat{\mathrm{\Omega }},\widehat{\rho }]$$
(2)
for the density matrix
$$\widehat{\rho }=\mathrm{exp}\left(\frac{ı\widehat{H}_0t}{\mathrm{}}\right)\widehat{\sigma }\mathrm{exp}\left(\frac{ı\widehat{H}_0t}{\mathrm{}}\right)$$
in the interaction representation. The formal solution of the equation (2) is obtained immediately
$$\widehat{\rho }=\mathrm{exp}\left(ı\widehat{\mathrm{\Omega }}t\right)\widehat{\rho }_0\mathrm{exp}\left(ı\widehat{\mathrm{\Omega }}t\right),$$
where $`\widehat{\rho }_0`$ is the initial density matrix of the system.
In order to obtain the average value
$$<\widehat{f}>=Tr\left(\widehat{\rho }\widehat{f}_I\right)$$
of any operator $`\widehat{f}`$ it is necessary to calculate the matrix elements
$$<n,J_\alpha ,m|\mathrm{exp}(ı\widehat{\mathrm{\Omega }}t)|n_1,J_\beta ,m_1>,\alpha ,\beta =b,c,$$
of the evolution operator. The explicit analytical expressions for these matrix elements may be derived with the use expansion
$$\mathrm{exp}(ı\widehat{\mathrm{\Omega }}t)=\underset{n=0}{\overset{\mathrm{}}{}}\frac{(ı\widehat{\mathrm{\Omega }}t)^n}{n!},$$
since the operator
$$\widehat{\mathrm{\Omega }}^2=\frac{\delta ^2}{4}\widehat{1}+g^2\left\{\widehat{R}_c\widehat{n}+\widehat{R}_b(\widehat{n}+1)\right\},$$
where
$$\widehat{R}_\beta =\underset{m}{}\alpha _m^2|J_\beta ,m><J_\beta ,m|,\beta =b,c,$$
is diagonal:
$$\widehat{\mathrm{\Omega }}^2|n,J_b,m>=\mathrm{\Omega }_{n+1,m}^2|n,J_b,m>,$$
$$\widehat{\mathrm{\Omega }}^2|n,J_c,m>=\mathrm{\Omega }_{n,m}^2|n,J_c,m>.$$
Here
$$\mathrm{\Omega }_{n,m}=\sqrt{\frac{\delta ^2}{4}+\alpha _m^2g^2n}.$$
(3)
So, the matrix elements of the evolution operator are:
$$<n,J_b,m|\mathrm{exp}(ı\widehat{\mathrm{\Omega }}t)|n_1,J_b,m_1>=$$
$$\delta _{n,n_1}\delta _{m,m_1}\left\{\mathrm{cos}(\mathrm{\Omega }_{n+1,m}t)+\frac{ı\delta }{2\mathrm{\Omega }_{n+1,m}}\mathrm{sin}(\mathrm{\Omega }_{n+1,m}t)\right\},$$
(4)
$$<n,J_c,m|\mathrm{exp}(ı\widehat{\mathrm{\Omega }}t)|n_1,J_c,m_1>=$$
$$\delta _{n,n_1}\delta _{m,m_1}\left\{\mathrm{cos}(\mathrm{\Omega }_{n,m}t)\frac{ı\delta }{2\mathrm{\Omega }_{n,m}}\mathrm{sin}(\mathrm{\Omega }_{n,m}t)\right\},$$
(5)
$$<n,J_b,m|\mathrm{exp}(ı\widehat{\mathrm{\Omega }}t)|n_1,J_c,m_1>=$$
$$\delta _{n+1,n_1}\delta _{m,m_1}g\alpha _m\sqrt{n+1}\frac{\mathrm{sin}(\mathrm{\Omega }_{n+1,m}t)}{\mathrm{\Omega }_{n+1,m}}.$$
(6)
In the experiment the average total population
$$n_b=Tr\left\{\widehat{n}_b\mathrm{exp}(ı\widehat{\mathrm{\Omega }}T)\rho _0\mathrm{exp}(ı\widehat{\mathrm{\Omega }}T)\right\}$$
of the upper resonant level $`b`$ after the atom passes through the resonant cavity, where T is the time of interaction, was detected. As follows from (4)-(6),
$$n_b=\underset{n,m}{}f_{nn}n_{mm}^b\left\{\mathrm{cos}^2(\mathrm{\Omega }_{n+1,m}T)+\frac{\delta ^2}{4\mathrm{\Omega }_{n+1,m}^2}\mathrm{sin}^2(\mathrm{\Omega }_{n+1,m}T)\right\},$$
(7)
where the atomic and field subsystems at the initial instant of time, when the atom enters the cavity, are independent and the initial density matrix of the system is
$$\widehat{\rho }_0=\widehat{\rho }_0^A\widehat{\rho }_0^F,$$
while
$$\widehat{\rho }_0^A=\underset{m,m_1}{}n_{mm_1}^b|J_b,m><J_b,m_1|,$$
$$\widehat{\rho }_0^F=\underset{n,n_1}{}f_{nn_1}|n><n_1|.$$
The cavity temperature in was low, so that the initial field may be considered to be in its vacuum state:
$$f_{n,n_1}=\delta _{n,0}\delta _{n_1,0}.$$
Then, in case of exact resonance $`\delta =0`$ the equation (7) simplifies to
$$n_b=\underset{m}{}n_{mm}^b\mathrm{cos}^2(\theta _m),\theta _m=\alpha _mgT.$$
(8)
Here $`n_{mm}^b`$ is the initial population of the Zeeman sublevel $`m`$ of the upper level $`b`$ . The resonant levels $`b`$ and $`c`$ in the experiment were the Rydberg states of the rubidium atom with the angular momenta $`J_b=3/2`$ and $`J_c=3/2`$ or $`J_c=5/2`$ . The upper level $`b`$ was excited from the ground state $`a`$ with the angular momentum $`J_a=1/2`$ by the linearly-polarized laser pulse. The evolution of the atomic density matrix under the action of the excitation pulse in the rotating-wave approximation is desribed by the equation
$$\frac{d\widehat{\rho }^A}{dt}=\frac{ı}{\mathrm{}}[\widehat{\rho }^A,\widehat{V}_e],$$
(9)
where
$$\widehat{V}_e=(\widehat{𝐝}_e^+𝐞_e+\widehat{𝐝}_e𝐞_e^{})$$
is the interaction operator of an atom with the cohernt resonant laser field,
$$\widehat{𝐝}_e=\underset{m_b,m_a}{}(𝐝_e)_{m_am_b}^{J_aJ_b}|J_a,m_a><J_b,m_b|,$$
is the dipole moment operator of the atomic transition $`J_bJ_a,`$ $`𝐞_e=e_e𝐥`$ is the slowly-varying amplitude of laser field, $`𝐥`$ is its unit polarization vector, which constitutes the angle $`\psi `$ with the polarization of the cavity field mode:
$$l_q=\mathrm{cos}\psi \delta _{q,0}+\frac{1}{\sqrt{2}}\mathrm{sin}\psi (\delta _{q,1}\delta _{q,1}).$$
For purposes of simplicity we shall consider the exciting pulses with small areas
$$\theta _e=\frac{|d_e|}{\mathrm{}}_0^{T_e}e_e(t)𝑑t1$$
(10)
(though in case of transition $`3/21/2`$ in the experiment the following results do not depend on the exciting pulse area), $`d_e=d(J_bJ_a)`$ is a reduced matrix element of the dipole moment operator for the transition $`J_bJ_a`$ , $`T_e`$ is the exciting pulse duration. Under the limitation (10) we obtain from (9) the density matrix of an atom (renormalized to unity trace)
$$\widehat{\rho }_0^A=\frac{(\widehat{𝐝}_e^+𝐥)\widehat{\rho }_{in}^A(\widehat{𝐝}_e𝐥)}{Tr\left\{(\widehat{𝐝}_e^+𝐥)\widehat{\rho }_{in}^A(\widehat{𝐝}_e𝐥)\right\}}$$
(11)
at an instant when it enters the cavity. Here
$$\widehat{\rho }_{in}^A=\frac{1}{(2J_a+1)}\underset{m}{}|J_a,m><J_a,m|$$
is the initial equilibrium atomic density matrix before the incidence of the exciting pulse. As follows from (11) the Zeeman sublevel populations in (8) are
$$n_{mm}^b=<J_b,m|\widehat{\rho }_0^A|J_b,m>=a_m\mathrm{cos}^2\psi +b_m\mathrm{sin}^2\psi ,$$
where
$$a_m=3\left(\begin{array}{ccc}J_b& 1& J_a\\ m& 0& m\end{array}\right)^2,$$
$$b_m=\frac{3}{2}\left\{\left(\begin{array}{ccc}J_b& 1& J_a\\ m& 1& m+1\end{array}\right)^2+\left(\begin{array}{ccc}J_b& 1& J_a\\ m& 1& m1\end{array}\right)^2\right\}.$$
In case of transitions $`J_b=3/2J_a=1/2`$
$$n_{1/2,1/2}^b=n_{1/2,1/2}^b=\frac{1}{2}\frac{3}{8}\mathrm{sin}^2\psi ,$$
$$n_{3/2,3/2}^b=n_{3/2,3/2}^b=\frac{3}{8}\mathrm{sin}^2\psi ,$$
and the total population (8) of the upper level after the atom leaves the cavity is
$$n_b=\left(1\frac{3}{4}\mathrm{sin}^2(\psi )\right)\mathrm{cos}^2(\theta )+\frac{3}{4}\mathrm{sin}^2(\psi )\mathrm{cos}^2(3\theta ),$$
$$\theta =\frac{gT}{2\sqrt{15}},$$
for the transitions $`J_b=3/2J_c=3/2`$ and
$$n_b=\left(1\frac{3}{4}\mathrm{sin}^2(\psi )\right)\mathrm{cos}^2(\theta )+\frac{3}{4}\mathrm{sin}^2(\psi )\mathrm{cos}^2\left(\sqrt{\frac{3}{2}}\theta \right),$$
$$\theta =\frac{gT}{\sqrt{10}},$$
for the transitions $`J_b=3/2J_c=5/2`$ .
The atom behaves like a two-level system - the population
$$n_b=\mathrm{cos}^2(\theta )$$
oscillates with a single Rabi frequency - only in case when the polarizations of the exciting laser pulse and of the cavity field mode coincide -
$`\psi =0,`$ otherwise the oscillations with more than one Rabi frequencies appear.
So, the Jaynes-Cummings model generalized to the case of the atomic levels degenerate in the projections of the angular momenta on the quantization axis is a useful tools for the description of the polarization properties of one-atom micromasers.
References
Jaynes E T, Cummings F W 1963 Proc. IEEE 51 89
Walther H 1995 Ann.N.Y.Acad.Sci. 755 133
Sobelman I I 1972 Introduction to the Theory of Atomic Spectra
(New York:Pergamon)
|
warning/0002/astro-ph0002385.html
|
ar5iv
|
text
|
# WHERE DO COOLING FLOWS COOL?
## 1 INTRODUCTION
Perhaps the most perplexing and long-standing problem associated with galactic and cluster cooling flows is the uncertain physical nature and spatial distribution of the gas that cools. The apparent absence of large masses of cooled gas in elliptical galaxies has led some to argue that little or no cooling actually occurs and to postulate some source of heating that offsets the radiative losses in X-ray emission. But the energy required to balance radiative losses is prohibitively large and appropriate heating sources may not be universally available. If the expected radiative cooling actually occurs, two questions arise: (1) What is the nature of the objects that condense from the hot gas? and (2) Where is most of the cooled mass located in the galaxy? Regarding the first question, a variety of physical arguments support the hypothesis, or even the inevitability, of low mass star formation (Fabian, Nulsen & Canizares 1982; Thomas 1986; Cowie & Binney 1988; Vedder, Trester, & Canizares 1988; Sarazin & Ashe 1989; Ferland, Fabian & Johnstone 1994; Mathews & Brighenti 1999). Here we shall address the second question in the context of cooling flows in elliptical galaxies where the known stellar mass and light profiles strongly constrain the spatial distribution of cooled gas.
We adopt the generally accepted hypothesis that only stars of very low mass (e.g.
<0.1
<absent0.1\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1 $`M_{}`$) form in cooling flows (e.g. Ferland, Fabian & Johnstone 1994), so that the mass to light ratio of the young stellar population formed from the cooling gas is essentially infinite. In view of the difficulties we encounter with this hypothesis, described below, it seems more likely that the stellar population formed from cooled gas extends to somewhat more massive stars that are optically luminous.
Our gas dynamical models for the evolution of hot interstellar gas in giant ellipticals indicate that the origin of the gas varies with galactic radius. Most of the gas in the inner, optically luminous regions originates from the ejected envelopes of evolving stars; gas in the outer halo is supplied by cosmological secondary infall or tidal acquisitions from neighboring galaxies (Mathews & Brighenti 1998b). Circumgalactic gas around massive ellipticals is enriched by Type II supernovae that accompanied early star formation. The variability of circumgalactic gas among luminous ellipticals is responsible for some of the enormous dispersion in X-ray luminosity $`L_x`$ among ellipticals of similar optical luminosity $`L_B`$ (Mathews & Brighenti 1998a).
Since the hot interstellar gas in a bright elliptical emits observable X-rays, it is clearly losing energy. However, as the gas loses energy it is compressed toward the galactic center by gravitational forces and $`Pdv`$ work maintains the high temperatures observed, $`T10^7`$ K, producing a galactic cooling flow. The positive interstellar temperature gradients typically observed within a few effective radii are often cited as evidence of radiative cooling in a cooling flow, but this cooling is due instead to the mixing of somewhat cooler, locally virialized gas ejected from stars with hotter gas arriving from larger galactic radii (Mathews & Brighenti 1998b; Brighenti & Mathews 1998, 1999a). If large entropy fluctuations are present in the hot gas, catastrophic cooling can occur at any radius in the flow. Regions of low entropy (low temperature, high density) radiate more and cool sooner. The amplitude distribution of entropy fluctuations in the interstellar gas determines the radius where cooling mass dropout occurs in the cooling flow. For example, if the entropy in some region in the flow is only slightly less than in the ambient flow, the differential radiative cooling will be small and the region will cool out of the flow at small galactic radii; conversely, localized regions with entropy much lower than the ambient flow cool rapidly and deposit their mass at large radii. Some possible sources of interstellar entropy variations are stellar winds, explosions of Type Ia supernovae, non-uniform SNII heating at early times, and mergers with small, gas-rich galaxies.
The total rate that mass cools and drops out of the flow is closely related to the X-ray luminosity $`L_x`$. The X-ray luminosity can be approximately expressed as the product of the total cooling rate $`\dot{M}`$ and the enthalpy per gram in the hot gas, or
$$\dot{M}=\left(\frac{2\mu m_p}{5kT}\right)L_{x,bol}2.5M_{}\mathrm{yr}^1.$$
Here we have used data from the giant Virgo elliptical NGC 4472: $`T1.3\times 10^7`$ K; $`L_x(0.54.5\mathrm{keV})=4.5\times 10^{41}`$; $`L_{x,bol}1.6L_x(0.54.5\mathrm{keV})`$. If $`L_x`$ and $`T`$ are reasonably constant over the Hubble time, a mass $`M_{cg}3\times 10^{10}`$ $`M_{}`$ of cold gas is expected to condense from the hot ISM somewhere within NGC 4472. Although this mass is very large, it is only about 4 percent of the total stellar mass in NGC 4472 today. The mass that cools can therefore be ignored if it is widely distributed throughout the galactic volume. However, the central concentration of H$`\alpha `$ emission in ellipticals (e.g. Macchetto et al. 1996) suggests that the cooling is concentrated toward the galactic center where the interstellar density is highest and the bulk of the X-ray energy is emitted.
The motivation of this paper is to explore a variety of options for the mass dropout profile of cooled gas in bright ellipticals appropriately constrained by the known radial distributions of total stellar and non-baryonic mass. The radial mass dropout profile of cooled gas cannot be determined from first principles because the distribution and amplitude of the entropy and magnetic fluctuations in the hot gas are unknown and difficult to evaluate from simple physical arguments. Nevertheless, the total mass of cooled gas inside an effective (half-light) radius $`r_e`$ must be consistent with the mass to light ratio determined from stellar velocities and with the total mass inferred from X-ray observations within $`r_e`$. Assuming that the stellar mass to light ratio is uniform with radius, the stellar mass $`M_{}(r)`$ and the X-ray mass $`M_x(r)`$ appear to be in nearly perfect agreement for two bright Virgo ellipticals in the range 0.1re
<r
<re
<0.1subscript𝑟𝑒𝑟
<subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e} (Brighenti & Mathews 1997a). Because of constraints on the mass distribution of cooled gas provided by X-ray and stellar dynamical observations, galactic cooling flows provide a critical venue for testing the physics of mass deposition in cooling flows.
The mass of cold (T
<104
<𝑇superscript104T\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}10^{4} K) gas $`M_{cg}`$ actually observed in ellipticals is many orders of magnitude less than the total cooled mass estimated above. For example, neither HI nor H<sub>2</sub> gas has been observed in NGC 4472, only upper limits, Mcg
<107
<subscript𝑀𝑐𝑔superscript107M_{cg}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}10^{7} $`M_{}`$ (Bregman, Roberts & Giovanelli 1988; Braine, Henkel & Wiklind 1988). If stars form they must either be indistinguishable from the old stellar population or non-luminous. We explore here the possibility that most of the cooled gas forms dense baryonic clouds or stars that are dark at optical and radio frequencies. The very low mass stars advanced by Ferland, Fabian & Johnstone (1994) satisfy this invisibility criterion, while the star formation models of Mathews & Brighenti (1999) indicate that (luminous) stars of mass $`12`$ $`M_{}`$ can form in galactic cooling flows.
X-ray studies indicate significant masses of cold, absorbing gas in cluster and galactic cooling flows (e.g. White et al. 1991; Allen et al. 1993; Fabian et al. 1994; Allen & Fabian 1997; Buote 1999), but these results are inconsistent with the absence of radio frequency emission from the cold gas (Braine & Dupraz 1994; Donahue & Vogt 1997) and should be regarded as controversial until this inconsistency is resolved. Even taken at face value, the total mass of cooled gas implied by the X-ray absorption in cluster cooling flows is typically only a small fraction of the total mass that should have cooled in a Hubble time (Allen & Fabian 1997; Wise & Sarazin 1999), implying that most of the cooled gas may have formed stars. If cooled gas forms into small stars, these stars will have apogalatica near their point of origin where they will spend most of their orbital time. We shall assume that the gas mass that cools and drops out of the flow contributes optically dark (stellar) mass at the radius where the cooling occurred.
In the following we describe a series of gas dynamical calculations for the evolution of X-ray emitting interstellar gas over the Hubble time and investigate a variety of assumptions about the radial distribution of optically dark cooled gas. To be specific, we compare our models with the well-observed massive elliptical NGC 4472. We find that the mass of cooled gas contributes significantly to dynamical mass to light determinations within $`r_e`$ based on stellar velocities.
## 2 KNOWN STELLAR AND DARK MASS DISTRIBUTION IN NGC 4472
The E2 elliptical NGC 4472 is a luminous, slowly rotating \[$`(v/\sigma )_{}=0.43`$\] galaxy in the Virgo cluster. With an adopted distance $`d=17`$ Mpc, its optical luminosity is $`L_B=7.89\times 10^{10}`$ $`L_B`$ and its half light or effective radius $`r_e=1.733^{}`$ is $`8.57`$ kpc (Faber et al. 1989). The total stellar mass $`M_t=7.26\times 10^{11}`$ $`M_{}`$ is found from the mass to light ratio $`M/L_B=9.2`$ determined by van der Marel (1991) with a two-integral stellar distribution function. This mass to light ratio is appropriate to the galactic region within about 0.4$`r_e`$ where stellar velocities are well determined, although the mass determined from X-ray observations suggests that $`M/L_B`$ remains constant to at least $`r_e`$ (Brighenti & Mathews 1997a). If $`M/L_B`$ is spatially constant, the stellar mass also has a de Vaucouleurs profile. Within a central core or break radius $`r_b=2.41^{\prime \prime }`$ ($`200`$ pc) the stellar density profile flattens (Faber et al. 1997), but we shall not consider this small feature here. NGC 4472 contains a central black hole of mass $`M_{bh}=2.9\times 10^9`$ $`M_{}`$ (Magorrian et al. 1998).
The total mass distribution in luminous ellipticals can most easily be determined from the radial variation of density and temperature in the hot interstellar gas, assuming hydrostatic equilibrium. Figure 1 illustrates the interstellar density and temperature profiles in NGC 4472. The filled circles in Figure 1 are Einstein HRI observations (Trinchieri, Fabbiano & Canizares 1986) and open circles are ROSAT HRI and PSPC data from Irwin & Sarazin (1996) (also see Forman et al. 1993). The $`T(r)`$ and $`n(r)`$ profiles have been fit with analytic curves as described by Brighenti & Mathews (1997a).
Hydrostatic equilibrium in the hot interstellar gas is an excellent approximation since the cooling flow velocity is very subsonic. The total mass interior to radius $`r`$ determined from X-ray observations is
$$M_x(r)=\frac{kT(r)r}{G\mu m_p}\left(\frac{d\mathrm{log}\rho }{d\mathrm{log}r}+\frac{d\mathrm{log}T}{d\mathrm{log}r}+\frac{P_m}{P}\frac{d\mathrm{log}P_m}{d\mathrm{log}r}\right)$$
(1)
where $`m_p`$ is the proton mass and $`\mu =0.61`$ is the mean molecular weight. The last term, representing the possibility of magnetic pressure $`P_m=B^2/8\pi `$, is negative if $`dP_m/dr<0`$ as seems likely. If the magnetic term is important but not included, the total mass will be underestimated. Assuming $`P_m=0`$, the total mass $`M_x(r)`$ in NGC 4472 is shown with a solid line in Figure 1b.
In the outer halo, r
>re
>𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}}r_{e}, the total mass is dominated by the dark halo. The dark halo mass distribution in NGC 4472 can be approximated with an NFW halo profile (Navarro, Frenk & White 1996) of virial mass $`M_h=4\times 10^{13}`$ $`M_{}`$ although the observed halo is somewhat less centrally peaked than NFW (Brighenti & Mathews 1999a). Within $`r_e`$ the contribution of the dark NFW halo mass in the model is small; for example at $`r<r_e/3`$ the total mass to light ratio is $`M/L_B=10.18`$, only 10 percent greater than the dynamic value 9.2 determined in r
<0.4re
<𝑟0.4subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.4r_{e}.
It is remarkable that the total mass found from the X-ray data $`M_x(r)`$ is nearly identical to the expected dynamical mass $`M_{}(r)`$ (based on stellar velocities and $`M/L`$) in the range 0.1re
<r
<re
<0.1subscript𝑟𝑒𝑟
<subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e} (Figure 1). An almost identical agreement in this radius range is indicated by X-ray observations of another bright Virgo elliptical, NGC 4649 (Brighenti & Mathews 1997a). In this important region the hot gas is confined by the stellar potential. The excellent agreement of the stellar and X-ray masses supports the consistency of two radically different mass determinations: from stellar velocities and from the radial equilibrium of hot interstellar gas. The apparent agreement of the X-ray and stellar masses in this range of galactic radii also indicates that the hydrostatic support of the hot gas is not strongly influenced by local magnetic fields and rotation.
However, it is not obvious why $`M/L_B`$ would be constant with galactic radius, particularly when the cooling dropout mass is considered, and why the agreement between stellar dynamic and X-ray masses no longer obtains in the central regions r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e}. In this central region the total mass indicated by the X-ray observations in Figure 1 is considerably less than the expected mass based on an assumed de Vaucouleurs profile and constant mass to light ratio. This type of deviation could be due to magnetic or other non-thermal pressure in this region, to rotation, or to local cooling dropout in the hot interstellar gas. The lower mean temperature in cooling regions lowers the total apparent gas temperature and results in an underestimate of the total internal mass (equation 1).
We assume that currently available Einstein and ROSAT observations are accurate in the central region of NGC 4472, r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e}. These observations have been reduced assuming no (non-Galactic) photoelectric absorption by low temperature gas in the central regions. If X-ray absorption is present, the true hot gas density would be more centrally peaked than shown in Figure 1 and the total mass indicated by the X-ray observations would increase. Buote (1999) finds that two-temperature models fit the X-ray spectrum for NGC 4472 quite well. The two temperatures do not necessarily need to be spatially mixed; they could also approximately represent the range of the radial temperature variation observed in NGC 4472. In Buote’s two temperature model, only the cooler component ($`T0.7`$ keV located in r
<re
<𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e}) requires an absorption column $`N_H=2.9\times 10^{21}`$ cm<sup>-2</sup> in excess of the Galactic value. However, the influence of cold gas having this column density on the hot gas density plotted in Figure 1 is small. For a worst case example, suppose that absorbing material with column density $`N_H=2.9\times 10^{21}`$ cm<sup>-2</sup> is in a disk oriented perpendicular to the line of sight and that this disk absorbs all X-rays from the back side of the galaxy. The radius of this disk would be $`<370`$ pc if it contained the maximum mass $`M_{cg}10^7`$ $`M_{}`$ allowed by CO and HI observations (Bregman, Roberts & Giovanelli 1988; Braine, Henkel & Wiklind 1997) or 20 kpc if it contained all of the gas that has cooled, $`M_{cg}=3\times 10^{10}`$ $`M_{}`$. The X-ray surface brightness within the opaque disk would be reduced by 2, but the corresponding gas density would be lowered by only $`2^{1/2}=10^{0.15}`$ since the volume emissivity $`n^2`$. Such a small correction in the density (gradient) could not account for the large mass discrepancy between the X-ray mass and the stellar dynamical mass shown in Figure 1 within 0.1$`r_e`$.
The densities and temperatures in Figure 1 were determined from X-ray data assuming the abundance of the hot gas is uniformly solar. Since the gas is likely to be more metal rich at smaller galactic radii (Matsushita 1997; Brighenti & Mathews 1999b), an allowance for this gradient would tend to lower the derived density gradient and the internal mass, increasing the discrepancy in r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1~{}r_{e} in Figure 1 by a small amount. In the following discussion we shall ignore the relatively small possible influence of absorption or metallicity gradients on the results shown in Figure 1.
## 3 HYDRODYNAMICAL MODELS
The hydrodynamical models we use in this paper are similar to those in our recent papers (e.g. Brighenti & Mathews 1999a) so we provide only a brief review here. Hot interstellar gas in ellipticals has a dual origin: (i) mass loss from an evolving old stellar population and (ii) secondary infall into the overdensity perturbation that formed the galaxy group within which the elliptical formed by early merging events. For a given set of cosmological parameters, dark and baryonic matter flow toward an overdensity region. The dark matter forms an NFW halo, growing in size with time. Spherical geometry is assumed. Within the accretion shock at time $`t_{}=2`$ Gyr, when enough baryons have accumulated, we form the current de Vaucouleurs stellar profile and release the energy of all Type II supernovae according to a Salpeter IMF (slope: $`x=1.35`$, mass limits: $`m_{\mathrm{}}=0.08`$ and $`m_u=100`$ $`M_{}`$). All stars greater than 8 $`M_{}`$ produce Type II supernovae each of energy $`E_{sn}=10^{51}`$ ergs. We assume that a fraction $`ϵ_{sn}=0.8`$ of this energy is delivered to the internal energy of the gas. We have shown (Brighenti & Mathews 1999b) that such a galaxy formation scheme can work in a variety of cosmologies: flat or low density, with or without a lambda term. The evolution of gas within the optical effective radius $`r_e`$, of most interest here, is insensitive to these cosmological parameters. For simplicity therefore we assume a simple flat universe with $`\mathrm{\Omega }=1`$, $`H_o=50`$ km s<sup>-1</sup> Mpc<sup>-1</sup> and $`\mathrm{\Omega }_b=0.05`$. We characterize the dark halo with an NFW profile (Navarro, Frenk & White 1996) having a virial mass $`M_h=4\times 10^{13}`$ $`M_{}`$ at the current time $`t_n=13`$ Gyrs. The models we discuss here are identical to the standard model of (Brighenti & Mathews 1999a) except we now use a finer central spatial zoning (65 pc for the innermost zone), a SNII efficiency $`ϵ_{sn}=0.8`$, and a mass “dropout” function $`q(r)`$ with more adjustable parameters (see below).
Our objective is to seek solutions of the gasdynamical equations including mass dropout that jointly satisfy several observational constraints at time $`t_n`$: (i) the observed hot gas density, temperature and X-ray surface brightness profiles, (ii) the known dynamical mass in the galactic center usually attributed to a massive black hole, (iii) the apparent dynamical mass to light ratio $`M/L_B=9.2`$ determined in r
<0.4re
<𝑟0.4subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.4r_{e}, and (iv) an apparent internal mass $`M_x(r)`$ in $`(0.11)r_e`$ based on Equation (1) that agrees with the constant-$`(M/L_B)`$ de Vaucouleurs profile as shown in Figure 1.
The baryonic component in our models has a complex evolution. Much of the initial baryonic mass is consumed in creating the stellar system. When the Type II supernova energy is released, a significant mass of gas is expelled as a galactic wind. After these early events, the interstellar gas is re-established and sustained by stellar mass loss and by inflow of circumgalactic gas (secondary infall), most of which was previously enriched and expelled by SNII. We assume that the stars form during a short epoch that can be described by a single burst Salpeter IMF as discussed above. The stellar mass loss rate for this IMF varies as $`\dot{M_t}=\alpha _{}(t)M_t`$ where $`\alpha _{}(t)=4.7\times 10^{20}[t/(t_nt_s)]^{1.3}`$ sec<sup>-1</sup>. Although galactic stars form at $`t_s=1`$ Gyr, their mass loss contribution to the ambient interstellar gas is assumed to begin at a later time, $`t_{}=2`$ Gyrs. Since galactic stars have been enriched by supernova ejecta, the single burst model cannot be strictly correct, but our approximation $`\alpha _{}(t<t_{})=0`$ is consistent with several early starbursts closely spaced in time and allows for metal enrichment of old galactic stars that are not in the first single-burst population. If the de Vaucouleurs profile is a result of largely dissipationless merging, some or most of the star formation must have occurred at a time $`t_s`$ before the important merging events at $`t_{}=2`$ Gyrs. By taking $`t_s<t_{}`$ we reduce by $`10^{10}`$ $`M_{}`$ the total amount of gas ejected by stars within the galactic potential. We recognize the inconsistencies in these approximations of complex stellar formation and dynamical processes that are poorly understood. However, once the galaxy is formed, we follow the interstellar gas dynamical evolution in full detail, conserving mass and energy.
Continued heating by Type Ia supernova is assumed to vary inversely with time, SNu$`(t)=`$ SNu$`(t_n)`$$`(t_n/t)`$, where the current rate, SNu$`(t_n)=0.03`$ SNIa per 100 yrs per $`10^{10}`$ $`L_B`$, is near the lower limit of observed values, SNu$`(t_n)=0.06\pm 0.03(H/50)^2`$ (Cappellaro et al. 1997), as required to maintain the low interstellar iron abundance.
For the models discussed here the equation of continuity includes a “mass dropout” term:
$$\frac{\rho }{t}+\frac{1}{r^2}\frac{}{r}\left(r^2\rho u\right)=\alpha \rho _{}q(r)\frac{\rho }{t_{do}},$$
where $`t_{do}=5m_pkT/2\mu \rho \mathrm{\Lambda }`$ is the time for gas to cool locally by radiative losses at constant pressure (see, e. g. Sarazin & Ashe 1989). The cooling is assumed to be instantaneous without advection in the cooling flow, i.e. $`t_{do}t_{flow}=r/v`$, although in practice this inequality may not always be satisfied. While this type of cooling dropout has been widely used in past models, there is no adequate physical model for mass dropout. Clearly, the gas must cool somewhere – the emission of X-rays indicates a large net energy (and mass) loss from the interstellar medium.
When small regions of low entropy cool, the pressure remains constant since the sound crossing time is much less than the flow time. Following Fabian, Nulsen & Canizares (1982) and Ferland, Fabian, & Johnstone (1994), we assume that cooled gas converts to a second population of optically dark, low mass stars. Regarding H$`\alpha `$ emission as a tracer for the cooling gas, Mathews & Brighenti (1999) have shown that the cooling occurs at a multitude ($`10^6`$) of cooling sites distributed throughout the inner galaxy and that only stars with mass
<12M
<absent12subscript𝑀direct-product\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1-2M_{\odot} can form at each cooling site. This is supported by the observed absence of young massive stars and SNII in elliptical galaxies. Nevertheless, in the following discussion we assume that the maximum stellar mass in the dropout population is sufficiently low that the optical light from these stars is unobservable. We expect cooling and low mass star formation to be concentrated within at least 2 kpc, which is the observed extent of H$`\alpha `$ emission in NGC 4472 (Macchetto et al. 1996). While the region containing optical emission lines provides a natural guideline for selecting the dropout profile, we consider a wider range of constant or variable dropout coefficients $`q(r)`$ parameterized by
$$q(r)=q_o\mathrm{exp}(r/r_{do})^m$$
(2)
which concentrates the cooling within radius $`r_{do}`$. Note that even when $`q`$ is constant, the mass dropout term is spatially concentrated, $`\rho /t_{do}\rho ^2`$.
## 4 MODELS WITH ZERO OR CONSTANT q
In a series of recent papers in this Journal we have presented evolutionary cooling flow models for NGC 4472 that agree quite well with the observed distributions of interstellar gas density, temperature and metallicity (Brighenti & Mathews 1999a; 1999b), particularly at intermediate and large radii. Since the total mass of cooled gas in these calculations was only a few percent of the stellar mass, the gravitational contribution of cooled gas was ignored, even in models including mass dropout. For a better understanding of the inner galaxy, we now include the gravity of all gas, both hot and cold, except when specifically noted. In this section we begin with models having $`q=0`$ everywhere, so that cooling to low temperatures occurs in a small central region, then we investigate models in which $`q`$ is constant throughout the cooling flow.
### 4.1 Models Without Distributed Mass Dropout ($`q=0`$)
We begin by considering a perfectly homogeneous interstellar flow in which all the gas reaches the central computational zone ($`r=65`$ pc) or its neighboring zones where it then cools to $`T10^7`$ K. For comparison we discuss two cases: in Model 1 we ignore the gravity of this cooled gas; in Model 2 (and subsequent models) we include its gravitational influence on the flow. The mass of gas that cools into the central gridzone is dynamically equivalent to a massive black hole, but we do not necessarily regard our calculation as a realistic model for black hole formation. In Figure 2a we illustrate the radial variation of gas density and temperature for Models 1 and 2 after the interstellar gas has evolved to $`t=t_n=13`$ Gyrs. Several global parameters for these and subsequent models are listed in Table 1.
When the gravity of the cooled gas is considered (Model 2), the interstellar gas within a few kpc of the galactic center is compressed and sharply heated in the local potential. Such hot central thermal cores are not generally observed (but see Colbert & Mushotzky 1998). The central mass in both models, $`M_{cent}=4.65\times 10^{10}`$ $`M_{}`$ is about 16 times larger than the black hole mass found in NGC 4472, $`M_{bh}=2.9\times 10^9`$ $`M_{}`$ (Magorrian et al. 1998). For these reasons neither Model 1 nor 2 provides a realistic interstellar mass distribution for this galaxy.
ROSAT band X-ray surface brightness profiles $`\mathrm{\Sigma }_x(R)`$ for Models 1 and 2 are shown in Figure 2b and the total ROSAT band luminosities are listed in Table 1. For Model 1 the X-ray brightness peaks strongly in the galactic center, diverging from the observations within about 3 kpc; it was just this sort of disagreement that initially led to the assumption of distributed mass dropout in cooling flows (Thomas 1986; Vedder, Trester, & Canizares 1988; Sarazin & Ashe 1989). If the efficiency of heating by early SNII were lowered, Models 1 and 2 could be made to agree better with $`\mathrm{\Sigma }_x`$ observations in r
>10
>𝑟10r\mathrel{\mathchoice{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}}10 kpc, but the computed $`\mathrm{\Sigma }_x`$ would rise even further above the observations within a few kpc. The X-ray luminosities $`L_x`$ for Models 1 and 2 listed in Table 1 are unreliable in part because of numerical inaccuracies caused by the extremely steep variation of gas temperature and density in the central 2 or 3 computational zones. This numerical difficulty is common to all cooling flows that proceed to the very center of the galaxy before cooling, for any reasonable central grid spacing. Since more $`Pdv`$ work is done on the flow in Model 2 (where the gravity of cooled gas is included), we expect that $`L_x`$ should also be larger than for Model 1. The opposite sense of the change in $`L_x`$ shown in Table 1 evidently results from numerical inaccuracies near the central singularity where zone to zone variations are no longer linear. If the gas has not cooled before flowing into the core (
<100
<absent100\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}100 pc), as in Models 1 and 2, a significant fraction of the total X-ray emission should come from this very central region, again in disagreement with observations.
In Models 1 and 2 a sphere of cold ($`T=10`$ K), dense gas accumulates in the central gridzones and grows in mass and size over the Hubble time. This sphere is an unrealistic artifact of our computational assumptions. Therefore, to explore further the central numerical difficulty with $`L_x`$ encountered in Models 1 and 2, we considered two additional models using higher spatial resolution (radius of central zone is only 15 pc), both of which include the gravitational influence of cooled gas. The first model (Model 2.1) is similar to Model 2 with $`q=0`$ in all zones. In the second model (Model 2.2) we set $`q=0`$ in all zones except the central zone where $`q=4`$. Model 2.2 is the appropriate limit of the series of models discussed below in which the cooling dropout is more and more centrally concentrated.
While the flow in r
>1
>𝑟1r\mathrel{\mathchoice{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}}1 kpc is very similar in Models 2.1 and 2.2, the behaviour at smaller radii is quite different. For example the flow parameters for Model 2.1 at 100 pc and $`t_n=13`$ Gyrs are: $`T=7.9\times 10^7`$ K, $`n=1`$ cm<sup>-3</sup>, and $`u=320`$ km s<sup>-1</sup>. At the same radius and time for Model 2.2 the flow is quite different: $`T=6.3\times 10^7`$ K, $`n=5`$ cm<sup>-3</sup>, and $`u=60`$ km s<sup>-1</sup>. At projected radius $`R=100`$ pc, the ROSAT X-ray surface brightness in Model 2.2 is 20 times larger than Model 2.1 and the total ROSAT X-ray luminosity integrated over the entire cooling flow in Model 2.2 is larger by a factor 32. The differences between these two models result from upstream propagation of information about the central boundary conditions made possible by subsonic flow near the origin. Fortunately, these numerical and physical difficulties near the origin do not arise in more realistic cooling flows discussed below in which $`q>0`$ at larger galactic radii.
### 4.2 True and Apparent Gas Density and Temperature
While solutions of the gasdynamical equations provide the temperature as a function of physical radius $`T(r)`$, the observed temperature $`T(R)`$ is an emission-weighted mean temperature along the line of sight at projected radius $`R`$. For symmetric galaxies and at small galactic radii, these two temperatures are nearly identical because of the steep radial variation in X-ray emissivity. The variation of temperature with physical radius $`T(r,t_n)`$ in the background flow is shown with light solid lines in Figure 2a.
In the presence of spatially distributed cooling dropout, the local temperature is an emission-weighted mean of the background (uncooled) gas and the cooling regions. Cooling is assumed to occur at a large number of cooling sites where the gas remains in pressure equilibrium as it cools, apparently unrestricted by magnetic stresses (see Mathews & Brighenti 1999 for details). The heavy solid lines in Figure 2a show the mean apparent temperature $`T_{eff}(r,t_n)`$ including contributions from locally cooling regions,
$$T_{eff}=\frac{T+qT\mathrm{\Delta }_1(T)}{1+q\mathrm{\Delta }_0(T)}$$
(3)
where $`T`$ is the background flow temperature and the slowly-varying functions $`\mathrm{\Delta }_i(T)`$ are plotted in Brighenti & Mathews (1998). Note that the effective temperature is independent of the local gas density. The temperature that is actually observed is an average of $`T_{eff}`$ along the line of sight; this temperature $`T_{eff}(R,t_n)`$ is shown with dotted lines in Figure 2a.
Similarly, the apparent hot gas density is increased because of additional emission from denser, cooling-out gas. The observed electron density shown in the Figure 2a is found by Abel inversion of the X-ray surface brightness distribution. When cooling sites are present, the local emissivity into the ROSAT energy band is
$$\epsilon _{\mathrm{\Delta }E}=(\rho _{eff}/m_p)^2\mathrm{\Lambda }_{\mathrm{\Delta }E}(T)$$
$$=(\rho /m_p)^2\mathrm{\Lambda }_{\mathrm{\Delta }E}(T)[1+q\mathrm{\Delta }_0(T)]\mathrm{ergs}/\mathrm{sec}\mathrm{cm}^3.$$
The effective density is therefore
$$n_{eff}=n[1+q\mathrm{\Delta }_0(T)]^{1/2}$$
(4)
where $`n`$ is the electron density of the background, uncooled gas. The observed (azimuthally-averaged) densities in NGC 4472 plotted in Figure 2a should be compared with $`n_{eff}(r,t_n)`$ shown with heavy solid lines.
### 4.3 Distributed Mass Dropout with Constant $`q`$
Lacking an acceptable physical model for spatially distributed mass dropout in galactic cooling flows, we are at liberty to choose any variation for the dropout coefficient $`q(r)`$, appropriately constrained a posteriori by the known dynamical mass from stellar velocities, the X-ray mass, the central black hole mass and the observed radial variation of hot gas density, temperature and X-ray surface brightness. It is natural to begin with constant $`q`$ solutions similar to those considered by Sarazin & Ashe (1989) in their steady state cooling flow solutions. For simplicity we assume that the mass of cooled gas remains at the cooling site where it contributes to the gravitational potential.
In the central panels of Figure 2a we illustrate the hot interstellar gas density and temperature that results after $`t_n=13`$ Gyrs assuming uniform $`q=1`$ and 4; these are listed in Table 1 as Models 3 and 4 respectively. When $`q`$ is a constant independent of radius, the ratios of true to apparent values – $`n/n_{eff}`$, $`T/T_{eff}`$ and $`M_{tot}/M_x`$ – are also approximately uniform with galactic radius. Here $`M_{tot}(r)`$ is the true total mass within $`r`$ and $`M_x(r)`$ is the value that would be determined from X-ray observations of the models by assuming hydrostatic equilibrium (as in Fig. 1). The influence of constant-$`q`$ mass dropout is similar at all galactic radii since the factors that convert background temperature and density to effective values in equations (3) and (4) depend only on $`T(r)`$ which is slowly varying, not on $`n(r)`$. Because of the enormous volume and time available to gas dropping out at large $`r`$, the total mass of cooled gas (listed in Table 1) becomes very large in the outer galaxy. In Figure 3 we compare the radial distribution of stellar and dark halo mass with the distributed dropout mass that results after 13 Gyrs with $`q=1`$ and 4. The de Vaucouleurs “stellar” mass profile in Figure 3 is constructed assuming uniform $`M/L_B=9.2`$.
As $`q`$ increases from 1 to 4, the background hot gas density (light solid lines in Fig. 2a) decreases and its radial gradient flattens, causing a rise in temperature to provide enough pressure to support the cooling flow atmosphere. Note that the total dropped-out mass within $`r_e/3`$ decreases with larger $`q`$ (Figure 3 and Table 1). Although more overall cooling dropout occurs as $`q`$ increases, most of this cooling occurs at very large $`r`$ and, somewhat paradoxically, less gas remains for dropout closer to the galactic center, r
<re
<𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e}. However, the effective (i.e. apparent) density (heavy solid lines in Fig. 2a) is less sensitive to $`q`$. The $`q=1`$ solution (Model 3) is preferred for its fit to observed temperatures while the $`q=4`$ solution (Model 4) agrees better with the observed density in r
<10
<𝑟10r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}10 kpc and almost exactly with the X-ray surface brightness (Fig. 2b).
In addition, the centrally concentrated dropped out mass in the $`q=1`$ solution (Model 3) contributes a larger fraction of the dynamical mass in r
<re/3
<𝑟subscript𝑟𝑒3r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e}/3 and the total mass within the central gridzone ($`r=65`$ pc) is almost equal to the known mass of the black hole in NGC 4472 (Table 1). Evidently, cooled masses in models with $`q<1`$ would exceed the central dark mass observed.
For both values of $`q`$ considered, the dropped out mass listed in Table 1 contributes appreciably to the total mass within $`r=r_e/3`$. When both old stellar and dropout mass are included, the $`M/L_B`$ values at $`r_e/3`$ shown in Table 1 – 12.19 and 11.21 – exceed both the dynamical value $`M/L_B=9.2`$ found by van der Marel (1991) within $`0.4r_e`$ and the value $`M/L_B(r_e/3)=10.18`$ in our models which includes a small additional contribution of non-baryonic dark matter. To quantify the influence of the dark baryonic dropout mass on the mass to light ratio we include in Table 1 an entry for
$$\mathrm{\Delta }_{m/l}(r_e/3)=\frac{(M/L_B)_{,dh,do}(r_e/3)}{(M/L_B)_{,dh}(r_e/3)}1$$
$$=\frac{M_{,dh,do}(r_e/3)}{M_{,dh}(r_e/3)}1$$
where $`M_{,dh}(r_e/3)=1.35\times 10^{11}`$ $`M_{}`$ is the combined mass of luminous stars and dark halo matter at $`r_e/3`$ and $`M_{,dh,do}(r_e/3)=M_{tot}(r_e/3)`$ also includes the dropout mass within this radius. Since our computed total mass $`M_{,dh,do}(r_e/3)`$ exceeds the observed dynamical value 9.2 (and also 10.18), to be fully consistent we should have chosen $`M/L_{B}^{}{}_{}{}^{}<9.2`$ for the old stellar population (see below), provided $`M/L_B`$ for stars produced from the cooled gas is infinite as we have assumed.
If these constant $`q`$ models are physically appropriate, the agreement between $`M_x(r)`$ and $`M_{}(r)`$ in $`(0.11)r_e`$, as shown in Figure 1 (and for NGC 4649), is surprising since $`M_x(r_e/3)`$ and $`M_{,dh}(r_e/3)`$ differ by 10 - 40 percent (Table 1) for the constant $`q`$ models. This suggests that the mass to light ratio of the dropout stellar population is not infinite as we have assumed, but comparable with that of the old stellar population.
## 5 MODELS WITH CENTRALLY CONCENTRATED DROPOUT
We now seek evolutionary gasdynamical solutions with variable dropout coefficients $`q(r)`$ that strongly concentrate the mass dropout in the inner galaxy, r
<re
<𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e}. The limited spatial extent of optical emission lines in the cores of bright ellipticals suggests that cooling dropout occurs well within $`r_e`$. For example, the H$`\alpha `$ \+ \[NII\] image of NGC 4472 is observed out to approximately $`0.25r_e`$ or 2 kpc (Macchetto et al. 1996). As hot interstellar gas cools in this region, its temperature pauses at $`T10^4`$ K where the gas is heated and ionized by stellar UV radiation. Therefore, we consider parameters $`r_{do}`$ and $`m`$ in equation (2) that concentrate the mass dropout in the central region, but not at the very center as in Model 2.
The dropout parameters $`q_o`$, $`r_{do}`$ and $`m`$, are listed in Table 1 for Models 5, 6 and 7. The mass dropout in these three models is progressively more concentrated toward the galactic center. The results for Model 5, for which the dropout scale length is very large, $`r_{do}=15`$ kpc, are similar in most respects to those of Model 4 except the massive dropout in the outer galaxy is no longer present. In particular, the mass to light ratio at $`r_e/3`$ in Table 1 is very similar for Models 4 and 5.
The current interstellar density and temperature variations for Model 6 with $`r_{do}=2`$ kpc are shown in Figure 2a and $`\mathrm{\Sigma }_x(R)`$ is plotted in Figure 2b. The apparent density and $`\mathrm{\Sigma }_x`$ (heavy solid lines) are considerably greater than observed values in r
<3
<𝑟3r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}3 kpc, although the projected apparent temperature is a reasonable fit to the NGC 4472 data. The radial distribution of dropout mass for Model 6 is shown in Figure 3. From Table 1 the total (old stars, halo and dropout) mass to light ratio at $`r_e/3`$ is $`M_{tot}/L_B=13.03`$, 29 percent higher than van der Marel’s value and 22 percent greater than the corresponding value for our background model galaxy.
Also shown in Figures 2a and 2b are similar results for Model 7 in which the mass dropout, now approximated with a Gaussian, is concentrated within $`r_{do}=800`$ pc. Model 7 is constructed so that most of the mass dropout occurs in r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e}, corresponding to the region of apparent disagreement between $`M_x`$ and $`M_{}`$ in Figure 1. It is of interest that the gas density and $`\mathrm{\Sigma }_x`$ within 1 kpc for Model 7, where the dropout is greatest, exceeds those of Model 2 in which there is no dropout at all. This can be understood from the flow velocity distribution. For Model 2 without distributed dropout the inward moving gas velocity increases through the entire region illustrated and shocks at $`r1`$ kpc. In the presence of dropout, the flow velocity at corresponding radii in Model 7 is much slower and reaches a maximum (negative) value near $`r=1`$ kpc then approaches zero subsonically at the origin. The density and $`\mathrm{\Sigma }_x`$ enhancements in the background flow for Model 7 at r
<3
<𝑟3r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}3 kpc are due to a local compression as the gas flows into slowly moving gas in the core. Within about 3 kpc, both the background and apparent densities exceed the observations by a larger factor than those of Model 6 and $`\mathrm{\Sigma }_x`$ also peaks unrealistically in this same region (Fig. 2b). The dropout mass for Model 7 shown in Figure 3 equals that of the old population stellar mass at $`r1`$ kpc ($`0.12r_e`$). For these reasons Model 7 seems less satisfactory than Model 6, but neither is as generally successful as Model 3 ($`q=1`$).
Although increased mass dropout can decrease the X-ray surface brightness $`\mathrm{\Sigma }_x`$, as when uniform $`q`$ increased from 1 to 4 (Models 3 and 4 in Fig. 2b), this is not always the case. When the dropout is concentrated more toward the galactic center, $`\mathrm{\Sigma }_x`$ actually increases, as in the transition from Model 6 to 7.
It is interesting to determine the influence of distributed dropout on the total apparent mass $`M_x(r)`$ found from the model by assuming hydrostatic equilibrium (equation 1). Due to the contribution of low temperature cooling regions to the total X-ray emission, $`T(r)`$ and therefore $`M_x(r)`$ is always lower than the true mass $`M_{,dh}(r)`$ in cooling dropout regions. A difference in this sense is apparent in Table 1 at $`r=r_e/3`$ for Models 3 - 7; this is similar to the mass discrepancy in Figure 1 at r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e}.
The high central apparent gas density and surface brightness for Model 6 shown in Figures 2a and 2b are obvious problems for this model. However, the gas density can be reduced if a strong magnetic field or other non-thermal energy density is present in r
<0.25re=2
<𝑟0.25subscript𝑟𝑒2r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.25r_{e}=2 kpc. Dynamically important magnetic fields may also be required to fit the X-ray data of NGC 4636 (Brighenti & Mathews 1997a) and may be generally expected in luminous ellipticals (Mathews & Brighenti 1997; Godon, Soker & White 1998). Additional non-thermal pressure support is also implied for Model 6 since the effect of central cooling dropout fails to lower the apparent mass $`M_x(r)`$ below the actual mass $`M_{tot}`$ as much as the observed deviation shown in Figure 1. Like many bright ellipticals, NGC 4472 has a weak non-thermal radio source within the central $`4`$ kpc, indicating $`B10100`$ $`\mu `$G (Ekers & Kolanyi 1978).
For all calculated models in Table 1 with distributed dropout, the total mass $`M_{tot}=M_{,dh,do}`$ significantly exceeds the mass of the old stellar population plus dark halo $`M_{,dh}`$ throughout the inner galaxy, indicating that dropout material makes an important additional contribution to the total mass. However, the apparent mass $`M_x`$ determined from equation (1) is less than $`M_{tot}`$ in the inner galaxy and, for Model 6, can also be less than $`M_{}`$. Of particular interest is the region 0.1re
<r
<re
<0.1subscript𝑟𝑒𝑟
<subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e} (i.e., 0.07
<logrkpc
<0.93
<0.07subscript𝑟𝑘𝑝𝑐
<0.93-0.07\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}\log r_{kpc}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.93) in Figure 1. Although the agreement in Figure 1 is excellent in this region, for Models 6 and 7 the total mass is larger than $`M_{,dh}(r)`$ and values of $`\mathrm{\Delta }_{m/l}`$ in Table 1 suggest that the dropped out mass contributes 25 - 35 percent of the total mass in this region. Therefore, if the dropout mass is optically dark, the true stellar mass to light ratio of the old stellar population must be $`M/L_B6`$ rather than the value $`M/L_B=9.2`$ found by van der Marel which includes the dropout mass.
To investigate such a possibly more self-consistent old stellar component, we consider Model 8 based on the same $`q(r)`$ used in Model 6, but with $`M/L_B=6`$ for the old stellar population. For additional consistency in Model 8, $`\alpha _{}(t)`$ is increased by the ratio of assumed stellar mass to light ratios 9.2/6 = 1.53 as described below; the total stellar mass ejected is identical to that in Model 6. Figure 3 illustrates the dropout mass profile for Model 8. The apparent density, temperature and $`\mathrm{\Sigma }_x`$ profiles for Model 8 shown in Figures 4a and 4b are very similar to those of Model 6, so this adjustment of the stellar $`M/L_B`$ has had little effect.
The radial mass profiles of Models 6 and 8 are compared in Figure 5. In this plot the open circles show the observed X-ray mass $`M_x(r)`$ for NGC 4472 using equation (1) and the solid lines shows $`M_x(r)`$ based on equation (1) using $`n(r)`$ and $`T(r)`$ from the models. The dashed lines show $`M_{tot}(r)`$ and the dotted lines are the stellar mass $`M_{}(r)`$ based on a de Vaucouleurs profile with $`M/L_B=9.2`$ in the upper panel (Model 6) and $`M/L_B=6`$ in the lower panel (Model 8). The superiority of Model 8 is evident from the closer agreement between the X-ray mass data points for NGC 4472 and the solid line for that model. This agreement for Model 8 would be even closer in the range $`\mathrm{log}r0.51`$ if we had used a dark halo mass profile less centrally peaked than NFW. Model 8 may provide the most self-consistent overall fit to the mass constraints for NGC 4472; if so, the old stars in NGC 4472 have a mass to light ratio $`M/L_B6`$, about $`30`$ percent lower than the mass to light ratio determined from stellar dynamics.
In summary, none of our models is fully satisfactory in every respect. The total mass to light ratio (including dropout mass) in r
<re/3
<𝑟subscript𝑟𝑒3r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e}/3 is 10 - 35 percent higher than the value for the underlying galaxy. However, in Model 8 in which $`M/L_B=6`$ for the old stars, the difference between the X-ray mass determined from the models and NGC 4472 is appreciably reduced. Nevertheless, the central apparent gas density and X-ray surface brightness in Model 8 are still larger than observed, requiring additional non-thermal support. There is no independent theoretical justification for the dropout profile $`q(r)`$ assumed in Models 6 and 8; as explained earlier, the dropout distribution depends on unknown interstellar entropy fluctuations. Nevertheless, for all models considered here the mass of cooled interstellar gas contributes significantly to the total mass and the dynamically determined mass to light ratio within the inner galaxy.
## 6 FINAL REMARKS AND CONCLUSIONS
In this series of calculations we have taken a census of all baryons involved in the evolution of a large elliptical galaxy: the original stellar component, the interstellar medium, and – of most interest – the small but troublesome mass of hot gas that cools over cosmic time. We have shown that the radial distribution of cooled interstellar gas influences dynamical and X-ray determinations of the total interior mass and the radial profiles of apparent density, temperature and X-ray brightness of the hot gas. In our models, cooled gas is slowly deposited in the central galaxy r
<re
<𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e} (see Figure 3) as indicated by the extent of observed H$`\alpha `$ emission. Since there is little or no direct observational evidence for the mass that has dropped out in ellipticals like NGC 4472, the cooled mass must either be dark at optical and radio frequencies or indistinguishable from the old stars. Low mass stars are an obvious and physically reasonable endstate for the cooled gas (Ferland, Fabian & Johnstone 1994; Mathews & Brighenti 1999). We also suppose that the cooled gas remains at the dropout site where it contributes to the galactic potential.
### 6.1 Reducing Stellar Mass Loss
In an attempt to reduce the influence of cooling and cooled gas on the models, we have altered many of the model parameters. The chosen cosmology ($`\mathrm{\Omega }=1`$; $`\mathrm{\Omega }=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, etc.) or the baryon fraction $`\mathrm{\Omega }_b`$ have little influence on the total dropout mass. Changing the times when the stars and the galactic potential form ($`t_s`$ and $`t_{}`$) or the spatial scale of the release of SNII energy within reasonable limits have only a modest influence on the dropout mass that accumulates by time $`t_n=13`$ Gyrs. Increasing the interval $`t_st_{}`$ between star and galaxy formation does reduce the total dropout mass, but this interval cannot be too large since luminous ellipticals are observed at large redshifts. The cooling dropout would not be dramatically reduced if massive ellipticals were all only a few Gyrs old since most of the stellar mass loss occurs just after star formation; however, many or most luminous ellipticals are thought to be very old.
Perhaps the most effective way to preserve the excellent agreement between $`M_{}`$ and $`M_x`$ in Figure 1 in 0.1re
<r
<1re
<0.1subscript𝑟𝑒𝑟
<1subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1r_{e}, without constraining the mass dropout profile $`q(r)`$, would be to reduce the total mass that has cooled over cosmic time. The most sensitive parameter influencing the cooled mass is the specific rate of stellar mass loss, $`\alpha _{}(t)`$. Although the total mass of hot gas increases with time due to the continued influx of secondary infalling gas, the mass of hot gas within the optical galaxy originates mostly from stellar mass loss and the X-ray luminosity there scales as $`L_x\alpha _{}(t)t^{1.3}`$ (Appendix B of Tsai & Mathews 1995). Computed models similar to those described here but with arbitrarily reduced $`\alpha _{}(t)`$ fit the $`n(r)`$ and $`\mathrm{\Sigma }_x(R)`$ data rather well at time $`t_n`$ and produce much less cooling dropout. However, $`\alpha _{}(t)`$ cannot be lowered without also increasing the stellar mass. For all reasonable power law initial mass functions, $`\alpha _{}(t)`$ varies inversely with the stellar mass to light ratio:
$$\alpha _{}(t)\frac{dM_{}/dt}{M_{}}=\left[\frac{dM_{}/dt}{L_B}\right]\frac{1}{(M_{}/L_B)}\frac{1}{(M_{}/L_B)}.$$
The physical explanation for the constancy of $`(dM_{}/dt)/L_B`$ is that both $`L_B`$ (dominated by post-main sequence stars) and $`dM_{}/dt`$ depend on the instantaneous rate that stars leave the main sequence so this IMF-dependent factor cancels out (e.g. Renzini & Buzzoni 1986). To illustrate this result, we use the Renzini-Buzzoni procedure and plot in Figure 6 the relationship between $`\alpha _{}`$ and $`M/L_B`$ at time $`t_n=13`$ Gyrs for 84 power law IMFs with slopes $`x=`$0.6, 0.8, 1.0, 1.2, 1.4, 1.6, and 1.8, each with lower and upper mass limits of $`m_{\mathrm{}}=`$0.01, 0.032, 0.1, 0.32 and $`m_u=`$10, 32, 100. The remarkable linearity in Figure 6 shows that the quantity in square brackets in the equation above is almost invariant to large changes in the slope or mass limits for power law IMFs.
Therefore, if $`\alpha _{}(t)`$ is reduced by 2 or 3 in an attempt to reduce the total dropout mass, the stellar mass (and $`M/L_B`$) must be increased by the same factor; as a result the total mass ejected, $`\alpha _{}M_{}𝑑t`$, and the total dropout mass do not change.
### 6.2 Effect of Galactic Rotation
Although massive ellipticals are not rotationally flattened, they do rotate significantly, e.g. $`(v/\sigma )_{}=0.43`$ for NGC 4472 (Faber et al. 1997). If the hot interstellar gas rotates in the same sense as the bulk of mass-losing stars, stars formed from the cooled gas should form into a disk of scale $`r_e`$ (Brighenti & Mathews 1997b), although the development of such disks is likely to be suppressed by the mass dropout process. Nevertheless, to the extent that the cooled gas has a disk-like distribution, its global influence on the stellar dynamics in r
<0.4re
<𝑟0.4subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.4r_{e} would be less than if the same dark mass were distributed spherically. Remarkably, there is no observational evidence at present for rotational flattening in the X-ray images of giant ellipticals like NGC 4472.
### 6.3 Non-Baryonic Dark Matter within $`r_e`$
Dynamical determinations of the mass to light ratio from stellar velocities reflect the entire mass within the stellar orbits, including non-baryonic mass. The NFW halo we use in our models for NGC 4472 agrees with X-ray observations in the extended halo but is slightly too massive near $`rr_e`$ relative to the X-ray mass $`M_x(r)`$ observationally determined for NGC 4472. This may indicate that dark halos are less centrally peaked than NFW (see Kravtsov et al. 1998). The mass of our NFW halo model contributes about 10 percent to the dynamical mass to light ratio measured within $`r_e/3`$, but the NFW profile is probably disturbed in this region. When the dominant baryonic mass in r
<re
<𝑟subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r_{e} compressed to form the de Vaucouleurs profile, we expect that the NFW halo was dragged inward and distorted. However, the dark halo cores of the earlier galactic condensations that merged to form the elliptical may have expanded due to starburst driven galactic winds. Because of these various counteracting effects, the small non-baryonic contribution to dynamical mass to light determinations is uncertain.
### 6.4 Contribution of Dropout to “Stellar” $`M/L`$
For all models studied here – based on a wide variety of mass dropout profiles $`q(r)`$ – the mass of cooled interstellar gas contributes substantially to the total mass within $`0.4r_e`$ where the stellar mass to light ratio is determined from stellar velocities. If optically dark low mass stars form from the cooled gas, the “stellar” mass to light ratios in the literature refer to two distinct stellar populations having radically different initial mass functions and spatial distributions. The stellar mass to light ratio of the original, optically luminous stellar population is lower than published values indicate.
A complete solution of this problem requires a better understanding of the physics of star formation and the processes that control the stellar IMF. We have assumed here that a Salpeter IMF provides a satisfactory approximation to the original single-burst star formation at early times. Yet we argue that the younger dropout IMF is strongly skewed toward low mass stars. The IMF has evolved over time. It is possible therefore that the early, more intense mass dropout resulted in a more nearly Salpeter-like IMF, producing a fraction of currently observed luminous stars in ellipticals. If so, this early dropout would not contribute to the excess dropout mass that we find in our models, but would have already been included in the original de Vaucouleurs population. High density, metal enriched stellar cores in elliptical galaxies may have derived from normal-IMF star formation from early, more intense interstellar cooling. This is similar to assumptions made for dissipative galactic core formation from the convergence of gas following major mergers (Mihos & Hernquist 1996). While such notions cannot be entirely dismissed, the approximate universality of the de Vaucouleurs light profile among ellipticals may argue against a dual formation process for the radial distribution of luminous stars: violent relaxation and cooling flow dropout.
Throughout this discussion we have assumed that the stellar population formed from cooling flow dropout is optically dark. Although there is little or no evidence that normal massive OB stars (or SNII) are present in elliptical galaxies, it is possible that younger stars having masses up to $`12`$ $`M_{}`$ are present and that such intermediate mass stars could form from the cooled gas (Mathews & Brighenti 1999). This type of dropout stellar population could contribute to the total optical light. If the mass to light ratio of dropout and old stellar populations are similar, the dropout component could be difficult to detect by the means we have discussed here and its perturbation on the observed $`M/L_B`$ would be greatly lessened. In this case the dropout population would introduce an additional radial light profile that would differ slightly from that of the old stellar population. The dropout mass profiles in Figure 3 indicate that Models 4 and 6 would be rather difficult to detect against the background stellar light. Intermediate mass dropout stars could therefore provide a satisfactory resolution to the problems we have discussed here.
### 6.5 Influence of Dropout on the Fundamental Plane
The ensemble of elliptical galaxies is known to have global parameters that deviate slightly from the assumptions of virial equilibrium and homologous structure. The deviation of this fundamental plane relationship is in the sense that the dynamical mass to light ratio increases with galactic mass, $`M/L_BM^{0.24}`$ (Dressler et al. 1987; Djorgovski & Davis 1987). Such a non-homologous deviation could in principle be produced by the small amount of cooled interstellar gas $`M_{cg}`$ provided it increases appropriately with $`M_{}`$. To test this idea, we performed an identical hydrodynamic calculation for an elliptical galaxy having a mass one fourth that of NGC 4472. The dark halo mass was also reduced by the same factor but the cosmological environment and mass dropout distribution were identical to those used for NGC 4472, scaled to a smaller $`r_e`$. We found that the amount of mass dropout $`M_{cg}`$ is higher in smaller ellipticals relative to the total baryonic mass $`M_{}`$. This is opposite to the trend observed in the fundamental plane. However, if hot gas and dark matter in the outer halos of smaller ellipticals is tidally stripped in group environments, as suggested by Mathews & Brighenti (1998b), then the mass of cooled gas would be reduced and its effect on the fundamental plane would be reduced. Nevertheless, explanations of the deviations of the fundamental plane from virial scaling must recognize the possible additional influence of dropout mass, regardless of the trend of $`M_{cg}/M_{}`$ with $`M_{}`$.
### 6.6 Conclusions
Using simple spherical gas-dynamical models for the evolution of interstellar gas and data from the well-observed elliptical NGC 4472, we reach the following conclusions:
(1) If the hot interstellar gas cools only in the very center of NGC 4472 for $`10`$ Gyrs, the total accumulated mass there would be
>10
>absent10\mathrel{\mathchoice{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}}10 times larger than the mass of the central black hole observed. If such large concentrated masses were generally present in bright ellipticals, interstellar gas in r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e} would be compressed and heated to temperatures $`>1`$ keV. Such hot thermal cores are not generally observed. We conclude that the cooling dropout in massive ellipticals must occur before the gas reaches the galactic center. The hypotheses of distributed mass dropout and low mass star formation were proposed many years ago (Fabian, Nulsen & Canizares 1982; Thomas 1986; Cowie & Binney 1988; Vedder, Trester, & Canizares 1988; Sarazin & Ashe 1989; Ferland, Fabian, & Johnstone 1994). However, these historical arguments were generally based on the notion that mass dropout would help reduce computed X-ray surface brightness profiles at small projected radii, as required by the observations, but we have shown here that in some cases enhanced dropout at small galactic radii can cause $`\mathrm{\Sigma }_x`$ to increase, not decrease. (The total bolometric X-ray luminosity $`L_x`$ should always be lower in distributed cooling models since the hot gas experiences only a fraction of the galactic potential.) The best arguments for distributed cooling dropout are (i) limits on the central black hole mass and (ii) the absence of rotational flattening in X-ray images.
(2) We have considered a wide variety of possible mass profiles for the radial deposition of cooled interstellar gas in NGC 4472. The dropout mass is assumed to be optically dark, consistent with the formation of very low mass stars. In every case the (stellar plus dropout) mass to light ratio at $`r_e/3`$ significantly exceeds the mass to light ratio determined from stellar velocities. If the dropout mass is optically dark, dynamical mass to light ratios in luminous ellipticals should be substantially enhanced by dark baryonic matter. In this case the true $`M/L_B`$ for luminous stars may be $`30`$ percent smaller than published values.
(3) The excellent agreement between the X-ray and “stellar” mass in NGC 4472 shown in Figure 1 (and also NGC 4649) in the range 0.1re
<r
<1re
<0.1subscript𝑟𝑒𝑟
<1subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1r_{e} may be a coincidence if our estimates of the cooling flow dropout mass are correct and if this mass is non-luminous.
(4) Dynamical mass to light determinations within $`r_e`$ refer to a superposition of two stellar populations: an old luminous population with a de Vaucouleurs galactic profile and a younger population having a bottom-heavy IMF and a different galactic mass profile. If the younger population is optically dark, the mass to light ratio is not likely to be constant with galactic radius within $`r_e`$.
(5) It is not possible to reduce the total amount of mass deposited from the cooling flow simply by lowering the specific rate of stellar mass loss $`\alpha _{}(t)(dM_{}/dt)/M_{}`$. For all reasonable power law initial mass functions we show that $`\alpha _{}(M_{}/L_B)^1`$. For given $`L_B`$, larger stellar masses $`M_{}(r)`$ must accompany lower values of $`\alpha _{}`$ so the total amount of mass ejected from stars $`\alpha _{}(t)M_{}𝑑t`$ is nearly independent of the IMF.
(6) Among the models we consider, those with centrally concentrated mass dropout perform best in minimizing the overall disagreement with the central $`M/L_B`$ and the X-ray determined mass in 0.1re
<r
<1re
<0.1subscript𝑟𝑒𝑟
<1subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1r_{e}. Constant $`q`$ models in which the dropout is proportional to the local gas emissivity at every radius deposit less mass in 0.1re
<r
<1re
<0.1subscript𝑟𝑒𝑟
<1subscript𝑟𝑒0.1r_{e}\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1r_{e}, but may have gas temperatures that are too low (q
>4
>𝑞4q\mathrel{\mathchoice{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}{\lower 3.0pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr>\crcr\sim\crcr}}}}4) or central masses that are too large (q
<1
<𝑞1q\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}1).
(7) Even in the presence of mass dropout, the computed central interstellar gas densities and X-ray surface brightnesses $`\mathrm{\Sigma }_x(R)`$ are generally too large. Such deviations would be reduced if the hot gas is partially supported in r
<0.1re
<𝑟0.1subscript𝑟𝑒r\mathrel{\mathchoice{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\displaystyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\textstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}{\lower 2.5pt\vbox{\halign{$\mathsurround 0pt\scriptscriptstyle\hfil#\hfil$\cr<\crcr\sim\crcr}}}}0.1r_{e} by magnetic or other non-thermal pressure associated with the extended radio source.
(8) If the stellar population formed from cooled interstellar gas extends to intermediate masses, $`12`$ $`M_{}`$, its mass to light ratio may blend with that of the older population. In this case the dropout mass would already be represented in the de Vaucouleurs profile representing the stellar mass distribution in our models. Dynamical determinations of $`M/L_B`$ would be a weighted mean of the two populations. If the dropout stellar population is luminous, some of the difficulties we have discussed here would be alleviated, but not those regarding $`\mathrm{\Sigma }_x(R)`$.
(9) If the influence of rotation and non-thermal pressure can be understood, high resolution images of the central regions of elliptical galaxies using the Chandra (AXAF) satellite may detect the presence of dark baryonic dropout material and an accurate determination of the mass to light ratio of the old stellar population.
Thanks to Karl Gebhardt for providing useful information. Studies of the evolution of hot gas in elliptical galaxies at UC Santa Cruz are supported by NASA grant NAG 5-3060 and NSF grant AST-9802994 for which we are very grateful. FB is supported in part by Grant MURST-Cofin 98.
|
warning/0002/hep-ph0002014.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Ever since the discovery of the Bjorken scaling of deep inelastic lepton-proton scattering in 1967, high energy lepton scattering experiments have provided increasingly detailed information on the flavor, momentum, and helicity distributions of the quarks and gluons in hadrons. The results are represented in the form of leading-twist light-cone momentum and helicity distributions $`q(x,\lambda ,Q)`$, $`\overline{q}(x,\lambda ,Q)`$ and $`g(x,\lambda ,Q)`$ at the resolution of $`Q`$. However, such distributions represent single-particle probabilities and thus do not contain information on the transverse momentum, spin, and flavor correlations of the bound quarks and gluons. In particular, structure functions cannot specify the phases needed to understand QCD processes at the amplitude level, the physics which underlies form factors, exclusive and diffractive scattering processes, and the hadronic decay amplitudes of heavy hadrons. The polarized beam and polarized target experiments now in progress and planned at Jefferson Laboratory, HERMES at DESY, BNL, CERN, and SLAC, and measurements of rare exclusive channels and their polarization correlations in $`e^+e^{}`$ and $`\gamma \gamma `$ annihilation at the high luminosity $`B`$ factories promise a new level of precision in testing QCD and determining fundamental properties of hadrons. A global unified interpretation of such inclusive and exclusive experiments is a challenging theoretical problem, mixing issues involving non-perturbative and perturbative dynamics.
Ideally, one wants to have a frame-independent, quantum-mechanical description of hadrons at the amplitude level capable of encoding all possible quark and gluon momentum, helicity, and flavor correlations in the form of universal process-independent hadron wavefunctions for each particle number configuration. Remarkably, the light-cone Fock expansion allows just such a unifying representation. Moreover, the light-cone formalism provides a physical factorization scheme which conveniently separates and factorizes soft non-perturbative physics from hard perturbative dynamics in both exclusive and inclusive reactions.
Formally, the light-cone expansion is constructed by quantizing QCD at fixed light-cone time $`\tau =t+z/c`$ and forming the invariant light-cone Hamiltonian: $`H_{LC}^{QCD}=P^+P^{}P_{}^2`$ where $`P^\pm =P^0\pm P^z.`$ The momentum generators $`P^+`$ and $`P_{}`$ are kinematic; i.e., they are independent of the interactions. The generator $`P^{}=i\frac{d}{d\tau }`$ generates light-cone time translations, and the eigen-spectrum of the Lorentz scalar $`H_{LC}^{QCD}`$ gives the mass spectrum of the color-singlet hadron states in QCD together with their respective light-cone wavefunctions. For example, the proton state satisfies: $`H_{LC}^{QCD}|\psi _p=M_p^2|\psi _p`$. The expansion of the proton eigensolution $`|\psi _p`$ on the color-singlet $`B=1,Q=1`$ eigen states $`\{|n\}`$ of the free Hamiltonian $`H_{LC}^{QCD}(g=0)`$ gives the light-cone Fock expansion: $`|\psi _p(P^+,P_{})>=_n\psi _n(x_i,k_i,\lambda _i)|n;x_iP^+,x_iP_{}+k_i,\lambda _i>`$.
The light-cone momentum fractions $`x_i=k_i^+/P^+`$ with $`_{i=1}^nx_i=1`$ and $`k_i`$ with $`_{i=1}^nk_i=0_{}`$ represent the relative momentum coordinates of the QCD constituents. The physical transverse momenta are $`p_i=x_iP_{}+k_i.`$ The $`\lambda _i`$ label the light-cone spin $`S_z`$ projections of the quarks and gluons along the quantization $`z`$ direction. The physical gluon polarization vectors $`ϵ^\mu (k,\lambda =\pm 1)`$ are specified in light cone gauge $`kϵ=0,\eta ϵ=ϵ^+=0.`$ Light-cone quantization is most conveniently carried out in the physical ghost-free light-cone gauge $`A^+=0;`$ however, light-cone quantization in Feynman gauge also has a number of attractive features, including manifest covariance and a straightforward passage to the Coulomb limit in the case of heavy static quarks.
The solutions of $`H_{LC}^{QCD}|\psi _p=M_p^2|\psi _p`$ are independent of $`P^+`$ and $`P_{}`$; thus given the eigensolution Fock projections $`n;x_i,k_i,\lambda _i|p=\psi _n(x_i,k_i,\lambda _i),`$ the wavefunction of the proton is determined in any frame. In contrast, in equal-time quantization, a Lorentz boost always mixes dynamically with the interactions, so that computing a wavefunction in a new frame requires solving a nonperturbative problem as complicated as the Hamiltonian eigenvalue problem itself.
The LC wavefunctions $`\psi _{n/H}(x_i,\stackrel{}{k}_i,\lambda _i)`$ are universal, process independent, and thus control all hadronic reactions. Given the light-cone wavefunctions, one can compute the moments of the helicity and transversity distributions measurable in polarized deep inelastic experiments. Similarly, the matrix elements of the currents as integrated squares of the LC wavefunctions. For example, the polarized quark distributions at resolution $`\mathrm{\Lambda }`$ correspond to
$`q_{\lambda _q/\mathrm{\Lambda }_p}(x,\mathrm{\Lambda })`$ $`=`$ $`{\displaystyle \underset{n,q_a}{}}{\displaystyle \underset{j=1}{\overset{n}{}}dx_jd^2k_j\underset{\lambda _i}{}|\psi _{n/H}^{(\mathrm{\Lambda })}(x_i,\stackrel{}{k}_i,\lambda _i)|^2}`$
$`\times \delta \left(1{\displaystyle \underset{i}{\overset{n}{}}}x_i\right)\delta ^{(2)}\left({\displaystyle \underset{i}{\overset{n}{}}}\stackrel{}{k}_i\right)\delta (xx_q)\delta _{\lambda _a,\lambda _q}\mathrm{\Theta }(\mathrm{\Lambda }^2_n^2)`$
where the sum is over all quarks $`q_a`$ which match the quantum numbers, light-cone momentum fraction $`x,`$ and helicity of the struck quark. Similarly, moments of transversity distributions and other off-diagonal helicity convolutions are defined as a density matrix of the light-cone wavefunctions. The light-cone wavefunctions also specify the multi-quark and gluon correlations of the hadron. For example, the distribution of spectator particles in the final state which could be measured in the proton fragmentation region in deep inelastic scattering at an electron-proton collider are in principle encoded in the light-cone wavefunctions.
The effective lifetime of each configuration in the laboratory frame is $`2P_{\mathrm{lab}}/`$ $`(_n^2M_p^2)`$ where $`_n^2=_{i=1}^n(k_i^2+m_i^2)/x_i<\mathrm{\Lambda }^2`$ is the off-shell invariant mass and $`\mathrm{\Lambda }`$ is a global ultraviolet regulator. The light-cone momentum integrals are thus be limited by requiring that the invariant mass squared of the constituents of each Fock state is less than the resolution scale $`\mathrm{\Lambda }`$. As I discuss below, this cutoff serves to define a factorization scheme for separating hard and soft regimes in both exclusive and inclusive hard scattering reactions.
The ensemble $`\psi _{n/H}`$ of light-cone Fock wavefunctions is a key concept for hadronic physics, providing the interpolation between physical hadrons (and also nuclei) and their fundamental quark and gluon degrees of freedom. Each Fock state interacts distinctly; e.g. Fock states with small particle number and small impact separation have small color dipole moments and can traverse a nucleus with minimal interactions. This is the basis for the predictions for “color transparency”.
Given the $`\psi _{n/H}^{(\mathrm{\Lambda })},`$ one can construct any spacelike electromagnetic or electroweak form factor or local operator product matrix element from the diagonal overlap of the LC wavefunctions. Similar results hold for the matrix elements which occur in deeply virtual Compton scattering. Exclusive semi-leptonic $`B`$-decay amplitudes such as $`BA\mathrm{}\overline{\nu }`$ can also be evaluated exactly. In this case, the timelike decay matrix elements require the computation of both the diagonal matrix element $`nn`$ where parton number is conserved and the off-diagonal $`n+1n1`$ convolution such that the current operator annihilates a $`q\overline{q^{}}`$ pair in the initial $`B`$ wavefunction. This term is a consequence of the fact that the time-like decay $`q^2=(p_{\mathrm{}}+p_{\overline{\nu }})^2>0`$ requires a positive light-cone momentum fraction $`q^+>0`$. Conversely for space-like currents, one can choose $`q^+=0`$, as in the Drell-Yan-West representation of the space-like electromagnetic form factors. However, as can be seen from the explicit analysis of timelike form factors in a perturbative model, the off-diagonal convolution can yield a nonzero $`q^+/q^+`$ limiting form as $`q^+0`$. This extra term appears specifically in the case of “bad” currents such as $`J^{}`$ in which the coupling to $`q\overline{q}`$ fluctuations in the light-cone wavefunctions are favored. In effect, the $`q^+0`$ limit generates $`\delta (x)`$ contributions as residues of the $`n+1n1`$ contributions. The necessity for such “zero mode” $`\delta (x)`$ terms has been noted by Chang, Root and Yan , Burkardt , and Ji and Choi.
The off-diagonal $`n+1n1`$ contributions give a new perspective for the physics of $`B`$-decays. A semi-leptonic decay involves not only matrix elements where a quark changes flavor, but also a contribution where the leptonic pair is created from the annihilation of a $`q\overline{q^{}}`$ pair within the Fock states of the initial $`B`$ wavefunction. The semi-leptonic decay thus can occur from the annihilation of a nonvalence quark-antiquark pair in the initial hadron. This feature carries over to exclusive hadronic $`B`$-decays, such as $`B^0\pi ^{}D^+`$. In this case the pion can be produced from the coalescence of a $`d\overline{u}`$ pair emerging from the initial higher particle number Fock wavefunction of the $`B`$. The $`D`$ meson is then formed from the remaining quarks after the internal exchange of a $`W`$ boson.
Light-cone Fock state wavefunctions thus encode all of the bound state quark and gluon properties of hadrons such as spin and flavor correlations in the form of universal process- and frame- independent amplitudes. Is there any hope of computing these wavefunctions from first principles? In the discretized light-cone quantization method (DLCQ), periodic boundary conditions are introduced in $`b_{}`$ and $`x^{}`$ so that the momenta $`k_i=n_{}\pi /L_{}`$ and $`x_i^+=n_i/K`$ are discrete. A global cutoff in invariant mass of the partons in the Fock expansion is also introduced. Solving the quantum field theory then reduces to the problem of diagonalizing the finite-dimensional hermitian matrix $`H_{LC}`$ on a finite discrete Fock basis. The DLCQ method has now become a standard tool for solving both the spectrum and light-cone wavefunctions of one-space one-time theories. Virtually any $`1+1`$ quantum field theory, including “reduced QCD” (which has both quark and gluonic degrees of freedom) can be completely solved using DLCQ. Hiller, McCartor, and I have recently shown that the use of covariant Pauli-Villars regularization with discrete light-cone quantization allows one to obtain the spectrum and light-cone wavefunctions of simplified theories in physical space-time dimensions, such as (3+1) Yukawa theory. Dalley et al. have also showed how one can use DLCQ with a transverse lattice to solve gluonic QCD. Remarkably, the spectrum obtained for gluonium states is in remarkable agreement with lattice gauge theory results, but with a huge reduction of numerical effort. One can also formulate DLCQ so that supersymmetry is exactly preserved in the discrete approximation, thus combining the power of DLCQ with the beauty of supersymmetry. The “SDLCQ” method has been applied to several interesting supersymmetric theories, to the analysis of zero modes, vacuum degeneracy, massless states, mass gaps, and theories in higher dimensions, and even tests of the Maldacena conjecture. Broken supersymmetry is interesting in DLCQ, since it may serve as a method for regulating non-Abelian theories.
Another remarkable advantage of light-cone quantization is that the vacuum state $`|\mathrm{\hspace{0.17em}0}`$ of the full QCD Hamiltonian coincides with the free vacuum. For example, as discussed by Bassetto, the computation of the spectrum of $`QCD(1+1)`$ in equal time quantization requires constructing the full spectrum of non perturbative contributions (instantons). However, light-cone methods such as DLCQ, give the correct result immediately, without any need for vacuum related contributions.
It is also possible to model the light-cone wavefunctions. For example one can find simple forms for the three valence quark wavefunctions $`\psi _{qqq/N}^{LC}(x_i,k_i,\lambda _i)`$ satisfying $`SU(6)`$ spin-flavor symmetry which can account for the “static” properties of the baryons: their magnetic moments, axial couplings $`g_A`$, and charged radii. Such LC models satisfy the rigorous constraint that the magnetic moment of a composite spin-half state must approach its Dirac moment $`\mu =e/2M`$ in the pointlike limit $`R0`$ with $`M`$ fixed, where $`R^2=dF_1(q^2)/dq^2|_{q^20}`$. In addition, the LC model predicts that the quark chirality measures $`\mathrm{\Delta }q,\mathrm{\Delta }\mathrm{\Sigma },`$ and $`g_A`$ vanish in the same pointlike limit. For the physical proton, their values are approximately $`0.75`$ of the nonrelativistic values. Physically, this reduction occurs because the quark chirality, (which can be identified with quark helicity $`\stackrel{}{S}_q\widehat{p}`$ in the massless limit) fluctuates strongly as the bound state becomes pointlike. Thus one cannot identify the chirality measures which appear in the Bjorken and Ellis-Jaffe-Gourdin sum rules in a relativistic theory with the spin projection of the equal-time wavefunction in the hadron rest-frame. One can also construct exact models based on the perturbative structure of the QED calculation of the anomalous moment using Pauli-Villars spectra. As discussed by Bo-Qiang Ma in these proceedings, the chirality sum rules are effectively measures of the light-cone spin projections, not the usual equal-time spin.
## 2 Intrinsic versus Extrinsic Sea
The deep inelastic scattering data show that the nonperturbative structure of nucleons is more complex than a simple three quark bound state. For example, if the sea quarks were generated solely by perturbative QCD evolution via gluon splitting, the anti-quark distributions would be approximately isospin symmetric. However, the $`\overline{u}(x)`$ and $`\overline{d}(x)`$ antiquark distributions of the proton at $`Q^210`$ GeV<sup>2</sup> are found to be quite different in shape and thus must reflect dynamics intrinsic to the proton’s structure. Evidence for a difference between the $`\overline{s}(x)`$ and $`s(x)`$ distributions has also been claimed. There have also been surprises associated with the chirality distributions $`\mathrm{\Delta }q=q_/q_/`$ of the valence quarks which show that a simple valence quark approximation to nucleon spin structure functions is far from the actual dynamical situation.
It is helpful to categorize the parton distributions as “intrinsic”—pertaining to the long-time scale composition of the target hadron, and “extrinsic”—reflecting the short-time substructure of the individual quarks and gluons themselves. Gluons carry a significant fraction of the proton’s spin as well as its momentum. Since gluon exchange between valence quarks contributes to the $`p\mathrm{\Delta }`$ mass splitting, it follows that the gluon distributions cannot be solely accounted for by gluon bremsstrahlung from individual quarks, the process responsible for DGLAP evolutions of the structure functions. Similarly, in the case of heavy quarks, $`s\overline{s}`$, $`c\overline{c}`$, $`b\overline{b}`$, the diagrams in which the sea quarks are multi-connected to the valence quarks are intrinsic to the proton structure itself.
The higher Fock state of the proton $`|uuds\overline{s}`$ should resemble a $`|K\mathrm{\Lambda }`$ intermediate state, since this minimizes its invariant mass $``$. In such a state, the strange quark has a higher mean momentum fraction $`x`$ than the $`\overline{s}`$. Similarly, the helicity intrinsic strange quark in this configuration will be anti-aligned with the helicity of the nucleon. This $`Q\overline{Q}`$ asymmetry is a striking feature of the intrinsic heavy-quark sea.
In a recent paper, Merino, Rathsman, and I have shown that the asymmetry in the fractional energy of charm versus anticharm jets produced in high energy diffractive photoproduction is sensitive to the interference of the Odderon $`(C=)`$ and Pomeron $`(C=+)`$ exchange amplitudes in QCD. We can predict the dynamical shape of the asymmetry in a simple model and have estimated its magnitude to be of the order 15% using an Odderon coupling to the proton which saturates constraints from proton-proton vs. proton-antiproton elastic scattering. Measurements of this asymmetry at HERA could provide the first evidence for the presence of Odderon exchange in the high energy limit of strong interactions.
The main features of the heavy sea quark-pair contributions of the Fock state expansion of light hadrons can be derived from perturbative QCD, since $`_n^2`$ grows with $`m_Q^2`$. One identifies two contributions to the heavy quark sea, the “extrinsic” contributions which correspond to ordinary gluon splitting, and the “intrinsic” sea which is multi-connected via gluons to the valence quarks. The intrinsic sea is thus sensitive to the hadronic bound state structure. The maximal contribution of the intrinsic heavy quark occurs at $`x_Qm_Q/_im_{}`$ where $`m_{}=\sqrt{m^2+k_{}^2}`$; i.e. at large $`x_Q`$, since this minimizes the invariant mass $`_n^2`$. The measurements of the charm structure function by the EMC experiment are consistent with intrinsic charm at large $`x`$ in the nucleon with a probability of order $`0.6\pm 0.3\%`$. Similarly, one can distinguish intrinsic gluons which are associated with multi-quark interactions and extrinsic gluon contributions associated with quark substructure. One can also use this framework to isolate the physics of the anomaly contribution to the Ellis-Jaffe sum rule. Thus neither gluons nor sea quarks are solely generated by DGLAP evolution, and one cannot define a resolution scale $`Q_0`$ where the sea or gluon degrees of freedom can be neglected.
Light-cone wavefunctions are the natural quantities to encode hadron properties and to bridge the gap between empirical constraints and theoretical predictions for the bound state solutions. We can thus envision a program to construct the $`\mathrm{\Psi }_n^P(x_i,k_i,\lambda _i)`$ using not only data, but theoretical constraints such as
(1) Since the state is far off shell at large invariant mass $``$, one can derive rigorous limits on the $`x1`$, high $`k_{}`$, and high $`_n^2`$ behavior of the wavefunctions in the perturbative domain.
(2) Ladder relations connecting state of different particle number follow from the QCD equation of motion and lead to Regge behavior of the quark and gluon distributions at $`x0`$. QED provides a constraint at $`N_C0.`$
(3) One can obtain guides to the exact behavior of LC wavefunctions in QCD from analytic or DLCQ solutions to toy models such as “reduced” $`QCD(1+1).`$
(4) QCD sum rules, lattice gauge theory moments, and QCD inspired models such as the bag model, chiral theories, provide important constraints.
(5) Since the LC formalism is valid at all scales, one can utilize empirical constraints such as the measurements of magnetic moments, axial couplings, form factors, and distribution amplitudes.
(6) In the nonrelativistic limit, the light-cone and many-body Schrödinger theory formalisms must match.
## 3 The Light-Cone Factorization Scheme
Factorization theorems for hard exclusive, semi-exclusive, and diffractive processes allow a rigorous separation of soft non-perturbative dynamics of the bound state hadrons from the hard dynamics of a perturbatively-calculable quark-gluon scattering amplitude.
Roughly, the direct proofs of factorization in the light-cone scheme proceed as follows: In hard inclusive reactions all intermediate states are divided according to $`_n^2<\mathrm{\Lambda }^2`$ and $`_n^2>\mathrm{\Lambda }^2`$ domains. The lower region is associated with the quark and gluon distributions defined from the absolute squares of the LC wavefunctions in the light cone factorization scheme. In the high invariant mass regime, intrinsic transverse momenta can be ignored, so that the structure of the process at leading power has the form of hard scattering on collinear quark and gluon constituents, as in the parton model. The attachment of gluons from the LC wavefunction to a propagator in the hard subprocess is power-law suppressed in LC gauge, so that the minimal $`22`$ quark-gluon subprocesses dominate. The higher order loop corrections lead to the DGLAP evolution equations, as well as the higher order in $`\alpha _s`$ corrections to the hard amplitude.
It is important to note that the effective starting point for the PQCD evolution of the structure functions cannot be taken as a constant $`Q_0^2`$ since as $`x1`$ the invariant mass $`_n`$ exceeds the resolution scale $`\mathrm{\Lambda }`$. Thus in effect, evolution is quenched at $`x1`$.
One of the most interesting aspects of deep inelastic lepton-proton scattering is the contribution to the $`g_1^p`$ spin-dependent structure function from photon-gluon fusion subprocesses $`\gamma ^{}(q)g(p)q\overline{q}`$. Naively, one would expect zero contributions from light mass $`q\overline{q}`$ pairs to the first moment $`_0^1𝑑xg_1^p(x,Q^2)`$ since the $`q`$ and $`\overline{q}`$ have opposite helicities. In fact, this is not the case if the quark mass $`m_q`$ is small compared to a scale set by the spacelike gluon virtuality $`p^2`$. This is the origin of the so-called anomalous correction $`3\frac{\alpha _s}{2\pi }\mathrm{\Delta }g`$ to the Ellis-Jaffe sum rule for isospin zero targets assuming three light flavors. Here $`\mathrm{\Delta }g`$ is the helicity carried by gluons in the hadron target, $`\mathrm{\Delta }g(Q)=_0^1𝑑x[g_{}(x,Q)g_{}(x,Q)]`$, at the factorization scale $`Q`$. If the sea quark mass is heavy compared to the gluon virtuality $`4m_q^2P^2=p^2`$, the photon-gluon fusion contribution to $`_0^1𝑑xg_1(x,Q^2)`$ vanishes to leading order in $`\alpha _s(Q^2)`$. This result follows from a general theorem based on the Drell-Hearn-Gerasimov sum rule which states that the integral
$$_{\nu _\pi }^{\mathrm{}}\frac{d\nu }{\nu }\sigma _{\gamma abc}(\nu )=0(\alpha ^3);$$
(2)
i.e., vanishes at order $`\alpha ^2`$ for any $`22`$ Standard Model process. In the present case the gluon (for $`p^2=0`$) takes the role of the target $`a`$. For large $`Q^2`$, the DHG integral evolves to the first moment of the helicity-dependent structure function $`g_1(x,Q^2)`$ for any photon virtuality. Thus the fusion $`\gamma ^{}gq\overline{q}`$ contribution to $`_0^1𝑑xg_1(x,Q^2)`$ vanishes for small gluon virtuality $`P^24m_q^2`$, $`P^2Q^2`$. This virtuality can be interpreted directly in the light-cone factorization scheme. If the off-shellness of the state is larger than the quark pair mass, one obtains the usual anomaly contribution. The specific contribution of a given sea quark pair $`q\overline{q}`$ thus depends not only on $`Q^2`$, but more critically on the ratio of scales $`p^2/4m_q^2`$. The spectrum $`N(p^2)`$ of gluon virtuality in the target nucleon in turn depends in detail on the physics of the nucleon light-cone wavefunction. Bass, Schmidt and I have discussed specific forms which allow one to estimate the effect of extrinsic and intrinsic $`s`$ and $`c`$ quarks on the anomaly. The application of the DHG theorem to photoabsorption is more general than leading twist. The fusion contribution to the DHG moment vanishes even if $`Q^2<4m_q^2`$, as long as the gluon virtuality can be neglected. The result also holds for the weak as well as electromagnetic current probes.
In exclusive amplitudes, the LC wavefunctions are the interpolating functions between the quark and gluon states and the hadronic states. In an exclusive amplitude involving a hard scale $`Q^2`$, the intermediate states can again be divided in invariant mass domains. The high invariant mass contributions to the amplitude has the structure of a hard scattering process $`T_H`$ in which the hadrons are replaced by their respective (collinear) quarks and gluons. In light-cone gauge only the minimal Fock states contribute to the leading power-law fall-off of the exclusive amplitude. The wavefunctions in the lower invariant mass domain can be integrated up to the invariant mass cutoff $`\mathrm{\Lambda }`$ and replaced by the gauge invariant distribution amplitudes, $`\varphi _H(x_i,\mathrm{\Lambda })`$. Final-state and initial-state corrections from gluon attachments to lines connected to the color-singlet distribution amplitudes cancel at leading twist. Thus the key non-perturbative input for exclusive processes is the gauge and frame independent hadron distribution amplitude defined as the integral of the valence (lowest particle number) Fock wavefunction; e.g. for the pion
$$\varphi _\pi (x_i,\mathrm{\Lambda })d^2k_{}\psi _{q\overline{q}/\pi }^{(\mathrm{\Lambda })}(x_i,\stackrel{}{k}_i,\lambda )$$
(3)
where the global cutoff $`\mathrm{\Lambda }`$ is identified with the resolution $`Q`$. The distribution amplitude controls leading-twist exclusive amplitudes at high momentum transfer, and it can be related to the gauge-invariant Bethe-Salpeter wavefunction at equal light-cone time. The logarithmic evolution of hadron distribution amplitudes $`\varphi _H(x_i,Q)`$ can be derived from the perturbatively-computable tail of the valence light-cone wavefunction in the high transverse momentum regime.
The features of exclusive processes to leading power in the transferred momenta are well known:
(1) The leading power fall-off is given by dimensional counting rules for the hard-scattering amplitude: $`T_H1/Q^{n1}`$, where $`n`$ is the total number of fields (quarks, leptons, or gauge fields) participating in the hard scattering. Thus the reaction is dominated by subprocesses and Fock states involving the minimum number of interacting fields. The hadronic amplitude follows this fall-off modulo logarithmic corrections from the running of the QCD coupling, and the evolution of the hadron distribution amplitudes. In some cases, such as large angle $`pppp`$ scattering, pinch contributions from multiple hard-scattering processes must also be included. The general success of dimensional counting rules implies that the effective coupling $`\alpha _V(Q^{})`$ controlling the gluon exchange propagators in $`T_H`$ are frozen in the infrared, i.e., have an infrared fixed point, since the effective momentum transfers $`Q^{}`$ exchanged by the gluons are often a small fraction of the overall momentum transfer. The pinch contributions are then suppressed by a factor decreasing faster than a fixed power.
(2) The leading power dependence is given by hard-scattering amplitudes $`T_H`$ which conserve quark helicity. Since the convolution of $`T_H`$ with the light-cone wavefunctions projects out states with $`L_z=0`$, the leading hadron amplitudes conserve hadron helicity; i.e., the sum of initial and final hadron helicities are conserved. Hadron helicity conservation thus follows from the underlying chiral structure of QCD. For example, hadron helicity conservation predicts the suppression of vector meson states produced with $`J_z=\pm 1`$ in $`e^+e^=`$ annihilation to vector-pseudoscalar final states. However, $`J/\psi \rho \pi `$ appears to occur copiously whereas $`\psi ^{}\rho \pi `$ has never been conserved. The PQCD analysis assumes that a heavy quarkonium state such as the $`J/\psi `$ always decays to light hadrons via the annihilation of its heavy quark constituents to gluons. However, as Karliner and I have shown, the transition $`J/\psi \rho \pi `$ can also occur by the rearrangement of the $`c\overline{c}`$ from the $`J/\psi `$ into the $`|q\overline{q}c\overline{c}`$ intrinsic charm Fock state of the $`\rho `$ or $`\pi `$. On the other hand, the overlap rearrangement integral in the decay $`\psi ^{}\rho \pi `$ will be suppressed since the intrinsic charm Fock state radial wavefunction of the light hadrons will evidently not have nodes in its radial wavefunction. This observation can provide a natural explanation of the long-standing puzzle why the $`J/\psi `$ decays prominently to two-body pseudoscalar-vector final states, whereas the $`\psi ^{}`$ does not.
I will mention here several other applications of the light-cone formalism and factorization scheme:
Diffractive vector meson photoproduction. The light-cone Fock wavefunction representation of hadronic amplitudes allows a simple eikonal analysis of diffractive high energy processes, such as $`\gamma ^{}(Q^2)p\rho p`$, in terms of the virtual photon and the vector meson Fock state light-cone wavefunctions convoluted with the $`gpgp`$ near-forward matrix element. One can easily show that only small transverse size $`b_{}1/Q`$ of the vector meson distribution amplitude is involved. The hadronic interactions are minimal, and thus the $`\gamma ^{}(Q^2)N\rho N`$ reaction can occur coherently throughout a nuclear target in reactions without absorption or shadowing. The $`\gamma ^{}AVA`$ process is thus a laboratory for testing QCD color transparency.
Regge behavior of structure functions. The light-cone wavefunctions $`\psi _{n/H}`$ of a hadron are not independent of each other, but rather are coupled via the equations of motion. The constraint of finite “mechanical” kinetic energy allows one to derive “ladder relations” which interrelate the light-cone wavefunctions of states differing by one or two gluons. We can then use these relations to derive the Regge behavior of both the polarized and unpolarized structure functions at $`x0`$, extending Mueller’s derivation of the BFKL hard QCD pomeron using the properties of heavy quarkonium light-cone wavefunctions at large $`N_C`$ QCD.
Structure functions at large $`x_{bj}`$. The behavior of structure functions where one quark has the entire momentum requires the knowledge of LC wavefunctions with $`x1`$ for the struck quark and $`x0`$ for the spectators. This is a highly off-shell configuration, and thus one can rigorously derive quark-counting and helicity-retention rules for the power-law behavior of the polarized and unpolarized quark and gluon distributions in the $`x1`$ endpoint domain. Modulo DGLAP evolution, the counting rule for finding parton $`a`$ in hadron $`a`$ at large $`x1`$ $`G_{a/A}(x,Q)(1x)^{2n_{\mathrm{spect}}1+2|\mathrm{\Delta }S_z|}`$ where $`n_{\mathrm{spect}}`$ is the minimum number of partons left behind when parton $`a`$ is removed from $`A`$, and $`\mathrm{\Delta }S_z`$ is the difference of the $`a`$ and $`A`$ helicities. This predicts $`(1x)^3`$ behavior for valence quarks aligned in helicity with the proton helicity, and $`(1x)^3`$ behavior for anti-aligned quarks. As noted above, DGLAP evolution is quenched in the large $`x`$ limit in the fixed $`W^2`$ domain. Burkardt, Schmidt, and I have discussed the phenomenological implications of this rule for gluon and sea distributions.
Materialization of far-off-shell configurations. In a high energy hadronic collisions, the highly-virtual states of a hadron can be materialized into physical hadrons simply by the soft interaction of any of the constituents. Thus a proton state with intrinsic charm $`|uud\overline{c}c`$ can be materialized by the interaction of a light-quark in the target, producing a $`J/\psi `$ at large $`x_F`$. The production occurs on the front-surface of a target nucleus, implying an $`A^{2/3}`$ $`J/\psi `$ production cross section at large $`x_F,`$ which is consistent with experiment, such as Fermilab experiments E772 and E866.
Comover phenomena. Light-cone wavefunctions describe not only the partons that interact in a hard subprocess but also the associated partons freed from the projectile. The projectile partons which are comoving (i.e., which have similar rapidity) with the final state quarks and gluons can interact strongly producing (a) leading particle effects, such as those seen in open charm hadroproduction; (b) suppression of quarkonium in favor of open heavy hadron production, as seen in the E772 experiment; (c) changes in color configurations and selection rules in quarkonium hadroproduction, as has been emphasized by Hoyer and Peigne. Further, more than one parton from the projectile can enter the hard subprocess, producing dynamical higher twist contributions, as seen for example in Drell-Yan experiments.
Jet hadronization in light-cone QCD. One of the goals of nonperturbative analysis in QCD is to compute jet hadronization from first principles. The DLCQ solutions provide a possible method to accomplish this. By inverting the DLCQ solutions, we can write the “bare” quark state of the free theory as $`|q_0=|nn|q_0`$ where now $`\{|n\}`$ are the exact DLCQ eigen states of $`H_{LC}`$, and $`n|q_0`$ are the DLCQ projections of the eigen-solutions. The expansion in automatically infrared and ultraviolet regulated if we impose global cutoffs on the DLCQ basis: $`\lambda ^2<\mathrm{\Delta }_n^2<\mathrm{\Lambda }^2`$ where $`\mathrm{\Delta }_n^2=_n^2(\mathrm{\Sigma }_i)^2`$. It would be interesting to study jet hadronization at the amplitude level for the existing DLCQ solutions to QCD (1+1) and collinear QCD.
Hidden Color. The deuteron form factor at high $`Q^2`$ is sensitive to wavefunction configurations where all six quarks overlap within an impact separation $`b_i<𝒪(1/Q);`$ the leading power-law fall off predicted by QCD is $`F_d(Q^2)=f(\alpha _s(Q^2))/(Q^2)^5`$, where, asymptotically, $`f(\alpha _s(Q^2))\alpha _s(Q^2)^{5+2\gamma }`$. The derivation of the evolution equation for the deuteron distribution amplitude and its leading anomalous dimension $`\gamma `$ is given by Ji, Lepage, and myself. In general, the six-quark wavefunction of a deuteron is a mixture of five different color-singlet states. The dominant color configuration at large distances corresponds to the usual proton-neutron bound state. However at small impact space separation, all five Fock color-singlet components eventually acquire equal weight, i.e., the deuteron wavefunction evolves to 80% “hidden color.” The relatively large normalization of the deuteron form factor observed at large $`Q^2`$ points to sizable hidden color contributions. Hidden color components can play a predominant role in the reaction $`\gamma dJ/\psi pn`$ at threshold if it is dominated by the multi-fusion process $`\gamma ggJ/\psi `$.
Spin-Spin Correlations and the Charm Threshold. One of the most striking anomalies in elastic proton-proton scattering is the large spin correlation $`A_{NN}`$ observed at large angles. At $`\sqrt{s}5`$ GeV, the rate for scattering with incident proton spins parallel and normal to the scattering plane is four times larger than that for scattering with anti-parallel polarization. This strong polarization correlation can be attributed to the onset of charm production in the intermediate state at this energy. A resonant intermediate state $`|uuduudc\overline{c}`$ has odd intrinsic parity and can thus couple to the $`J=L=S=1`$ initial state, thus strongly enhancing scattering when the incident projectile and target protons have their spins parallel and normal to the scattering plane. The charm threshold can also explain the anomalous change in color transparency observed at the same energy in quasi-elastic $`pp`$ scattering. A crucial test is the observation of open charm production near threshold with a cross section of order of $`1\mu `$b. Analogous strong spin effects should also appear at the strangeness threshold and in exclusive photon-proton reactions such as large angle Compton scattering and pion photoproduction near the strangeness and charm thresholds.
## 4 Self-Resolved Diffractive Reactions and Light Cone Wavefunctions
Diffractive multi-jet production in heavy nuclei provides a novel way to measure the shape of the LC Fock state wavefunctions and test color transparency. For example, consider the reaction $`\pi A\mathrm{Jet}_1+\mathrm{Jet}_2+A^{}`$ at high energy where the nucleus $`A^{}`$ is left intact in its ground state. The transverse momenta of the jets have to balance so that $`\stackrel{}{k}_i+\stackrel{}{k}_2=\stackrel{}{q}_{}<R_{}^{1}{}_{A}{}^{},`$ and the light-cone longitudinal momentum fractions have to add to $`x_1+x_21`$ so that $`\mathrm{\Delta }p_L<R_A^1`$. The process can then occur coherently in the nucleus. Because of color transparency, i.e., the cancelation of color interactions in a small-size color-singlet hadron, the valence wavefunction of the pion with small impact separation, will penetrate the nucleus with minimal interactions, diffracting into jet pairs. The $`x_1=x`$, $`x_2=1x`$ dependence of the di-jet distributions will thus reflect the shape of the pion distribution amplitude; the $`\stackrel{}{k}_1\stackrel{}{k}_2`$ relative transverse momenta of the jets also gives key information on the underlying shape of the valence pion wavefunction. The QCD analysis can be confirmed by the observation that the diffractive nuclear amplitude extrapolated to $`t=0`$ is linear in nuclear number $`A`$, as predicted by QCD color transparency. The integrated diffractive rate should scale as $`A^2/R_A^2A^{4/3}`$. A diffractive dissociation experiment of this type, E791, is now in progress at Fermilab using 500 GeV incident pions on nuclear targets. The preliminary results from E791 appear to be consistent with color transparency. The momentum fraction distribution of the jets is consistent with a valence light-cone wavefunction of the pion consistent with the shape of the asymptotic distribution amplitude, $`\varphi _\pi ^{\mathrm{asympt}}(x)=\sqrt{3}f_\pi x(1x)`$. Data from CLEO for the $`\gamma \gamma ^{}\pi ^0`$ transition form factor also favor a form for the pion distribution amplitude close to the asymptotic solution to the perturbative QCD evolution equation. It will also be interesting to study diffractive tri-jet production using proton beams $`pA\mathrm{Jet}_1+\mathrm{Jet}_2+\mathrm{Jet}_3+A^{}`$ to determine the fundamental shape of the 3-quark structure of the valence light-cone wavefunction of the nucleon at small transverse separation. One interesting possibility is that the distribution amplitude of the $`\mathrm{\Delta }(1232)`$ for $`J_z=1/2,3/2`$ is close to the asymptotic form $`x_1x_2x_3`$, but that the proton distribution amplitude is more complex. This would explain why the $`p\mathrm{\Delta }`$ transition form factor appears to fall faster at large $`Q^2`$ than the elastic $`pp`$ and the other $`pN^{}`$ transition form factors. Conversely, one can use incident real and virtual photons: $`\gamma ^{}A\mathrm{Jet}_1+\mathrm{Jet}_2+A^{}`$ to confirm the shape of the calculable light-cone wavefunction for transversely-polarized and longitudinally-polarized virtual photons. Such experiments will open up a direct window on the amplitude structure of hadrons at short distances.
The diffractive dissociation of a hadron or nucleus can also occur via the Coulomb dissociation of a beam particle on an electron beam (e.g. at HERA or eRHIC) or on the strong Coulomb field of a heavy nucleus (e.g. at RHIC or nuclear collisions at the LHC). The amplitude for Coulomb exchange at small momentum transfer is proportional to the first derivative $`_ie_i\frac{}{\stackrel{}{k}_{Ti}}\psi `$ of the light-cone wavefunction, summed over the charged constituents. The Coulomb exchange reactions fall off less fast at high transverse momentum compared to pomeron exchange reactions since the light-cone wavefunction is effective differentiated twice in two-gluon exchange reactions.
For example, consider the Coulomb dissociation of a high energy proton at HERA. The proton can dissociate into three jets corresponding to the three-quark structure of the valence light-cone wavefunction. We can demand that the produced hadrons all fall outside of an “exclusion cone” of opening angle $`\theta `$ in the proton’s fragmentation region. Effectively all of the light-cone momentum $`_jx_j1`$ of the proton’s fragments will thus be produced outside the exclusion cone. This requirement then limits the invariant mass of the Fock state $`_n^2>\mathrm{\Lambda }^2=P^{+2}\mathrm{sin}^2\theta /4`$ from below, so that perturbative QCD counting rules can predict the fall-off in the jet system invariant mass $``$. At large invariant mass one expects the three-quark valence Fock state of the proton to dominate. The segmentation of the forward detector in azimuthal angle $`\varphi `$ can be used to identify structure and correlations associated with the three-quark light-cone wavefunction. A further discussion is in progress.
The light-cone formalism is also applicable to the description of nuclei in terms of their nucleonic and mesonic degrees of freedom. Self-resolving diffractive jet reactions in high energy electron-nucleus collisions and hadron-nucleus collisions at moderate momentum transfers can thus be used to resolve the light-cone wavefunctions of nuclei.
## 5 Semi-Exclusive Processes: New Probes of Hadron Structure
A new class of hard “semi-exclusive” processes of the form $`A+BC+Y`$, have been proposed as new probes of QCD. These processes are characterized by a large momentum transfer $`t=(p_Ap_C)^2`$ and a large rapidity gap between the final state particle $`C`$ and the inclusive system $`Y`$. Here $`A,B`$ and $`C`$ can be hadrons or (real or virtual) photons. The cross sections for such processes factorize in terms of the distribution amplitudes of $`A`$ and $`C`$ and the parton distributions in the target $`B`$. Because of this factorization, semi-exclusive reactions provide a novel array of generalized currents, which not only give insight into the dynamics of hard scattering QCD processes, but also allow experimental access to new combinations of the universal quark and gluon distributions.
## 6 Summary
In this talk I have discussed how universal, process-independent and frame-independent light-cone Fock-state wavefunctions can be used to encode the properties of a hadron in terms of its fundamental quark and gluon degrees of freedom. Given the proton’s light-cone wavefunctions, one can compute not only the moments of the quark and gluon distributions measured in deep inelastic lepton-proton scattering, but also the multi-parton correlations which control the distribution of particles in the proton fragmentation region and dynamical higher twist effects. Light-cone wavefunctions also provide a systematic framework for evaluating exclusive hadronic matrix elements, including time-like heavy hadron decay amplitudes and form factors. The formalism also provides a physical factorization scheme for separating hard and soft contributions in both exclusive and inclusive hard processes. A new type of jet production reaction, “self-resolving diffractive interactions” can provide direct information on the light-cone wavefunctions of hadrons in terms of their QCD degrees of freedom, as well as the composition of nuclei in terms of their nucleon and mesonic degrees of freedom. Progress in QCD is driven by experiment, and we are fortunate that there are new experimental facilities such as Jefferson laboratory, new studies of exclusive processes $`e^+e^{}`$ and $`\gamma \gamma `$ processes at the high luminosity $`B`$ factories, as well as the new accelerators and colliders now being planned to further advance the study of QCD phenomena.
## Acknowledgments
Work supported by the Department of Energy under contract number DE-AC03-76SF00515. I wish to thank Shunzo Kumano and Akihisa Kohama for their kind hospitality at this symposium and the RIKEN Institute. I also thank Paul Hoyer, Markus Diehl, Stephane Peigne, Bo-Qiang Ma, Ivan Schmidt, and Dae Sung Hwang for helpful conversations.
|
warning/0002/cond-mat0002265.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
For non–equilibrium systems in low dimensions, an understanding can often be gained by studying rather simple models . One of the important examples of these systems are reaction–diffusion processes on a one–dimensional lattice, which their dynamics are fully specified by their master equation . In some cases, it is possible to solve the master equation exactly. In recent years, there has been enormous progress in the field of exactly solvable non–equilibrium processes. These developments were mainly triggered by the observation that the Liouville operator of certain (1+1)–dimensional reaction–diffusion models may be related to Hamiltonians of previously known quantum spin systems , .
One of the simplest examples of reaction–diffusion processes are Asymmetric Simple Exclusion Processes (ASEP) , , , which has been used to describe various problems in different fields of interest, such as the kinetics of bipolymerization, dynamical models of interface growth , and traffic models. The totally ASEP model has been solved exactly by imposing the appropriate boundary condition on the probabilities appear in the master equation. The totally ASEP model describes a process in which each lattice site can be occupied by at most one particle and the particle hops with rate one to its right neighboring site if it is not already occupied, otherwise the attempted move is rejected.
There are some other interesting and more complicated processes which can be solved by the method developed in , namely by choosing a suitable boundary condition for the master equation. For example, it has been shown that the so called “generalized totally ASEP model” can be solved exactly by this method . In this model, even if the right neighboring site of a particle is occupied, the particle hops to the next right site by pushing all the neighboring particles to their next right sites, with a rate which depends on the number of right neighboring particles. This model has been further generalized in by considering both the right and left hopping of the particles.
In this paper we are going to consider a class of integrable models in which there are two species of particles which can hop to their right neighboring sites if those are not occupied, and also the particles interact with each other if they are in adjacent sites. The details of this nearest–neighboring interaction depends on the specific considered model (see for some recent works in two– and three–species reaction–diffusion processes). The important point in integrable reaction–diffusion processes with more than one type of particle is that, as we will show, the two–particle S–matrix of the reaction, which specify the $`N`$–point functions, must satisfy the Quantum Yang–Baxter Equation (QYBE). Therefore, as we expect, the number of integrable models, in the sense that its $`N`$–particle S–matrix can be factorized into a product of two–particle S–matrices, is very few. In this paper we will find all two–species integrable reaction–diffusion processes which have the following properties:
1. the particles hop to their right neighboring sites if these sites are not occupied,
2. the interaction occurs only between nearest–neighbor particles,
3. the particles can be annihilated or created, with the only restriction that the total number of particles is fixed,
4. all the interactions, including diffusions, occur with the same rate.
We show that among the $`2^{12}=4096`$ types of the interactions which have the above–mentioned properties and can be modeled by a master equation and an appropriate boundary condition, there are only 42 interactions which are integrable (their two–particle S–matrices satisfy the QYBE), and from these 42 interactions, only 28 of them are independent. Some of these may be new solutions of QYBE.
The plan of the paper is as following. In section 2, we introduce the first kind of this interactions, which was our initial motivation in this work, in which besides the usual hopping, the two types of particles interact as : $`A+BB+B`$ and $`B+AB+B.`$ We show that this interaction can be modeled by the usual master equation of ASEP and four boundary conditions. We also show that the model is integrable. Note that one can look at this model (see eq.(1)) as a simple one–dimensional model of spread of disease. If we consider $`A`$ particles to be the healthy individuals and the $`B`$ particles the diseased ones, then we expect that when $`A`$ and $`B`$ particle are near to each other, healthy one may become diseased (in other words $`B`$ transmits disease to $`A`$). In section 3, we compute the exact two–particle conditional probabilities of this interaction and study the long–time behavior of this probabilities. And finally in section 4, we investigate the class of integrable models which have the four above–mentioned properties and deduce that there are 28 different models, which the totally ASEP model and our first model introduced in section 2, are two of them.
## 2 $`ABBB`$ and $`BABB`$ reaction diffusion process
### 2.1 The master equation
Suppose there are $`N`$ particles of two types $`A`$ and $`B`$ on an infinite one dimensional lattice, with interactions
$$\begin{array}{ccc}\hfill A\mathrm{}& & \mathrm{}A,\hfill \\ \hfill B\mathrm{}& & \mathrm{}B,\hfill \\ \hfill AB& & BB,\hfill \\ \hfill BA& & BB,\hfill \end{array}$$
(1)
all occur with *equal* rate, which can be scaled to one. In eq.(1), we denote the vacancy by notation $`\mathrm{}`$. The basic quantities we are interested in are the probabilities $`P_{\alpha _1,\alpha _2,\mathrm{},\alpha _N}(x_1,x_2,\mathrm{},x_N;t)`$ for finding at time $`t`$ a particle of type $`\alpha _1`$ at site $`x_1`$, a particle of type $`\alpha _2`$ at site $`x_2`$, etc. Each $`\alpha _i`$ can be $`A`$ or $`B`$. Following , we take these functions to define probabilities only in the physical region $`x_1<x_2<\mathrm{}<x_N`$. The surfaces where any of the two adjacent coordinates are equal, are the boundaries of the physical region. In the subset of the physical region where $`x_{i+1}x_i>1,i`$, we have only hopping of the particles and therefore the master equation can be written as
$$\begin{array}{ccc}\hfill \frac{}{t}P_{\alpha _1,\alpha _2,\mathrm{},\alpha _N}(x_1,x_2,\mathrm{},x_N;t)& =& P_{\alpha _1,\mathrm{},\alpha _N}(x_11,x_2,\mathrm{},x_N;t)+\mathrm{}+\hfill \\ & & +P_{\alpha _1,\mathrm{},\alpha _N}(x_1,x_2,\mathrm{},x_N1;t)\hfill \\ & & NP_{\alpha _1,\mathrm{},\alpha _N}(x_1,\mathrm{},x_N;t).\hfill \end{array}$$
(2)
As is clear from eq.(2), when $`x_{i+1}=x_i+1`$ for some $`i`$’s, the one or more of the probability functions go out from the physical region and therefore the eq.(2) has to be supplemented by some boundary conditions. The particular choice of the boundary condition depends on the details of the interactions of particles. For example, it can be shown that in the totally ASEP model, the suitable boundary condition is ,
$$P(x,x)=P(x,x+1),x,$$
(3)
in which the time variable and also all the other coordinates have been suppressed for simplicity. The master equation (2) and the boundary condition (3) replace the very large number of equations which one should write by considering the multitude of cases which arises in different possible configurations.
To model the interaction (1), we claim that the suitable boundary conditions are :
$$P_{AA}(x,x)=P_{AA}(x,x+1),$$
(4)
$$P_{BB}(x,x)=P_{BB}(x,x+1)+P_{AB}(x,x+1)+P_{BA}(x,x+1),$$
(5)
$$P_{AB}(x,x)=P_{BA}(x,x)=0,$$
(6)
where we have again suppressed the positions of all the other particles. By looking at (1), it is obvious that if we have only $`A`$ particles , the process is exactly the same as totally ASEP. It is the reason of appearing eq.(4) which is the same as (3). To justify the other three boundary conditions, we provide a few examples in the two– and three–particle sectors, instead of giving a general proof.
First we consider the two–particle sector, for example $`P_{BA}(x,x+1)`$. By master equation (2) we have
$$\frac{}{t}P_{BA}(x,x+1)=P_{BA}(x1,x)+P_{BA}(x,x)2P_{BA}(x,x+1).$$
(7)
Using (6), (7) reduces to
$$\frac{}{t}P_{BA}(x,x+1)=P_{BA}(x1,x)2P_{BA}(x,x+1).$$
(8)
This is exactly what we expect, as the source of configuration $`(\mathrm{}BA\mathrm{})`$ is $`(B\mathrm{}A\mathrm{})`$ and its sinks are two configurations $`(\mathrm{}B\mathrm{}A)`$ and $`(\mathrm{}BB\mathrm{}).`$ The second example is $`P_{BB}(x,x+1)`$. Using again the master equation (2) and the boundary condition (5), we obtain
$$\frac{}{t}P_{BB}(x,x+1)=P_{BB}(x1,x)+P_{AB}(x,x+1)+P_{BA}(x,x+1)P_{BB}(x,x+1).$$
(9)
This equation also predicts the true source and sink terms, because $`(\mathrm{}BB\mathrm{})`$ has three sources $`(B\mathrm{}B\mathrm{}),(\mathrm{}AB\mathrm{}),`$ and $`(\mathrm{}BA\mathrm{})`$ and only one sink $`(\mathrm{}B\mathrm{}B).`$ As a three–particle sector example, let us consider the most nontrivial case $`P_{BBB}(x,x+1,x+2).`$ Using (3) and (5), we find
$$\begin{array}{c}\frac{}{t}P_{BBB}(x,x+1,x+2)=P_{BBB}(x1,x+1,x+2)+P_{BBB}(x,x,x+2)+\\ P_{BBB}(x,x+1,x+1)3P_{BBB}(x,x+1,x+2)\\ =P_{BBB}(x1,x+1,x+2)+2P_{BAB}(x,x+1,x+2)+\\ P_{BBA}(x,x+1,x+2)+P_{ABB}(x,x+1,x+2)P_{BBB}(x,x+1,x+2).\end{array}$$
(10)
This is also the true equation, because the configuration $`(\mathrm{}BBB\mathrm{})`$ has five sources namely $`(B\mathrm{}BB\mathrm{}),2(\mathrm{}BAB\mathrm{}),(\mathrm{}BBA\mathrm{})`$, and $`(\mathrm{}ABB\mathrm{})`$ and one sink $`(\mathrm{}BB\mathrm{}B).`$ The reason of appearing the factor 2 in $`(\mathrm{}BAB\mathrm{})`$ is that $`(\mathrm{}BA\mathrm{})`$ can go to $`(\mathrm{}BB\mathrm{})`$ and also $`(\mathrm{}AB\mathrm{})`$ can go to $`(\mathrm{}BB\mathrm{}).`$
### 2.2 The Bethe ansatz solution
We now try to solve the master equation (2) with boundary conditions ( 4)–(6) by Bethe ansatz method. If we define $`\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_1,\mathrm{},x_N)`$ through
$$P_{\alpha _1,\mathrm{},\alpha _N}(x_1,\mathrm{},x_N;t)=e^{ϵ_Nt}\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_1,\mathrm{},x_N),$$
(11)
and substitute (11) in master equation (2) and boundary conditions (4)–(6), we find
$`\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_11,x_2,\mathrm{},x_N)+\mathrm{}+\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_1,x_2,\mathrm{},x_N1)`$ (12)
$`=`$ $`(Nϵ_N)\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_1,x_2,\mathrm{},x_N),`$
and
$`\mathrm{\Psi }_{AA}(x,x)`$ $`=`$ $`\mathrm{\Psi }_{AA}(x,x+1),`$
$`\mathrm{\Psi }_{BB}(x,x)`$ $`=`$ $`\mathrm{\Psi }_{BB}(x,x+1)+\mathrm{\Psi }_{AB}(x,x+1)+\mathrm{\Psi }_{BA}(x,x+1),`$ (13)
$`\mathrm{\Psi }_{AB}(x,x)`$ $`=`$ $`\mathrm{\Psi }_{BA}(x,x)=0.`$
Following , it becomes easier if we use a compact notation as follows: $`𝚿`$ is a tensor of rank $`N`$ with components $`\mathrm{\Psi }_{\alpha _1,\mathrm{},\alpha _N}(x_1,\mathrm{},x_N)`$. Then the boundary conditions (13) can be written as
$$𝚿(\mathrm{},\xi ,\xi ,\mathrm{})=𝐛_{k,k+1}𝚿(\mathrm{},\xi ,\xi +1,\mathrm{}),$$
(14)
where $`𝐛_{k,k+1}`$ is the embedding of $`𝐛`$ (the matrix derived from (13)) in the location $`k`$ and $`k+1`$:
$$\begin{array}{ccc}𝐛_{k,k+1}=\mathrm{𝟏}\mathrm{}& \underset{}{𝐛}& \mathrm{}\mathrm{𝟏},\\ & k,k+1& \end{array}$$
(15)
with
$$𝐛=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 1& 1& 1\end{array}\right).$$
(16)
To solve eq.(12), we write the coordinate Bethe ansatz for $`𝚿`$ in the form:
$$𝚿(x_1,\mathrm{},x_N)=\underset{\sigma }{}𝐀_\sigma e^{i\sigma (𝐩).𝐱},$$
(17)
where $`𝐱`$ and $`𝐩`$ denote the $`N`$–tuples coordinates and momenta, respectively, the summation runs over all the elements of permutation group, and $`𝐀_\sigma `$’s (tensors of rank $`N`$) are coefficients that must be determined from boundary condition (14). Inserting (17) into (12), yields:
$$\underset{\sigma }{}𝐀_\sigma e^{i\sigma (𝐩).𝐱}(e^{i\sigma (p_1)}+\mathrm{}+e^{i\sigma (p_N)}+ϵ_NN)=0,$$
(18)
from which one can find the eigenvalues $`ϵ_N`$ as:
$$ϵ_N=\underset{k=1}{\overset{N}{}}(1e^{ip_k}).$$
(19)
To find the coefficients $`𝐀_\sigma `$, we insert the wavefunction (17) in (14), which yields
$$\underset{\sigma }{}e^{i\underset{jk,k+1}{}\sigma (p_j)x_j+i(\sigma (p_k)+\sigma (p_{k+1}))\xi }\left[(1e^{i\sigma (p_{k+1})}𝐛_{k,k+1})𝐀_\sigma \right]=0.$$
(20)
We note that the first part of the above equation is symmetric with respect to interchange of $`p_k`$ and $`p_{k+1}`$, so if we symmetrize the bracket with respect to this interchange, we obtain
$$(1e^{i\sigma (p_{k+1})}𝐛_{k,k+1})𝐀_\sigma +(1e^{i\sigma (p_k)}𝐛_{k,k+1})𝐀_{\sigma \sigma _k}=0,$$
(21)
where $`\sigma _k`$ is an element of permutation group which only interchange $`p_k`$ and $`p_{k+1},`$ and $`\sigma \sigma _k`$ stands for the product of two group elements $`\sigma `$ and $`\sigma _k`$. Thus we obtain:
$$𝐀_{\sigma \sigma _k}=𝐒_{k,k+1}(\sigma (p_k),\sigma (p_{k+1}))𝐀_\sigma ,$$
(22)
where
$$\begin{array}{ccc}𝐒_{k,k+1}(z_1,z_2)=\mathrm{𝟏}\mathrm{}& \underset{}{S(z_1,z_2)}& \mathrm{}\mathrm{𝟏},\\ & k,k+1& \end{array}$$
(23)
and
$$S(z_1,z_2)=(\mathrm{𝟏}z_1𝐛)^1(\mathrm{𝟏}z_2𝐛).$$
(24)
In the above equations, we have denoted $`e^{ip_k}`$ by $`z_k`$. In this way, we can calculate all the coefficients $`𝐀_\sigma `$’s in term of $`𝐀_1`$ by using eq.(22), and *it seems that* the problem is solved for arbitrary boundary condition $`𝐛`$ matrix. But this is not the case and we should moreover check the consistency of the solutions, which is highly nontrivial and depends on the details of the interaction, i.e. the elements of the $`𝐛`$ matrix .
To see this, let us find two coefficients $`𝐀_{\sigma _1\sigma _2\sigma _1}`$ and $`𝐀_{\sigma _2\sigma _1\sigma _2}.`$ Note that $`\sigma _1\sigma _2\sigma _1`$and $`\sigma _2\sigma _1\sigma _2`$ are equal as the elements of permutation group, that is both of them when act on $`(p_1,p_2,p_3,p_4,\mathrm{})`$ will result in
$$(p_1,p_2,p_3,p_4,\mathrm{})(p_3,p_2,p_1,p_4,\mathrm{}).$$
(25)
So we should have
$$𝐀_{\sigma _1\sigma _2\sigma _1}=𝐀_{\sigma _2\sigma _1\sigma _2}.$$
(26)
Now using (22), we have:
$$\begin{array}{ccc}\hfill 𝐀_{\sigma _1\sigma _2\sigma _3}& =& S_{12}(\sigma _1\sigma _2(p_1),\sigma _1\sigma _2(p_2))𝐀_{\sigma _1\sigma _2}=S_{12}(p_2,p_3)𝐀_{\sigma _1\sigma _2}\hfill \\ & =& S_{12}(p_2,p_3)S_{23}(\sigma _1(p_2),\sigma _1(p_3))𝐀_{\sigma _1}=S_{12}(p_2,p_3)S_{23}(p_1,p_3))𝐀_{\sigma _1}\hfill \\ & =& S_{12}(p_2,p_3)S_{23}(p_1,p_3)S_{12}(p_1,p_2)𝐀_1,\hfill \end{array}$$
(27)
and in the same way ,
$$𝐀_{\sigma _2\sigma _1\sigma _2}=S_{23}(p_1,p_2)S_{12}(p_1,p_3)S_{23}(p_2,p_3)𝐀_1.$$
(28)
Therefore, (26) yields:
$$S_{12}(p_2,p_3)S_{23}(p_1,p_3)S_{12}(p_1,p_2)=S_{23}(p_1,p_2)S_{12}(p_1,p_3)S_{23}(p_2,p_3),$$
(29)
which is the familiar Quantum Yang–Baxter equation. Therefore , we must check if the S–matrices defined in (23) and (24) , with $`𝐛`$ from eq.(16) , satisfy (29) or not.
Using (16) , it can be shown that (24) is equal to
$$S(z,w)=\frac{1}{z1}\left[\begin{array}{cccc}1w& 0& 0& 0\\ 0& 1z& 0& 0\\ 0& 0& 1z& 0\\ 0& zw& zw& 1w\end{array}\right],$$
(30)
and if we define $`z=e^{ip_1},w=e^{ip_2},`$ and $`t=e^{ip_3}`$, the eq. (29) can be written as
$$(S(w,t)\mathrm{𝟏})(\mathrm{𝟏}S(z,t))(S(z,w)\mathrm{𝟏})=(\mathrm{𝟏}S(z,w))(S(z,t)\mathrm{𝟏})(\mathrm{𝟏}S(w,t)).$$
(31)
Now it is not too difficult to show that the matrix (30) really *satisfy* eq.(31), and therefore the solutions of interactions (1) are wavefunctions (17) with the coefficients that can be found by eq.(22).
## 3 The two–particle conditional probabilities
In this section we want to calculate the two–particle conditional probabilities $`P(\alpha _1,\alpha _2,x_1,x_2;t|\beta _1,\beta _2,y_1,y_2;0),`$ which is the probability of finding $`\alpha _1`$ at site $`x_1`$ and particle $`\alpha _2`$ at site $`x_2`$ at time $`t`$, if at time $`t=0`$ we have the particle $`\beta _1`$ at site $`y_1`$ and particle $`\beta _2`$ at site $`y_2`$. As has been discussed in and , we should take a linear combination of eigenfunctions $`P(x_1,x_2)`$ (from (11) and (17)) with suitable coefficients, to find these two–particle conditional probabilities. Therefore,
$$\begin{array}{c}\left(\begin{array}{c}P_{AA}\\ P_{AB}\\ P_{BA}\\ P_{BB}\end{array}\right)(𝐱;t|\beta ,𝐲;0)=\hfill \\ =f(p_1,p_2)e^{ϵ_2t}𝚿(x_1,x_2)𝑑p_1𝑑p_2\hfill \\ =\frac{1}{(2\pi )^2}e^{ϵ_2t}e^{i𝐩.𝐲}\left\{\left(\begin{array}{c}a\\ b\\ c\\ d\end{array}\right)e^{i(p_1x_1+p_2x_2)}+S_{12}(p_1,p_2)\left(\begin{array}{c}a\\ b\\ c\\ d\end{array}\right)e^{i(p_2x_1+p_1x_2)}\right\}𝑑p_1𝑑p_2.\hfill \end{array}$$
(32)
In the above expansion $`P_{\alpha _1\alpha _2}(𝐱;t|\beta ,𝐲;0)`$ is $`P(\alpha _1,\alpha _2,x_1,x_2;t|\beta _1,\beta _2,y_1,y_2;0),`$ $`f(p_1,p_2)`$ is the coefficient of expansion which in the second equality we choose it $`\frac{1}{(2\pi )^2}e^{i𝐩.𝐲},`$ and $`ϵ_2=2e^{ip_1}e^{ip_2}`$ (see (19)). $`𝚿(x_1,x_2)`$ is the two–particle wave function where from (17) and (22) we obtain
$$\begin{array}{ccc}\hfill 𝚿(x_1,x_2)& =& 𝐀_1e^{i(p_1x_1+p_2x_2)}+𝐀_{\sigma _1}e^{i\sigma _1(𝐩).𝐱}\hfill \\ & =& 𝐀_1e^{i(p_1x_1+p_2x_2)}+S_{12}(p_1,p_2)𝐀_1e^{i(p_1x_2+p_2x_1)}\hfill \end{array}.$$
(33)
In the two–particle sector, $`𝐀_1`$ is a 4–column vector whose components must be determined by initial conditions, and $`S_{12}(p_1,p_2)`$ can be read from (30):
$$S_{12}(p_1,p_2)=\left(\begin{array}{cccc}s^{}& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& s& s& s^{}\end{array}\right),$$
(34)
with
$`s^{}`$ $`=`$ $`{\displaystyle \frac{1e^{ip_2}}{e^{ip_1}1}},`$
$`s`$ $`=`$ $`{\displaystyle \frac{e^{ip_1}e^{ip_2}}{e^{ip_1}1}}.`$ (35)
Let us first calculate eq.(32) irrespective of initial conditions , that is for arbitrary $`a,b,c,d.`$ Substituting (34) in (32), we find:
$$\left(\begin{array}{c}P_{AA}\\ P_{AB}\\ P_{BA}\\ P_{BB}\end{array}\right)(𝐱;t|\beta ,𝐲;0)=\left(\begin{array}{c}a(F_1(t)+F_4(t))\\ b(F_1(t)F_2(t))\\ c(F_1(t)F_2(t))\\ d(F_1(t)+F_4(t))+(b+c)F_3(t)\end{array}\right),$$
(36)
in which
$`F_1(t)`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle e^{ϵ_2t}e^{i𝐩.(𝐱𝐲)}𝑑p_1𝑑p_2},`$ (37)
$`F_2(t)`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle e^{ϵ_2t}e^{i(\stackrel{~}{𝐩}.𝐱𝐩.𝐲)}𝑑p_1𝑑p_2},`$ (38)
$`F_3(t)`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle e^{ϵ_2t}\frac{e^{ip_1}e^{ip_2}}{e^{ip_1}1}e^{i(\stackrel{~}{𝐩}.𝐱𝐩.𝐲)}𝑑p_1𝑑p_2},`$ (39)
$`F_4(t)`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle e^{ϵ_2t}\frac{1e^{ip_2}}{e^{ip_1}1}e^{i(\stackrel{~}{𝐩}.𝐱𝐩.𝐲)}𝑑p_1𝑑p_2},`$ (40)
where in the above equations we have suppressed the $`𝐱`$ and $`𝐩`$ dependence of $`F_i`$’s for simplicity and $`\stackrel{~}{𝐩}=(p_2,p_1).`$ To avoid the singularity in $`s`$ and $`s^{}`$, we set $`p_1p_1+i\epsilon `$ , and then by some simple contour integration we find
$`F_1(0)`$ $`=`$ $`\delta _{x_1,y_1}\delta _{x_2y_2},`$
$`F_2(0)`$ $`=`$ $`F_3(0)=F_4(0)=0,`$ (41)
and at $`t0,`$
$$\begin{array}{ccc}\hfill F_1(t)& =& e^{2t}\frac{t^{x_1y_1}}{(x_1y_1)!}\frac{t^{x_2y_2}}{(x_2y_2)!},\hfill \\ \hfill F_2(t)& =& e^{2t}\frac{t^{x_2y_1}}{(x_2y_1)!}\frac{t^{x_1y_2}}{(x__1y_2)!},\hfill \\ \hfill F_3(t)& =& e^{2t}\left\{\frac{t^{x_1y_2+1}}{(x_1y_2+1)!}\underset{k=0}{\overset{\mathrm{}}{}}\frac{t^{x_2y_1+k}}{(x_2y_1+k)!}\frac{t^{x_1y_2}}{(x_1y_2)!}\underset{k=0}{\overset{\mathrm{}}{}}\frac{t^{x_2y_1+k+1}}{(x_2y_1+k+1)!}\right\},\hfill \\ \hfill F_4(t)& =& e^{2t}\left\{\frac{t^{x_1y_2+1}}{(x_1y_2+1)!}\frac{t^{x_1y_2}}{(x_1y_2)!}\right\}\underset{k=0}{\overset{\mathrm{}}{}}\frac{t^{x_2y_1+k}}{(x_2y_1+k)!}.\hfill \end{array}$$
(42)
Now we can study the different initial conditions.
a) Case of $`\beta _1=\beta _2=A`$
If at $`t=0,`$ both particles are $`A`$ type, then our initial condition is
$$\left(\begin{array}{c}P_{AA}\\ P_{AB}\\ P_{BA}\\ P_{BB}\end{array}\right)(𝐱;0|A,A,𝐲;0)=\left(\begin{array}{c}\delta _{x_1,y_1}\delta _{x_2,y_2}\\ 0\\ 0\\ 0\end{array}\right).$$
(43)
Using (36) and (41) we find
$$a=1,b=c=d=0,$$
(44)
and therefore
$$P_{AA}(𝐱;t|A,A,𝐲;0)=F_1(t)+F_4(t),$$
(45)
and all other $`P`$’s are zero. Note that eq.(45) is exactly the same conditional probability that has been found in for simple ASEP model, as we expect.
b) Case of $`\beta _1=A,\beta _2=B`$
In this case the only non–zero element of conditional probability, at $`t=0,`$ is $`P_{AB}=\delta _{x_1,y_1}\delta _{x_2,y_2}.`$ Therefore one finds
$$b=1,a=c=d=0,$$
(46)
and therefore
$`P_{AB}(𝐱;t|A,B,𝐲;0)`$ $`=`$ $`F_1(t)F_2(t),`$
$`P_{BB}(𝐱;t|A,B,𝐲;0)`$ $`=`$ $`F_3(t),`$ (47)
and $`P_{AA}=P_{BA}=0,`$ which is consistent with our processes (1). It can be also checked that the conservation of probability holds,
$$\underset{x_2=y_2}{\overset{\mathrm{}}{}}\underset{x_1=y_1}{\overset{x_21}{}}(P_{AB}+P_{BB})(𝐱;t|A,B,𝐲;0)=1,$$
(48)
for arbitrary $`y_1,y_2`$ and $`t.`$
c) Case of $`\beta _1=B,\beta _2=A`$
In this case, the final result is :
$`P_{BA}(𝐱;t|B,A,𝐲;0)`$ $`=`$ $`F_1(t)F_2(t),`$
$`P_{BB}(𝐱;t|B,A,𝐲;0)`$ $`=`$ $`F_3(t)`$ (49)
and $`P_{AA}=P_{AB}=0.`$
d) Case of $`\beta _1=\beta _2=B`$
In this case the final result is the same as the case 3.1, as we expect,
$$P_{BB}(𝐱;t|B,B,𝐲;0)=F_1(t)+F_4(t),$$
(50)
and $`P_{AA}=P_{AB}=P_{BA}=0.`$
Another interesting quantity that can be calculated, is the long time behavior of this functions . The only nontrivial case, is the case (b) (or equivalently (c)). We expect that if at $`t=0`$ , we have $`A`$ and $`B`$ particles, (one healthy and one diseased individuals), the healthy one becomes diseased finally, or we have two $`B`$ particles finally. In other words, we expect that (in case (b)),
$$\underset{x_2=y_2}{\overset{\mathrm{}}{}}\underset{x_1=y_1}{\overset{x_21}{}}P_{AB}(𝐱;t\mathrm{}|A,B,𝐲;0)0,$$
(51)
and
$$\underset{x_2=y_2}{\overset{\mathrm{}}{}}\underset{x_1=y_1}{\overset{x_21}{}}P_{BB}(𝐱;t\mathrm{}|A,B,𝐲;0)1.$$
(52)
After some calculations , one can show that
$$\underset{x_2=y_2}{\overset{\mathrm{}}{}}\underset{x_1=y_1}{\overset{x_{21}}{}}P_{AB}(𝐱;t|A,B,𝐲;0)=e^{2t}\left[2\underset{m=1}{\overset{y_2y_11}{}}I_m(2t)+I_0(2t)+I_{y_2y_1}(2t)\right],$$
(53)
where $`I_n(x)`$ is the $`n`$-th order Bessel function of the first kind :
$$I_n(x)=\underset{k=0}{\overset{\mathrm{}}{}}\frac{(x/2)^{n+2k}}{k!(n+k)!}.$$
(54)
Now at $`x\mathrm{}`$, we have
$$I_n(x)\frac{e^x}{\sqrt{2\pi x}},$$
(55)
therefore,
$$\underset{x_2=y_2}{\overset{\mathrm{}}{}}\underset{x_1=y_1}{\overset{x_21}{}}P_{AB}(𝐱;t\mathrm{}|A,B,𝐲;0)\frac{M}{\sqrt{4\pi t}},$$
(56)
which goes to zero. $`M`$ is the number of the $`I_n(2t)`$’s in the left hand side of (53). Using (48), we see that both limits in (51) and (52) are satisfied.
## 4 The class of models
Now we want to find the class of all two–species integrable models which have the four properties introduced in the introduction section. If we look at the preceding sections, we notice that all the information about the model are abbreviated in the $`𝐛`$ matrix (16), because this matrix comes from the boundary conditions (4) to (6), and the latter induce our interactions. We also note that the sum of each column of $`𝐛`$ matrix is one. Now we claim that each $`𝐛`$ matrix which has the following properties:
1– the non–diagonal elements are one or zero,
2–the sum of elements in each column is one,
represents a model that its interaction(s) can be induced by the master equation (2) plus the boundary condition(s) which can be read from $`𝐛.`$
The reason of the first requirement is that the non–diagonal elements are the sources of our reactions, as can be seen from the example solved in section 2.1, and if we want all the reactions to occur with equal rate one, the pre–factors of all source terms must be one (or zero, if we don’t want the corresponding source terms). Note that we should take all the rates equal to each other, otherwise for the reactions we are interested in (i.e. those in which, particles change their type), the factorization (11) will not yield the time independent boundary condition(s), which is wrong.
The reason for the second requirement lies in the conservation of probability. Suppose that in the first column of $`𝐛`$, for example, the sum of the non–diagonal elements is $`m`$ and the diagonal element is $`n`$. So we have $`m`$ possible interactions each can be a sink for $`AA`$. Therefore if our configuration is $`(\mathrm{}AA\mathrm{}),`$ we must have $`m+1`$ sinks, one sink for diffusion $`(\mathrm{}AA\mathrm{})(\mathrm{}A\mathrm{}A),`$ and $`m`$ sinks for reactions: $`(\mathrm{}AA\mathrm{})(\mathrm{}\alpha \beta \mathrm{}).`$ Now we consider the master equation for this process:
$`{\displaystyle \frac{}{t}}P_{AA}(x,x+1)`$ $`=`$ $`P_{AA}(x1,x+1)+P_{AA}(x,x)2P_{AA}(x,x+1)`$ (57)
$`=`$ $`P_{AA}(x1,x+1)+{\displaystyle \underset{1}{\overset{m^{}}{}}}P_{\alpha \beta }(x,x+1)(2n)P_{AA}(x,x+1).`$
in which we have supposed that the $`m^{}`$ elements of the first row of $`𝐛`$ (besides b<sub>11</sub>) are different from zero and therefore the corresponding probabilities appear in $`P_{AA}(x,x)`$. Now as the number of sinks must be $`m+1,`$ so $`2n=m+1`$ which yields $`n+m=1.`$ Therefore the sum of the elements of the first column of $`𝐛`$ must be one. By the same reasoning, it is true for other columns.
In this way we have $`2^{12}=4096`$ possibilities for matrices $`𝐛`$ (there are 12 non–diagonal elements, each can be one or zero), each plus master equation (2) can model a reaction–diffusion process. But as we have seen in reaction (2), these $`𝐛`$’s must be consistent with the QYBE (31). Therefore the domain of $`𝐛`$’s is much smaller. So it is sufficient to check which of these $`𝐛`$’s (or more carefully, the S–matrices that are constructed by these $`𝐛`$’s from eq.(24)) satisfy (31). Using a symbolic manipulator (e.g. MAPLE), we found that there are 42 different $`𝐛`$’s that satisfy eq.(31). By a closer inspection of these matrices it is observed that 14 one of them can be obtained from the others by interchanging $`AB`$, so they do not represent any new physical interactions. The 42-14=28 $`𝐛`$’s (interactions) are as follows:
$$\begin{array}{cc}𝐛_1=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right),\mathrm{pure}\mathrm{diffusion}& 𝐛_2=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 1& 1& 0\\ 0& 0& 0& 1\end{array}\right),ABBA\end{array}$$
$$\begin{array}{cc}𝐛_3=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 1& 1\end{array}\right),BABB& 𝐛_4=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 1\end{array}\right),ABBB\end{array}$$
$$\begin{array}{cc}𝐛_5=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 1& 1& 1\end{array}\right),\begin{array}{c}AB\\ BA\end{array}\}BB& 𝐛_6=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 1& 1& 0\\ 1& 0& 0& 1\end{array}\right),\begin{array}{c}ABBA\\ AABB\end{array}\end{array}$$
$$\begin{array}{cc}𝐛_7=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 1& 0& 1& 0\\ 0& 1& 0& 1\end{array}\right),\begin{array}{c}AABA\\ ABBB\end{array}& 𝐛_8=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 1& 1& 1& 0\\ 0& 0& 0& 1\end{array}\right),\begin{array}{c}AA\\ AB\end{array}\}BA\end{array}$$
$$\begin{array}{cc}𝐛_9=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 1& 1& 0\\ 0& 0& 0& 0\\ 1& 0& 0& 1\end{array}\right),\begin{array}{c}BAAB\\ AABB\end{array}& 𝐛_{10}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 1& 1& 0& 0\\ 0& 0& 1& 1\\ 0& 0& 0& 0\end{array}\right),\begin{array}{c}AAAB\\ BBBA\end{array}\end{array}$$
$$\begin{array}{cc}𝐛_{11}=\left(\begin{array}{cccc}1& 0& 1& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 1& 0& 1\end{array}\right),\begin{array}{c}BAAA\\ ABBB\end{array}& 𝐛_{12}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 1& 1& 1& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 1\end{array}\right),\begin{array}{c}AA\\ BA\end{array}\}AB\end{array}$$
$$\begin{array}{cc}𝐛_{13}=\left(\begin{array}{cccc}1& 1& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 1& 1\end{array}\right),\begin{array}{c}ABAA\\ BABB\end{array}\hfill & 𝐛_{14}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 1& 1& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 1& 1\end{array}\right),\begin{array}{c}AAAB\\ BABB\end{array}\hfill \end{array}$$
$$\begin{array}{cc}𝐛_{15}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 1& 0& 1\\ 1& 0& 1& 0\\ 0& 0& 0& 0\end{array}\right),\begin{array}{c}BBAB\\ AABA\end{array}\hfill & 𝐛_{16}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 1& 1& 1\\ 0& 1& 0& 0\end{array}\right),\begin{array}{c}ABBB\\ \begin{array}{c}BB\\ AB\end{array}\}BA\end{array}\hfill \end{array}$$
$$\begin{array}{cc}𝐛_{17}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ 1& 1& 1& 1\end{array}\right),\begin{array}{c}AA\\ AB\\ BA\end{array}\}BB\hfill & 𝐛_{18}=\left(\begin{array}{cccc}1& 0& 1& 0\\ 0& 0& 0& 0\\ 0& 1& 0& 1\\ 0& 0& 0& 0\end{array}\right),\begin{array}{c}BAAA\\ \begin{array}{c}AB\\ BB\end{array}\}BA\end{array}\hfill \end{array}$$
$$\begin{array}{cc}𝐛_{19}=\left(\begin{array}{cccc}1& 1& 1& 0\\ 0& 0& 0& 1\\ 0& 0& 0& 1\\ 0& 0& 0& 1\end{array}\right),\begin{array}{c}\begin{array}{c}AB\\ BA\end{array}\}AA\\ BB\{\begin{array}{c}AB\\ BA\end{array}\end{array}\hfill & 𝐛_{20}=\left(\begin{array}{cccc}0& 0& 1& 0\\ 0& 0& 0& 1\\ 1& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right),\begin{array}{c}BAAA\\ BBAB\\ AABA\\ ABBB\end{array}\hfill \end{array}$$
$$\begin{array}{cc}𝐛_{21}=\left(\begin{array}{cccc}0& 1& 0& 0\\ 1& 0& 0& 0\\ 0& 0& 0& 1\\ 0& 0& 1& 0\end{array}\right),\begin{array}{c}ABAA\\ AAAB\\ BBBA\\ BABB\end{array}\hfill & 𝐛_{22}=\left(\begin{array}{cccc}0& 0& 0& 1\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 1& 0& 0& 0\end{array}\right),\begin{array}{c}BBAA\\ BAAB\\ ABBA\\ AABB\end{array}\hfill \end{array}$$
$$\begin{array}{cc}𝐛_{23}=\left(\begin{array}{cccc}0& 0& 1& 0\\ 1& 1& 1& 0\\ 0& 1& 1& 1\\ 0& 1& 0& 0\end{array}\right),\begin{array}{c}BAAA\\ ABBB\\ \begin{array}{c}BA\\ AA\end{array}\}AB\\ \begin{array}{c}BB\\ AB\end{array}\}BA\end{array}\hfill & 𝐛_{24}=\left(\begin{array}{cccc}1& 1& 0& 1\\ 0& 0& 0& 1\\ 1& 0& 0& 0\\ 1& 0& 1& 1\end{array}\right),\begin{array}{c}\begin{array}{c}BB\\ AB\end{array}\}AA\\ \begin{array}{c}AA\\ BA\end{array}\}BB\\ BBAB\\ AABA\end{array}\hfill \end{array}$$
$$𝐛_{25}=\left(\begin{array}{cccc}1& 0& 1& 1\\ 0& 1& 1& 1\\ 1& 1& 1& 0\\ 1& 1& 0& 1\end{array}\right),\begin{array}{c}\begin{array}{c}BB\\ BA\end{array}\}AA\\ \begin{array}{c}BB\\ BA\end{array}\}AB\end{array},\begin{array}{c}\begin{array}{c}AA\\ AB\end{array}\}BA\\ \begin{array}{c}AA\\ AB\end{array}\}BB\end{array}$$
$$𝐛_{26}=\left(\begin{array}{cccc}1& 1& 0& 1\\ 1& 1& 1& 0\\ 0& 1& 1& 1\\ 1& 0& 1& 1\end{array}\right),\begin{array}{c}\begin{array}{c}BB\\ AB\end{array}\}AA\\ \begin{array}{c}AA\\ BA\end{array}\}AB\end{array},\begin{array}{c}\begin{array}{c}BB\\ AB\end{array}\}BA\\ \begin{array}{c}AA\\ BA\end{array}\}BB\end{array}$$
$$𝐛_{27}=\left(\begin{array}{cccc}1& 1& 1& 0\\ 1& 1& 0& 1\\ 1& 0& 1& 1\\ 0& 1& 1& 1\end{array}\right),\begin{array}{c}\begin{array}{c}BA\\ AB\end{array}\}AA\\ \begin{array}{c}BB\\ AA\end{array}\}AB\end{array},\begin{array}{c}\begin{array}{c}AA\\ BB\end{array}\}BA\\ \begin{array}{c}AB\\ BA\end{array}\}BB\end{array}$$
$$𝐛_{28}=\left(\begin{array}{cccc}2& 1& 1& 1\\ 1& 2& 1& 1\\ 1& 1& 2& 1\\ 1& 1& 1& 2\end{array}\right),\begin{array}{c}\begin{array}{c}AB\\ BB\\ BA\end{array}\}AA\\ \begin{array}{c}AA\\ BB\\ BA\end{array}\}AB\end{array},\begin{array}{c}\begin{array}{c}BB\\ AA\\ AB\end{array}\}BA\\ \begin{array}{c}BA\\ AA\\ AB\end{array}\}BB\end{array}$$
It should be mentioned that in above, the reaction processes of each $`𝐛`$ have been given only and the diffusion processes (which exist in all cases) have been suppressed. Also note that $`𝐛_1`$ is the pure diffusion process of , and $`𝐛_5`$ is nothing but eq. (16).
In all the above cases the probabilities, $`𝚿`$ and $`𝐀_\sigma `$ are given by (11), (17) and (22), respectively. Obviously, $`S(z,w)`$ must be calculated from eq. (24) for each case, and then the calculations of section 3 can be repeated for them.
Acknowledgement
We would like to thank V. Karimipour for useful discussions, and R. Faraji–Dana for helping us in computer programming. M. Alimohammadi would also like to thank the research council of the University of Tehran for partial financial support.
|
warning/0002/math-ph0002050.html
|
ar5iv
|
text
|
# 1 Estimation; the Cramer-Rao inequality
## 1 Estimation; the Cramer-Rao inequality
Let $`\rho _\eta (x)`$ be a probability density, depending on a parameter $`\eta R`$. The Fisher information of $`\rho _\eta `$ is defined to be
$$G:=\rho _\eta (x)\left(\frac{\mathrm{log}\rho _\eta (x)}{\eta }\right)^2𝑑x.$$
(1)
We note that this is the variance of the random variable $`Y=\mathrm{log}\rho _\eta /\eta `$, which has mean zero. $`G`$ is associated with the family $`=\{\rho _\eta \}`$ of distributions, rather than any one of them. This concept arises in the theory of estimation as follows. Let $`X`$ be a random variable whose distribution is believed or hoped to be one of those in $``$. We estimate the value of $`\eta `$ by measuring $`X`$ independently $`m`$ times, getting the data $`x_1,\mathrm{},x_m`$. An estimator $`f`$ is a function of $`(x_1,\mathrm{},x_m)`$ that is used for this estimate. So $`X`$ is a function of $`m`$ independent copies of $`X`$, and so is a random variable. To be useful, the estimator must be independent of $`\eta `$, which we do not (yet) know. We say that an estimator is unbiased if its mean is the desired parameter; it is usual to take $`f`$ as a function of $`X`$ and to regard $`f(x_i)`$, $`i=1,\mathrm{},m`$ as samples of $`f`$. Then the condition that $`f`$ is unbiased becomes
$$\rho _\eta .f:=\rho _\eta (x)f(x)𝑑x=\eta .$$
(2)
We use the notation $`\rho .f`$ for the expectation of $`f`$ in the state $`\rho `$. A good estimator should also have only a small chance of being far from the correct value, which is its mean if it is unbiased. This chance is measured by the variance. Fisher stated, and Rao and Cramer proved, that the variance V of an unbiased estimator $`f`$ obeys the inequality $`VG^1`$. For the proof, differentiate eq. (2) w. r. t. $`\eta `$ to get
$$\frac{\rho _\eta (x)}{\eta }f(x)𝑑x=1,$$
(3)
which can be written as
$$Y(x)(f(x)\eta )\rho _\eta (x)𝑑x=\left(\frac{\mathrm{log}\rho }{\eta }\right)\left(f(x)\eta \right)\rho _\eta (x)𝑑x=1.$$
(4)
We note that this is the correlation of $`Y`$ and $`f`$, so the covariance matrix becomes
$$\left(\begin{array}{cc}G& 1\\ 1& V\end{array}\right).$$
(5)
This is positive semi-definite, giving the result.$`\mathrm{}`$
If we do $`N`$ independent measurements of the estimator, and average them, we improve the inequality to $`VG^1/N`$. This inequality expresses that, given the family $`\rho _\eta `$, there is a limit to the reliability with which we can estimate $`\eta `$. Fisher termed $`V/G^1`$ the efficiency of the estimator $`f`$. Equality in the Schwarz inequality occurs if and only if the two functions are proportional. Let $`\xi /\eta `$ denote the factor of proportionality. Then the optimal estimator occurs when
$$\mathrm{log}\rho _\eta (x)=\xi /\eta (f(x)\eta )d\eta .$$
(6)
Doing the integral, and adjusting the integration constant by normalisation, leads to
$$\rho _\eta (x)=Z^1\mathrm{exp}\{\xi f(x)\}$$
(7)
which is the ‘exponential family’.
This can be generalised to any $`n`$-parameter manifold $`=\{\rho _\eta \}`$ of distributions, $`\eta =(\eta _1,\mathrm{},\eta _n)`$ with $`\eta R^n`$. Suppose we have unbiased estimators $`(f_1,\mathrm{},f_n)`$, with covariance matrix $`V`$. Fisher introduced the information matrix
$$G^{ij}=\rho _\eta (x)\frac{\mathrm{log}\rho _\eta (x)}{\eta _i}\frac{\mathrm{log}\rho _\eta (x)}{\eta _j}𝑑x.$$
(8)
We note that $`Y^j:=\mathrm{log}\rho /\eta _j`$ is a random variable with zero mean, and that $`G^{ij}`$ is its covariance matrix. Rao remarked that $`G^{ij}`$ provides a Riemannian metric for $``$. We now derive the analogue of the inequality when $`n>1`$. Put $`V_{ij}=\rho _\eta .[(f_i\eta _i)(f_j\eta _j)]`$, the covariance matrix of $`\{f_i\}`$. Differentiate the condition for being unbiased,
$$\rho _\eta (x)f_i(x)𝑑x=\eta _i$$
(9)
with respect to $`\eta _j`$, and rearrange as above, to get
$$\rho _\eta (x)Y^i(x)(f_j(x)\eta _j)𝑑x=\delta _{ij}.$$
(10)
This is the correlation between $`Y^i`$ and $`f_j`$. The covariance matrix of the $`2n`$ random variables $`Y^i,f_j`$ therefore is
$$\left(\begin{array}{cc}G& I\\ I& V\end{array}\right).$$
(11)
This is therefore a positive semi-definite matrix. If it is not definite, it has zero as an eigenvalue, which leads to $`GV=I`$, and the manifold must be the exponential family, as before. If it is definite, so is its inverse, which is found to be
$$\left(\begin{array}{cc}\left(GV^1\right)^1& G^1\left(VG^1\right)^1\\ V^1\left(GV^1\right)^1& \left(VG^1\right)^1\end{array}\right).$$
(12)
It follows that the leading submatrices $`(GV^1)^1`$ and $`(VG^1)^1`$ are positive definite, and thus so are their inverses. It follows that we get the matrix inequality $`VG^1`$.
## 2 Entropy methods, exponential families
Gibbs knew that the state of maximum entropy, given the mean energy, is the canonical state. More generally, let $`\mathrm{\Omega }`$ be a countable sample space, and let $`\mathrm{\Sigma }`$ denote the set of probabilities (or states) on $`\mathrm{\Omega }`$. Let $`f_1,\mathrm{},f_n`$ be $`n`$ linearly independent random variables, whose means we can measure. We want to find the ‘best’ choice for the the state, given these means. The least prejudiced choice of $`\rho `$ (Jaynes) is to maximise the entropy $`S`$ subject to the $`n+1`$ constraints given by normalisation and the means of $`f_j,j=1,\mathrm{},n`$. We use $`\lambda ,\xi ^j`$ as Lagrange multipliers; then we must maximise
$$\underset{\omega \mathrm{\Omega }}{}\rho (\omega )\mathrm{log}\rho (\omega )\lambda \underset{\omega }{}\rho (\omega )\underset{j=1}{\overset{n}{}}\xi ^j\rho (\omega )f_j(\omega )$$
by varying $`\rho (\omega )`$ subject to no constraints. We get
$$\rho _\xi (\omega )=Z^1\mathrm{exp}\{\underset{j}{}\xi ^jf_j(\omega )\}\text{ where }Z=\underset{\omega }{}\mathrm{exp}\{\underset{j}{}\xi ^jf_j(\omega )\}.$$
(13)
These make up the exponential manifold M determined by $`:=\mathrm{Span}\{f_1,\mathrm{},f_n\}`$ and parametrised by $`\xi ^1,\mathrm{},\xi ^n`$; these are called the canonical coordinates on $``$, which has dimension $`n`$. At least one, say $`f_1`$, must be bounded below, to ensure $`Z<\mathrm{}`$ holds for some $`\xi `$.
The $`\xi ^j`$ are determined by the given expectation values by the conditions $`\rho _\xi .f_j=\eta _j`$, $`j=1,\mathrm{},n`$. The $`\eta _j`$ are thus also coordinates for the manifold (the mixture coords.) It is easy to show that
$$\eta _j=\frac{\mathrm{\Psi }}{\xi ^j},j=1,\mathrm{},n;V_{jk}=\frac{\eta _j}{\xi ^k},j,k=1,\mathrm{},m,$$
(14)
where $`\mathrm{\Psi }=\mathrm{log}Z`$, and that $`\mathrm{\Psi }`$ is a convex function of $`\xi ^j`$. The Legendre dual to $`\mathrm{\Psi }`$ is $`\mathrm{\Psi }\xi ^i\eta _i`$ and this is the entropy $`S=\rho .\mathrm{log}\rho `$. The dual relations are
$$\xi ^j=\frac{S}{\eta _j}G^{jk}=\frac{\xi ^j}{\eta _k}.$$
(15)
By the rule for Jacobians, $`V`$ and $`G`$ are mutual inverses. Therefore, the method of maximum entropy leads to the exponential family, which allows the optimisation of the Cramer-Rao bound, and gives us estimators of 100% efficiency.
## 3 Manifolds modelled by Orlicz spaces
Pistone and Sempi have developed a version of information geometry, which does not depend on a choice of $``$, the span of a finite number of estimators. Let $`(\mathrm{\Omega },\mu )`$ be measure space and let $``$ be the set of all probability measures that are equivalent to $`\mu `$; such a measure is determined by its Radon-Nikodym derivative $`\rho `$ relative to $`\mu `$. The topology on $``$ is not given by the $`L^1`$-distance, but by an Orlicz norm.
Given $`\rho `$, the Cramer class at $`\rho `$ is the set of all random variables $`X`$ on $`(\mathrm{\Omega },\mu )`$ such that the moment-generating function
$$\widehat{X}_\rho (t):=e^{tX}\rho 𝑑\mu $$
(16)
is finite in a ’hood of the origin. This is enough to ensure that it is analytic in an interval about $`t=0`$. The Cramer class $`C_\rho `$ at a point $`\rho `$ in $``$ is furnished with the Luxemburg norm
$$X_\rho =inf\{r>0:E_\rho \left[\mathrm{cosh}\left(\frac{u}{r}\right)1\right]1\}.$$
(17)
The Cramer class $`C`$ at $`\rho `$ is an Orlicz space, and so is a Banach space with this norm. The centred Cramer class $`C(0)`$ is defined as the subset of $`C`$ at $`\rho `$ with zero mean in the state $`\rho `$; this is a closed subspace. A sufficiently small ball in the quotient Banach space $`C/C(0)`$ then parametrises a ’hood of $`\rho `$, and can be identified with the tangent space at $`\rho `$; namely, the ’hood contains those points $`\sigma `$ of $``$ such that
$$\sigma =Z^1e^X\rho \text{for some }XC.$$
(18)
where $`Z`$ is a normalising factor. Pistone and Sempi show that the bilinear form
$$G(X,Y)=E_\rho \left[XY\right]$$
(19)
is a Riemannian metric on the tangent space $`C/C_0`$, thus generalising the Fisher-Rao theory.
This theory is called non-parametric estimation theory, because we do not limit the distributions to those specified by a finite number of parameters, but allow any ‘shape’ for the density $`\rho `$. It is this construction that we take over to the quantum case, except that the spectrum is discrete and the distributions are not always equivalent.
## 4 Efron, Dawid and Amari
A Riemannian metric $`G`$, eq. (15) gives us a notion of parallel transport, namely, that given by the Levi-Civita affine connection. Recall that an affine map, $`U`$ (acting on the right) from one vector space $`𝒯_1`$ to another, $`𝒯_2`$, is one that obeys
$$(\lambda XU+(1\lambda )YU)=\lambda XU+(1\lambda )YU,\text{ for all }X,Y𝒯_1\text{ and all }\lambda [0,1].$$
(20)
The same definition works on an affine space, that is, a convex subset of a vector space. This leads to the concept of an affine connection.
Let $``$ be a manifold and denote by $`T_\rho `$ the tangent space at $`\rho `$. Consider an affine map $`U_\gamma (\rho ,\sigma ):T_\rho T_\sigma `$ defined for each pair of points $`\rho ,\sigma `$ and each continuous path $`\gamma `$ in the manifold starting at $`\rho `$ and ending at $`\sigma `$. Let $`\rho ,\sigma `$ and $`\tau `$ be any three points and $`\gamma _1`$ a path from $`\rho `$ to $`\sigma `$, and $`\gamma _2`$ any path from $`\sigma `$ to $`\tau `$.
###### Definition 1
We say that $`U`$ is an affine connection, if $`U_{\mathrm{}}=Id`$ and
$$U_{\gamma _1\gamma _2}=U_{\gamma _1}U_{\gamma _2}.$$
(21)
Let $`X`$ be a tangent vector at $`\rho `$; we call $`XU_{\gamma _1}`$ the parallel transport of $`X`$ to $`\sigma `$, along the path $`\gamma _1`$.
We also require $`U`$ to be smooth in $`\rho `$ in a ’hood of the point $`\rho `$, when we identify a ball in the tangent space with part of the manifold by the exponential map. In physics it is usually the differential of $`U`$ along a specified direction that is called ‘affine connection’. Equivalently, a connection defines a covariant derivative of a vector field on the manifold:
$$_YX:=d/dtXU_\gamma (\rho ,\gamma (t))|_{t=0}$$
(22)
where $`\{\gamma (t)\},\mathrm{\hspace{0.33em}0}t1`$ is any path from $`\rho `$ to $`\sigma `$, which starts at $`\rho `$ in the direction $`YT_\rho `$. This is designed to convert vector fields to tensor fields. Conversely, a covariant derivative defines a connection. This concept allows us to specify that two tangent vectors to the manifold at points $`\rho `$ and $`\sigma `$ are parallel if the parallel transport (along a specified curve) of one from $`\rho `$ to $`\sigma `$ is proportional to the other. A geodesic is a self-parallel curve on $``$: the tangent vectors to the curve at different points are parallel, when transported along the curve. Geodesics relative to the Levi-Civita connection are lines of minimal length, as measured by the metric.
Estimation theory might be considered geometrically as follows. For theoretical reasons, we expect the distribution of a random variable to lie on a submanifold $`_0`$ of states. The data give us a histogram, which is a distribution, but not a pretty one. We seek the point on $`_0`$ that is ‘closest’ to the data. Suppose that the sample space is $`\mathrm{\Omega }`$, with $`|\mathrm{\Omega }|<\mathrm{}`$. Let us place all positive distributions, including the experimental one, in a common manifold, $``$. This manifold will have the Riemannian structure, $`G`$, provided by the Fisher metric. We then draw the geodesic curve through the data point that has shortest distance to the sub-manifold $`_0`$; where it cuts $`_0`$ is our estimate for the state. This procedure, however, does not always lead to unbiased estimators. Efron and Dawid noticed that the Levi-Civita connection is not the only useful one, and that there are others that might be used in estimation theory. First, the ordinary mixtures of densities $`\rho _1,\rho _2`$ leads to
$$\rho =\lambda \rho _1+(1\lambda )\rho _2,0<\lambda <1.$$
(23)
Done locally, this leads to a connection on the manifold, now called the $`(1)`$-Amari connection: two tangents are parallel if they are proportional as functions on the sample space. This differs from the parallelism given by the Levi-Civita connection. We need to use $`(1)`$-geodesics to give unbiased estimates for $`f`$.
There is another obvious convex structure, that obtained from the linear structure of the space of centred random variables, also known as the scores. Take $`\rho _0`$ and write $`f_0=\mathrm{log}\rho _0`$. Consider a perturbation $`\rho __X`$ of $`\rho _0`$, which we write as
$$\rho __X=Z_X^1e^{f_0X}.$$
(24)
The random variable $`X`$ is not uniquely defined by $`\rho _X`$, since by adding a constant to $`X`$, we can adjust the partition function to give the same $`\rho _X`$. Among all these equivalent $`X`$ we can choose the score which has zero expectation in the state $`\rho _0`$: $`\rho _0.X=0`$. We can define a sort of mixture of two such perturbed states, $`\rho __X`$ and $`\rho __Y`$ by
$$\mathrm{`}\lambda \rho __X+(1\lambda )\rho __Y\text{}:=\rho _{_{\lambda X+(1\lambda )Y}}.$$
(25)
This is a convex structure on the space of states, and differs from that given in eq. (23). It leads to an affine connection, now called the $`(+1)`$-Amari connection. How do these connections relate to the metric?
###### Definition 2
Let $`G`$ be a Riemannian metric on the manifold $``$. A connection $`\gamma U_\gamma `$ is called a metric connection if
$$G_\sigma (XU_\gamma ,YU_\gamma )=G_\rho (X,Y)$$
(26)
for all tangent vectors $`X,Y`$ and all paths $`\gamma `$ from $`\rho `$ to $`\sigma `$.
The Levi-Civita connection is a metric connection, but the $`(\pm )`$ Amari connections are not; they are, however, dual relative to the Rao-Fisher metric; let $`\gamma `$ be a path connecting $`\rho `$ with $`\sigma `$; then for all $`X,Y`$:
$$G_\sigma (XU^+(\rho ,\sigma ),YU^{}(\rho ,\sigma ))=G_\rho (X,Y).$$
(27)
Let $`^\pm `$ be the two covariant derivatives obtained from the connections $`U^\pm `$. Amari defines intermediate covariant derivatives
$$^\alpha =\frac{1}{2}(1+\alpha )^++\frac{1}{2}(1\alpha )^{}.$$
(28)
These uniquely define connections, $`U^{(\alpha )}`$, whose dual relative to $`G`$ is $`U^{(\alpha )}`$. The Levi-Civita covariant derivative is the case $`\alpha =0`$, which is self-dual and therefore metric, as is known. Amari shows that $`^{(\pm )}`$ define flat connections without torsion. Flat means that the transport is independent of the path, and ‘no torsion’ means that $`U`$ takes the origin of $`T_\rho `$ to the origin of $`T_\rho `$ around any loop; it is linear, and not a general affine map. In that case there are affine coordinates, that is, global coordinates in which the respective convex structure is obtained by simply mixing coordinates linearly. Amari shows that for $`\alpha \pm 1`$, $`^\alpha `$ is not flat, but that the manifold is a sphere in the Banach space $`\mathrm{}^p`$, $`p=\alpha /2+1/2`$. In particular, the case $`\alpha =0`$ leads to the unit sphere in the Hilbert space $`L^2`$, and the Levi-Civita parallel transport is vector translation in this space. The metric distance between measures is the Hellinger distance, and the natural coordinates are the square-roots of the densities, imitating the wave-functions of quantum mechanics. Similar results were obtained in infinite dimensions in .
In estimation theory, the method of maximum entropy for unbiased estimators makes use of the $`^{}`$ connection. This is true also in the dynamics of neural nets, dense liquids, Onsager theory, Brownian particles in a potential and the Soret and Dufour effects ; the micro-state after a small time is replaced by a macrostate, which is the same as the max-entropy estimation of the state by one on the manifold generated by exponentials of the macrovariables (or, slow variables). The (intractible) microdynamics is continuously projected in a rolling construction onto the (easier) manifold of exponential states. This idea was proposed by Kossakowski , Ingarden, et al. , and beautifully expounded by Balian, et al. . The resulting non-linear dynamics can be described thus: after each time-step of the linear dynamics of the system, Nature makes the best estimate of the state among those lying on the manifold.
## 5 The finite quantum info manifold
Chentsov asked whether the Fisher-Rao metric was unique. Any manifold has a large number of different metrics on it; apart from those that differ just by a constant factor, one can multiply a metric by a space-dependent factor. There are many others. Chentsov therefore imposed conditions on the metric. He saw the metric (and the Fisher metric in particular) as a measure of the distinguishability of two states. He argued that if this is to be true, then the distance between two states must be reduced by any stochastic map; for, a stochastic map must ‘muddy the waters’, reducing our ability to distinguish states. He therefore considered the class of metrics $`G`$ that are reduced by any stochastic map on the random variables.
###### Definition 3
A stochastic map is a linear map on the algebra of random variables that preserves positivity and takes 1 to itself.
Chentsov was able to prove that the Fisher-Rao metric is unique, among all metrics, being the only one (up to a constant multiple) that is reduced by any stochastic map. It is therefore uniquely defined up to this factor within the category of commutative function algebras, with stochastic maps as morphisms.
In quantum mechanics, instead of the abelian algebra of random variables we use the algebra of matrices $`M_n`$. Measures on $`\mathrm{\Omega }`$ are replaced by ‘states’, that is, $`n\times n`$ density matrices. For convenience we limit discussion to the interior of the set of states; these are positive-definite matrices of trace 1, which are faithful states and invertible matrices. We take this set to be the manifold $``$; it is a genuine manifold, and not one of the non-commutative manifolds without points that occur in Connes’s theory. The natural morphisms of the quantum info manifold are the completely positive maps that preserve the identity. Chentsov found that uniqueness of the metric is not true for quantum mechanics. (Actually, Petz completed the analysis after Chentsov died; see ).
As in the classical case, there are several affine structures on this manifold. The first comes from the mixing of the states, and is called the $`1`$-affine structure. Coordinates for a state $`\rho `$ in a hood of $`\rho _0`$ provided by $`\rho \rho _0`$, a small traceless matrix. The whole tangent space at $`\rho `$ is thus identified with the set of traceless matrices, and this is a vector space with the usual rules for adding matrices. Obviously, the manifold is flat relative to this affine structure.
The $`+1`$-affine structure is constructed as follows. Since a state $`\rho _0`$ is faithful we can write $`H_0:=\mathrm{log}\rho _0`$ and any $`\rho `$ near $`\rho _0`$ as
$$\rho =Z_X^1\mathrm{exp}(H_0+X)$$
(29)
for some Hermitian matrix $`X`$, which is ambiguous up to a multiple of the identity. We choose to fix $`X`$ by requiring $`\rho _0.X=0`$, and call $`X`$ the ‘score’ of $`\rho `$. Then the tangent space at $`\rho `$ can be identified with the set of scores, and the $`+1`$-linear structure is given by matrix addition of the scores. Corresponding to these two affine structures, there are two affine connections, whose covariant derivatives are denoted $`^{(\pm )}`$. Following Hasagawa , one can also form interpolating affine structures from eq. (28).
As an example of a metric on $``$, let $`\rho `$, and for $`X,Y`$ in $`T_\rho `$ define the GNS metric by
$$G_\rho (X,Y)=\mathrm{Re}\mathrm{Tr}[\rho XY].$$
(30)
This metric is reduced by all cp stochastic maps $`F`$; that is, it obeys
$$G_{F^{}\rho }(XF,XF))G_\rho (X,X),$$
(31)
in accordance with Chentsov’s idea. $`G`$ is just the real part of the scalar product in the Gelfand-Naimark-Segal construction, and is positive definite since $`\rho `$ is faithful. This has been adopted by Helstrom and others in the theory of quantum estimation theory. However, Nagaoka has noted that if we take this metric, then the $`(+1)`$ and the $`(1)`$ affine connections are not dual; the dual to the $`(1)`$ affine connection, relative to this metric, is not flat and has torsion. This failure of duality is confirmed in .
In estimation theory we naturally seek a quantum analogue of the Cramer-Rao inequality. Given a family $``$ of density operators, parametrised by a real parameter $`\eta `$, we seek an estimator $`X`$ whose mean we can measure in the true state $`\rho _\eta `$. To be unbiased, we require $`\mathrm{Tr}\rho _\eta X=\eta `$, which, as in the classical case gives
$$\mathrm{Tr}\left\{\rho _\eta \rho _\eta ^1\frac{\rho _\eta }{\eta }(X\eta )\right\}=1.$$
(32)
It is tempting to regard $`L_r=\rho ^1\rho /\eta `$ as a quantum analogue of the Fisher info; it has zero mean, and the above equation says that its covariance with $`X\eta `$ is equal to 1. The Schwarz inequality then leads to $`𝒱(X)[\rho _\eta .(L_r^{}L_r)]^1`$, where we use $`\rho .X`$ to denote $`\mathrm{Tr}[\rho X]`$. For several estimators, the method used earlier gives this as a matrix inequality.
However, $`\rho `$ and its derivative do not (in general) commute, so $`Y`$ is not Hermitian, and is not popular as a measure of quantum information. Helstrom, and Petz and Toth get round this by using the idea of a logarithmic derivative. Let $`g`$ be a real or complex scalar product on the space of matrices; we say that a matrix $`L`$ is the $`g`$-logarithmic derivative of the family $`\rho _\eta `$ if for any matrix $`X`$,
$$\frac{\rho _\eta .X}{\eta }=g(L^{},X).$$
(33)
The symmetric logarithmic derivative uses the real part of the GNS metric for $`g`$, so that
$$\frac{}{\eta }\mathrm{Tr}(\rho _\eta X)=\frac{1}{2}\mathrm{Tr}[\rho _\eta (L_sX+XL_s)].$$
(34)
Another metric in Chentsov’s allowed class is the Bogoliubov-Kubo-Mori metric; let $`X`$ and $`Y`$ have zero mean in the state $`\rho `$. Then put
$$g_\rho (X,Y)=_0^1\mathrm{Tr}\left[\rho ^\alpha X\rho ^{1\alpha }Y\right]𝑑\alpha .$$
(35)
This is one of the family of scalar products found by Petz to obey the Chentsov property (a similar property was proved in , with detailed balance replacing complete positivity). The corresponding logarithmic derivative, $`L_B`$, is defined such that
$$\frac{}{\eta }\rho _\eta .X=_0^1\rho _\eta ^\lambda L_B\rho _\eta ^{1\lambda }X𝑑\lambda $$
(36)
and is given explicitly by
$$L_B=_0^{\mathrm{}}(\lambda +\rho _\eta )^1\frac{\rho _\eta }{\eta }(\lambda +\rho _\eta )^1𝑑\lambda .$$
(37)
Each metric leads to a Cramer-Rao inequality, also in matrix form for several estimators, and some of these are stronger than others .
The $`BKM`$ metric has other desirable properties, apart from entering in Kubo’s ‘theory of linear response’. For the metric $`g`$, the connections with covariant derivatives $`^{(\pm \alpha )}`$ are dual, and there are affine coordinates for $`^\alpha `$, namely, it is the unit sphere in the (finite-dim.) Banach space $`𝒞_p`$, the Schatten class with norm $`X_p=\left(\mathrm{Tr}|X|^p\right)^{1/p}`$. The case $`p=1/2`$, or $`\alpha =0`$, leads to the Hilbert space of Hilbert-Schmidt operators, which has been used in . More, the Massieu function $`\mathrm{log}Z`$ is the generating function for all the connected Kubo functions, and in particular, the mean is the first derivative, and the metric is the second, as in eq. (14). The entropy is again the Legendre transform of the Massieu function, and the reciprocal relations of eq. (15) hold. It follows that the Cramer-Rao inequality for the $`BKM`$-metric is achieved exactly for the exponential family, agreeing with the method of maximum entropy.
## 6 Araki’s expansionals and the analytic manifold
Araki has considered the case where $`\rho `$ is a KMS state on a $`W^{}`$-algebra. He then perturbed the state by adding bounded operators to the KMS Hamiltonian; the perturbed KMS state has a convergent Kubo-Mori perturbation expansion, which defines an analytic function in the Banach space of bounded perturbations. We try to follow this for unbounded perturbations.
Let $`\mathrm{\Sigma }`$ be the set of density operators on $``$, and let $`\mathrm{int}\mathrm{\Sigma }`$ be its interior, the faithful states. We shall deal only with systems described by $`\rho \mathrm{int}\mathrm{\Sigma }`$; this means that for a free Schrödinger particle, or system of such, we are limited to systems inside a finite volume of real space. Then we would expect the entropy to be finite. The following class of states turns out to be tractable. Let $`p(0,1)`$ and let $`𝒞_p`$, denote the set of operators $`C`$ such that $`|C|^p`$ is of trace class. This is like the Schatten class, except that we are in the bad case, $`0<p<1`$, for which $`C(\mathrm{Tr}[|C|^p])^{1/p}`$ is only a quasi-norm. Let
$$𝒞_<=\underset{0<p<1}{}𝒞_p.$$
(38)
One can show that the entropy
$$S(\rho ):=\mathrm{Tr}[\rho \mathrm{log}\rho ]$$
(39)
is finite for all states in $`𝒞_<`$. We take the underlying set of the quantum info manifold to be
$$=𝒞_<\mathrm{int}\mathrm{\Sigma }.$$
(40)
We shall cover $``$ with balls, each belonging to a Banach space, and shall show that we have a Banach manifold when $``$ is furnished with the topology induced by the norms; for this, the main problem is to ensure that various Banach norms are equivalent.
Let $`\rho _0`$ and write $`H_0=\mathrm{log}\rho _0+cI`$. We choose $`c`$ so that $`H_0I`$, and we write $`R_0=H_0^1`$ for the resolvent at $`0`$. We define a ’hood of $`\rho _0`$ to be the set of states of the form
$$\rho _V=Z_V^1\mathrm{exp}\left(H_0+V\right),$$
(41)
where $`V`$ is a sufficiently small $`H_0`$-bounded form perturbation of $`H_0`$. The necessary and sufficient condition to be Kato-bounded is that
$$V_0:=R_0^{1/2}VR_0^{1/2}_{\mathrm{}}<\mathrm{}.$$
(42)
The set of such $`V`$ make up a Banach space, $`𝒯(0)`$, with (42) as norm. The first result is that $`\rho _V`$ for $`V`$ inside a small ball in $`𝒯(0)`$. For the proof, let $`a`$ be the form-bound of $`V`$, and let $`q__V`$ be the form of $`H_0+V`$. Then we have for some $`b0`$,
$$bI+(1a)q_0q_VbI+(1+a)q_0.$$
(43)
Let $`L`$ be any finite dimensional subspace of $`\mathrm{Dom}q_0`$, and put
$$\lambda (q,L)=sup\{q(\psi ,\psi ):\psi =1,\psi L\}.$$
(44)
Then the ordered eigenvalues of $`q`$ are given by
$$\lambda (q,n)=inf\{\lambda (q,L):dimL=n\}.$$
(45)
From (43) we have for each $`L`$,
$$b+(1a)\lambda (q_0,n)\lambda (q_V,L).$$
(46)
Since $`\lambda (q_0,n)\mathrm{}`$ with $`n`$, the spectrum of $`H_V`$ is purely discrete. Thus
$$\mathrm{exp}\beta \left(b(1a)\lambda (q_0,n)\right)\mathrm{exp}\beta \lambda (q_V,n).$$
(47)
Summing over $`n`$ gives the traces
$$\mathrm{Tr}e^{\beta H_V}e^{\beta (b(1a)H_0)}$$
which is of trace class for some $`\beta <1`$ if $`a`$ is small enough.
We now consider the special case when $`V`$ is an $`H_0`$-bounded as an operator; the condition for this is $`R_0V<\mathrm{}.`$ Then $`V`$ is also form-bounded, since
$$R_0^{1/2}VR_0^{1/2}_{\mathrm{}}R_0V_{\mathrm{}}<\mathrm{}.$$
(48)
In this case we can use the larger norm to provide a topology. This is not equivalent to the topology we get using the norm (42); we are moving from $`\rho _0`$ in a direction more regular than the general direction in the tangent space, and this allows us to furnish this slice of the manifold with a stronger topology. The state defined by $`V`$ is given by
$$\rho _V:=Z_V^1\mathrm{exp}(H_0+V).$$
(49)
Thus, $`V`$ and $`V+cI`$ give rise to the same state; near $`\rho _0`$ the regular directions in $``$ are thus parametrised by the quotient space
$$\widehat{𝒯}=𝒯/\{cI\}.$$
(50)
We may therefore use the score, $`V\rho _0.V`$, as coordinates for the ‘regular’ manifold, now using just the operator bounded perturbations. We show that these are displacements of the state in analytic directions; in we find a more general class of analytic directions, which together make up the ‘analytic’ manifold. This is an attempt to find the quantum analogue of the Cramer class. We shall come to this later.
The norms $`R_0V_{\mathrm{}}`$ on overlapping regions are equivalent. For, around $`\rho _V`$ we perturb with $`X`$ such that $`R_VX_{\mathrm{}}<\mathrm{}`$, and
$$R_VX_{\mathrm{}}=R_VH_0R_0X_{\mathrm{}}R_VH_0.R_0V_{\mathrm{}},$$
(51)
and the converse inequality holds similarly. We define the $`(+)`$-affine connection by transporting the score $`V\mathrm{Tr}\rho V`$ at the point $`\rho `$ to the score $`V\mathrm{Tr}\sigma V`$ at $`\sigma `$. This connection is flat and torsion-free, since it patently does not depend on the path between $`\rho `$ and $`\sigma `$. The $`()`$-connection can be defined in $``$ since each $`𝒞_p`$ is a vector space. It is likely, but not proved, that the $`()`$-mixture of states is continuous in the topology we have defined here.
A case between operator bounded and form bounded is $`ϵ`$-bounded:
$$V_ϵ:=R_0^{1/2ϵ}VR_0^{1/2+ϵ}_{\mathrm{}}<\mathrm{},0ϵ1/2.$$
(52)
This is the analogue of the Cramer class, since we prove that $`Z`$ is an analytic function of $`V`$ in this case.
Araki proved that if $`V`$ is bounded, the Kubo-Mori expansion converges:
$$\mathrm{log}Z_V=\underset{n=0}{\overset{\mathrm{}}{}}(n!)^1_0^1d\alpha _i\delta (\alpha _i1)K_n$$
(53)
where
$$K_n:=\mathrm{Tr}\left(\rho ^{\alpha _1}V\mathrm{}\rho ^{\alpha _n}V\right).$$
(54)
We prove (with Grasselli) that the series converges also for $`ϵ`$\- bounded perturbations, and that the $`V_ϵ`$ are equivalent on overlapping regions. We now give an outline of the method.
We need an economical estimate for the $`n`$-Kubo function. If $`V`$ were bounded, we could use the Hölder inequality for traces, with $`p_i=1/\alpha _i`$ using that $`\alpha _i=1`$:
$$|\mathrm{Tr}\left[\rho ^{\alpha _1}V_1\mathrm{}\rho ^{\alpha _n}V_n\right]|\mathrm{Tr}\rho V_1_{\mathrm{}}\mathrm{}V_n_{\mathrm{}}.$$
(55)
We do better, since there is $`\beta <1`$ such that $`\rho ^\beta `$ is of trace class, so we can replace $`\rho `$ by $`\rho ^\beta `$. We can thus borrow $`\rho ^{(1\beta )\alpha _j}`$ to help bound the potentials. Also, as $`\alpha _j=1`$, the region of integration is the (overlapping) union of regions $`S_j`$ where $`\alpha _j1/n`$. By cyclicity, we may take $`j=n`$. We then write $`\rho ^{\alpha _j}V_j`$ as
$$\mathrm{}\left[\rho ^{\alpha _j\beta }\right]\left[H^{1\delta _{j1}+\delta _j}\rho ^{(1\beta )\alpha _j}\right]\left[R^{\delta _j}V_jR^{1\delta _j}\right]\mathrm{}$$
(56)
The dots are factors taken with other terms. We bound the middle $`[\mathrm{}]`$ by the spectral theorem, arranging the parameters $`\delta _j`$ so that we get an integrable function of $`\alpha _j`$ in $`S_n`$, $`1jn1`$. We bound the final $`[\mathrm{}]`$ using the $`ϵ`$-boundedness of $`V`$, by a suitable choice of the $`\delta _j`$. We end up with a factorial bound on the $`n`$-point function, so the series converges as a geometric series.
The manifold can be furnished by a real-analytic structure, by asserting that the ring of germs of analytic functions on the manifold consists of functions that are analytic in these analytic directions. The mixture coordinates $`\eta `$ are examples of analytic functions; we say that we have an analytic parametrisation of the manifold by $`\eta `$. It remains to prove that the $`\xi `$ are analytic functions of $`\eta `$, before we can say that $`\eta `$ are analytic coordinates.
## 7 Singular perturbations
Every point of our manifold has some directions in its tangent space that remain within $``$ but are not analytic directions. Consider the anharmonic oscillator,
$$H=(p^2+q^2)/2+\lambda q^{2n},\lambda >0.$$
(57)
It is known that $`\mathrm{exp}\beta H`$ is of trace-class for all $`\beta >0`$, so these states are in $``$. It is also known that there is a singularity at $`\lambda =0`$. Our result shows that if we start at $`\lambda >0`$ then there is a region around this state where the manifold has analytic directions. Obviously, any point in $``$ has many analytic directions: the bounded perturbations, provide many such. The metric is finite in a much wider class of directions: if $`\rho ^\beta `$ is of trace-class, and $`V`$ is a form such that $`\rho ^\delta V`$ is bounded for $`\delta =(1\beta )/2`$, the a regularised BKM metric in the $`V`$-direction is finite at $`\rho `$.
The natural class of states, the analogue of the Orlicz space of , is the set $`_{\mathrm{max}}`$ of states of finite entropy. The natural class of states $`\sigma `$ in a ’hood of a state $`\rho `$ of finite entropy consists of states of finite entropy whose entropy relative to $`\rho `$ is also finite. This ’hood will consist of many non-analytic perturbations of $`\rho `$. It is known that the $`1`$-mixture (the usual mixture) of states of finite entropy has finite entropy, so $`_{\mathrm{max}}`$ has the $`1`$-affine structure. Here is a simple proof.
###### Theorem 4
$$S(\lambda \rho +(1\lambda )\sigma )\lambda S(\rho )+(1\lambda )S(\sigma )+\lambda \mathrm{log}(1/\lambda )+(1\lambda )\mathrm{log}(1/(1\lambda )).$$
(58)
Proof.
$`\mathrm{log}x`$ is an operator monotone decreasing function. Since $`\lambda \rho +(1\lambda )\sigma \lambda \rho `$, we have
$$\mathrm{log}(\lambda \rho +(1\lambda )\sigma )\mathrm{log}(\lambda \rho ).$$
Hence
$$\lambda \rho .\mathrm{log}(\lambda \rho +(1\lambda )\sigma )\lambda \rho .\mathrm{log}(\lambda \rho ).$$
Similarly
$$(1\lambda )\mathrm{log}(\lambda \rho +(1\lambda )\sigma )(1\lambda )\sigma \mathrm{log}((1\lambda )\sigma ).$$
Adding, gives
$`S(\lambda \rho +(1\lambda )\sigma )`$ $``$ $`\lambda \rho .(\lambda \rho )(1\lambda )\sigma .\mathrm{log}((1\lambda )\sigma )`$
$`=`$ $`\lambda S(\rho )+(1\lambda )S(\sigma )+\lambda \mathrm{log}(1/\lambda )+(1\lambda )\mathrm{log}(1/(1\lambda ))<\mathrm{}.`$
So the space $`_{\mathrm{max}}`$ of density matrices of finite entropy is a $`(1)`$-affine space.
In we propose a Luxemburg norm for the tangent space at a point $`\rho _{\mathrm{max}}`$. We expect that a ’hood of a point $`\rho `$ will consist of all states $`\sigma _{\mathrm{max}}`$ having finite relative entropy, thus: $`S(\sigma |\rho ):=\rho .(\mathrm{log}\rho \mathrm{log}\sigma )<\mathrm{}`$.
Acknowledgements
It is a pleasure to thank M. Ohya for the invitation to the conference, H. Araki for discussions, and H. Hasagawa for arranging the trip.
|
warning/0002/hep-ph0002077.html
|
ar5iv
|
text
|
# Heavy quark production as sensitive test for an improved description of high energy hadron collisions
## Abstract
QCD dynamics at small quark and gluon momentum fractions or large total energy, which plays a major role for HERA, the Tevatron, RHIC and LHC physics, is still poorly understood. For one of the simplest processes, namely $`b\overline{b}`$ production, next-to-leading-order perturbation theory fails. We show that the combination of two recently developed theoretical concepts, the $`k_{}`$-factorization and the next-to-leading-logarithmic-approximation BFKL vertex, gives perfect agreement with data. One can therefore hope that these concepts provide a valuable foundation for the description of other high energy processes.
preprint: TPR-00-04
Existing QCD calculations describe many high energy observables which involve partonic transverse momentum rather poorly. This is also true for the theoretically especially clean case of $`b\overline{b}`$ production, which was investigated experimentally at Fermilab . Since central quark-antiquark production at $`\sqrt{s}=1.8`$ TeV is sensitive to very small gluon momentum fraction $`x10^210^4`$, one probes the gluon content of the nucleon at small $`x`$, which is a central issue of current research. We reconsider this process and combine as essential new ingredients the $`k_t`$-factorization scheme with the next-to-leading-logarithmic-approximation (NLLA) BFKL production vertex derived in . The $`k_t`$-factorization approach for the description of high energy processes differs strongly from the conventional NLO collinear approximation (e.g. ) because it takes the non-vanishing transverse momenta of the scattering partons into account. The usual gluon densities are replaced by unintegrated gluon distributions which depend on the transverse momentum $`k_t`$. These together with the $`k_t`$-factorization form a basis for a general calculation scheme for high energy (i.e. small $`x`$). The standard collinear approximation has the advantage of being closely related to the operator product expansion. It is, however, only justified for the processes dominated by $`x=𝒪(1)`$. In application to processes governed by small $`x`$ the $`k_t`$-factorization approach has the advantage that its approximations correspond to the dominant kinematics. Essential small x contributions are included in the Born approximation which in the collinear approach are accounted for in higher orders only. This is well known from the case of structure functions where the DGLAP evolution is appropriate for $`x=𝒪(1)`$ and the BFKL evolution for small $`x`$.
While the $`k_t`$-factorization formalism is very attractive theoretically, its phenomenological usefullness has been mostly tested in the case of the structure function $`F_2`$ . The NLLA BFKL vertices are just the ones needed to treat semi-hard central production at collider energies in this approach.
In our calculation we use one particular element of the NLLA BFKL formalism , namely the effective vertex for quark-antiquark production. Thus our calculation can be seen as a first phenomenological application of this vertex which decides whether the NLLA BFKL formalism can be hoped to converge.
One special aspect of the reaction we investigate is the possible loss of gauge invariance when a $`q\overline{q}`$ production vertex is incorporated into an amplitude with off-shell gluons. In the BFKL approach, however, gauge invariance is ensured automatically by the use of the just mentioned NLL effective vertex which is valid in quasi multi Regge kinematics (QMRK), i.e. when the $`q`$ and $`\overline{q}`$ have similar rapidities and form a cluster (in contrast to LLA, where the particles are produced with a large rapidity gap).
We begin with the following definition for the light cone coordinates and the momenta of the scattering hadrons in the c.m. frame
$`k^+=k^0+k^3,k^{}=k^0k^3,k_{}=(0,k^1,k^2,0)=(0,𝐤,0).`$ (1)
$`P_1^+=P_2^{}=\sqrt{s},P_1^{}=P_2^+=0,P_1=P_2=0.`$ (2)
The Mandelstam variable $`s`$ is as usual the c.m. energy squared. As defined in fig. 1, $`q_1`$ and $`q_2`$ are the momenta of the gluons and the on-shell quark and antiquark have momentum $`k_1`$ respectively $`k_2`$. In the high energy (large $`s`$) regime we have
$`k_1^++k_2^+=q_1^+q_2^+q_1^+,`$
$`k_1^{}+k_2^{}=q_1^{}q_2^{}q_2^{},`$
$`q_1^2q_1^2,q_2^2q_2^2.`$
The longitudinal momentum fractions of the gluons are $`x_1=q_1^+/P_1^+`$, $`x_2=q_2^{}/P_2^{}`$.
The cross section for heavy quark pair production in the $`k_t`$-factorization approach is then given by
$`\sigma `$ $`{}_{P_1P_2q\overline{q}X}{}^{}={\displaystyle \frac{1}{16(2\pi )^4}}{\displaystyle }{\displaystyle \frac{d^3k_1}{k_1^+}}{\displaystyle \frac{d^3k_2}{k_2^+}}d^2q_1d^2q_2`$ (5)
$`\delta ^2(q_1q_2k_1k_2)(x_1,q_1){\displaystyle \frac{1}{(q_1^2)^2}}`$
$`\left\{{\displaystyle \frac{\psi ^{c_2c_1}\psi ^{c_2c_1}}{(N^21)^2}}\right\}{\displaystyle \frac{1}{(q_2^2)^2}}(x_2,q_2).`$
The factor $`(N^21)^2`$ reflects the projection on color singlet, where $`N`$ is the number of colors. The hard amplitude $`\psi ^{c_2c_1}(x_1,x_2,q_1,q_2,k_1,k_2)`$ is calculable in perturbation theory, whereas the unintegrated gluon distribution $`(x,q_{})`$ has to be measured or modelled. We choose the argument $`\mu ^2`$ of the strong coupling constant $`\alpha _S(\mu ^2)`$ in the hard amplitude $`\psi ^{c_2c_1}`$ to be equal to $`𝐪_1^2=q_1^2`$ respectively $`𝐪_2^2=q_2^2`$ .
We generalize the results on the $`q\overline{q}`$ production vertex presented in for massless QCD in an obvious way in order to take the masses $`m`$ of the produced quarks into account. The resulting vertex $`\mathrm{\Psi }^{c_2c_1}`$ is given by a sum of two terms
$$\mathrm{\Psi }^{c_2c_1}=g^2\left(t^{c_1}t^{c_2}b(k_1,k_2)t^{c_2}t^{c_1}b^T(k_2,k_1)\right),$$
(6)
where $`t^c`$ are the colour group generators in the fundamental representation. The connection between $`\psi ^{c_2c_1}`$ in eq. (5) and $`\mathrm{\Psi }^{c_2c_1}`$ in eq. (6) is given by
$`\psi ^{c_2c_1}=\overline{\text{u}}(k_1)\mathrm{\Psi }^{c_2c_1}\text{v}(k_2),`$
with the on-shell quark and antiquark spinors u$`(k)`$ and v$`(k)`$. The expression for $`b(k_1,k_2)`$ is a sum of two terms
$$b(k_1,k_2)=\gamma ^{}\frac{\mathit{}_1\mathit{}_1m}{(q_1k_1)^2m^2}\gamma ^+\frac{\gamma _\beta \mathrm{\Gamma }^{+\beta }(q_2,q_1)}{(k_1+k_2)^2},$$
(7)
The first term on the r.h.s. of eq. (7) describes the production of a $`q\overline{q}`$ pair by means of usual vertices (see fig. 2), the second term involves the light-cone projection of the effective vertex $`\mathrm{\Gamma }^{+\beta }(q_2,q_1),`$ which describes the transition of two $`t`$-channel gluons (reggeons) with momenta $`q_1`$ and $`q_2`$ to a gluon with momentum $`k_1+k_2`$
$`\mathrm{\Gamma }^{+\beta }(q_2,q_1)`$ $`=`$ $`2(q_1+q_2)^\beta 2q_1^+n^\beta 2q_2^{}n^{+\beta }`$ (9)
$`2t_1{\displaystyle \frac{n^\beta }{q_1^{}q_2^{}}}+2t_2{\displaystyle \frac{n^{+\beta }}{q_1^+q_2^+}},`$
with $`t_{1/2}=q_{1/2}^2`$. This effective vertex differs from the usual triple-gluon vertex by the appearence of the last two terms. They are related to Feynman diagrams in which the $`q\overline{q}`$ pair is not produced by the $`t`$-channel gluons but in other ways. These two last terms in eq. (9) are also required by gauge invariance, $`\mathrm{\Gamma }^{+\beta }(q_2,q_1)(q_1q_2)_\beta =0`$. Another consequence of gauge invariance is the vanishing of the matrix element of the effective vertex $`\mathrm{\Psi }^{c_2c_1}`$ between on-mass-shell quark and antiquark states in the limit of small $`q_1`$ or $`q_2`$
$`\overline{\text{u}}(k_1)\mathrm{\Psi }^{c_2c_1}\text{v}(k_2)0\text{ for }q_1\text{ or }q_20.`$
The function $`b^T(k_2,k_1)`$ is very similar to (7)
$`b^T(k_2,k_1)=\gamma ^+{\displaystyle \frac{\mathit{}_1\mathit{}_2+m}{(q_1k_2)^2m^2}}\gamma ^{}{\displaystyle \frac{\gamma _\beta \mathrm{\Gamma }^{+\beta }(q_2,q_1)}{(k_1+k_2)^2}}.`$
The unintegrated gluon distribution is related to the standard gluon distribution by
$`xg(x,𝐪^2)={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d𝐤^2}{𝐤^2}}\mathrm{\Theta }(𝐪^2𝐤^2)(x,𝐤).`$
Taking the derivative of this expression makes it obvious that $`(x,𝐤)`$ includes the evolution of $`xg(x,𝐪^2)`$, which is given by the BFKL and/or DGLAP equation. Since the unintegrated gluon distribution is not known at small $`𝐤`$, we write this equation as
$$xg(x,𝐪^2)=xg(x,q_0^2)+_{q_0^2}^{\mathrm{}}\frac{d𝐤^2}{𝐤^2}\mathrm{\Theta }(𝐪^2𝐤^2)(x,𝐤).$$
(10)
This formula has been repeatedly used and introduces the a priori unknown initial scale $`q_0`$ and the initial gluon distribution $`xg(x,q_0^2)`$. Following , one may neglect the hard cross section dependence on $`𝐪`$ in the soft region $`|𝐪|<q_0`$, so that
$`{\displaystyle \frac{1}{q_1^2}}`$ $`\left\{{\displaystyle \frac{\psi ^{c_2c_1}\psi ^{c_2c_1}}{(N^21)^2}}\right\}{\displaystyle \frac{1}{q_2^2}}S(q_1,q_2)`$ (12)
$`S(q_1,q_2)\mathrm{\Theta }(𝐪_1^2q_0^2)\mathrm{\Theta }(𝐪_2^2q_0^2)`$
$`+`$ $`S(q_1,0)\mathrm{\Theta }(𝐪_1^2q_0^2)\mathrm{\Theta }(q_0^2𝐪_2^2)`$ (13)
$`+`$ $`S(0,q_2)\mathrm{\Theta }(𝐪_2^2q_0^2)\mathrm{\Theta }(q_0^2𝐪_1^2)`$ (14)
$`+`$ $`S(0,0)\mathrm{\Theta }(q_0^2𝐪_1^2)\mathrm{\Theta }(q_0^2𝐪_2^2).`$ (15)
Note that the very existence of the finite limit $`q_{}0`$ follows from the decrease of the production amplitude due to gauge invariance. Substituting this formula in (5) using eq. (10) one may easily perform the integration over $`q_{}`$. As a result, $`S(0,0)`$ produces the standard expression of collinear factorization (ref. ), while $`S(q_1,0)`$, $`S(0,q_2)`$ correspond to the asymmetric configurations, where one of the gluons is described by the unintegrated distribution and the other by the integrated one. Here it is important to notice that when we insert (10), (15) in (5) the coupling constant $`\alpha _s`$ in the term proportional to $`xg(x,q_0^2)`$ is taken to be $`\alpha _s(q_0^2)`$.
In all our numerical calculations we used for the unintegrated gluon distribution $`(x,𝐤)`$ the code by Kwiecinski, Martin and Staśto , because they use a combination of DGLAP and BFKL equations which governs simultaneously the evolution in $`Q^2`$ and $`x`$. They obtain an excellent description of $`F_2(x,Q^2)`$ in a very large $`x`$-$`Q^2`$-window. According to our knowledge this is the only unintegrated gluon distribution which has given such a satisfactory result, which justifies our choice. As in the case of the usual gluon distribution function one has to choose an initial scale and an initial distribution function which in the case of are given by
$`q_0^2`$ $`=`$ $`1\text{ GeV}^2,xg(x,q_0^2)=1.57(1x)^{2.5}.`$ (16)
We use these values, which are fixed by the fit to $`F_2(x,Q^2)`$, in our calculation.
We consider the production of $`b\overline{b}`$-pairs. For the computation we use eqs. (5) and (15) with the unintegrated gluon distribution from and the corresponding values (16). The rapidities and the transverse masses of the produced quark and antiquark are defined by
$`y_{1/2}={\displaystyle \frac{1}{2}}\mathrm{ln}({\displaystyle \frac{k_{1/2}^+}{k_{1/2}^{}}}),m_{1/2}=\sqrt{m^2k_{1/2}^2}.`$
The Bjorken-variables of the gluons can then be written as
$`x_1`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{s}}}(m_1e^{y_1}+m_2e^{y_2}),`$
$`x_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{s}}}(m_1e^{y_1}+m_2e^{y_2}).`$
In fig. 3 we show our results for inclusive $`b`$ production, together with experimental results measured by the D0 Collaboration (see Table II) in$`\sqrt{s}=1.8`$ TeV $`p\overline{p}`$ collisions. We obtain this cross section by integrating out all antibottom variables in eq. (5).
The variable $`k_{1\mathrm{min}}`$ is the lower integration cut on the transverse momentum of the produced $`b`$ quark. To get an indication of the theoretical uncertainties apart from higher order contributions which are not available at the moment we proceed in a similar way as the authors of ref. and present our calculations for three different choices of $`\mathrm{\Lambda }_{\text{QCD}}`$ and the bottom quark mass
high $`:`$ $`\mathrm{\Lambda }^{(5)}=180\text{ MeV, }m_b=4.5\text{ GeV,}`$
central $`:`$ $`\mathrm{\Lambda }^{(5)}=150\text{ MeV, }m_b=4.7\text{ GeV,}`$
low $`:`$ $`\mathrm{\Lambda }^{(5)}=100\text{ MeV, }m_b=4.9\text{ GeV,}`$
Our result is in very good quantitative agreement with data over the whole range of $`k_{1\mathrm{min}}`$. The corresponding central QCD NLO calculation has a similar shape, but is about a factor of $`23`$ smaller than our central result (see for example fig. 11 in ).
We now turn to $`b\overline{b}`$ correlations in $`\sqrt{s}=1.8`$ TeV $`p\overline{p}`$ collisions, which have been measured by the CDF collaboration at Fermilab . The correlations of the quark and antiquark give an insight into the dynamics of the production mechanism and are important in order to study the limits of the collinear ($`k_1=k_2)`$ LO QCD approximation. We present a comparison between our results and the experimental data in fig. 4 and fig. 5. The data points and uncertainties were taken from . We find good agreement with experiment for both $`k_{1\mathrm{min}}=6.5`$ GeV (fig. 4) and $`k_{1\mathrm{min}}=8.75`$ GeV (fig. 5). In this case QCD NLO calculations underestimate the measured cross section roughly by factor of $`3`$ (compare with fig.6 in ).
An interesting parameter concerning the correlation is the opening angle $`\varphi `$ between the momentum vectors of the produced quarks in the plane transverse to the beam axis. Our predictions for the corresponding differential cross sections at Fermilab and LHC energies are shown in fig. 6. As expected we find a peak at $`\varphi =180^{}`$ which shows the dominance of the collinear part.
Additionally we present our predictions for rapidity distributions of the $`\overline{b}`$ for the rapidity of the $`b`$ being $`0`$ and $`\sqrt{s}=1.8`$ TeV respectively $`\sqrt{s}=16`$ TeV in fig. 7. Our cross section for $`\sqrt{s}=1.8`$ TeV at $`y_2=0`$ is about a factor of $`3`$ larger than the corresponding QCD NLO result from .
Let us conclude. We have studied quark-antiquark hadroproduction within the $`k_t`$-factorization approach using an unintegrated gluon distribution and a specific effective BFKL vertex for $`q\overline{q}`$ production. We found very good agreement with experiment for both single $`b`$ production and $`b\overline{b}`$ correlations at $`\sqrt{s}=1.8`$ TeV. Our approach leads to nontrivial $`b\overline{b}`$ correlations already at LO perturbation theory, whereas traditional collinear factorization gives them only at NLO and beyond. In contrast, the available NLO caculations are not in agreement with the Tevatron data we compare with .
Our results show that at least those features of the effective $`q\overline{q}`$ vertex which we tested provide a substantial improvement with respect to the standard collinear treatment.
If further tests of other observables should be equally successfull, the NLL BFKL vertices will also allow for a much improved description of many processes to be studied at RHIC and LHC.
We thank J. Kwiecinski, A. Martin and A. Staśto for supplying us with their code for the calculation of the unintegrated gluon distribution.
L. S. would like to acknowledge the support by the DFG.
|
warning/0002/hep-ph0002131.html
|
ar5iv
|
text
|
# CURRENT STATUS OF RADIATIVE B DECAYS
## 1 Introduction
The experimental challenge of finding new physics in direct search may still take some time if new particles or their effects set in only at several hundred GeV. Complementary to these direct signals at highest available energies are the measurements of the effects of new ”heavy” particles in loops, through either precision measurements or detection of processes occurring only at one loop in the standard model (SM). Among these are the transitions induced by flavor-changing neutral currents (FCNC), such as $`bs\gamma `$.
The first observations of the exclusive $`BK^{}\gamma `$ and inclusive $`BX_s\gamma `$ decays were reported in 1993/94 by the CLEO collaboration. Skwarnicki from the CLEOII$`\&`$II.V reported the recent exclusive branching ratios:
$$BR(B^0K^0\gamma )=4.5\pm 0.7\pm 0.3\times 10^5,$$
(1)
$$BR(B^{}K^{}\gamma )=3.8\pm 0.9\pm 0.3\times 10^5,$$
(2)
$$BR(BK_2^{}(1430)\gamma )=1.7\pm 0.5\pm 0.1\times 10^5,$$
(3)
$$BR(BK^{}(1410)\gamma )<12.7\times 10^5,$$
(4)
$$BR(BK_2^{}(1430)\gamma )/BR(BK^{}\gamma )=0.4\pm 0.1.$$
(5)
The measurements envolving higher resonant $`K^{}`$-states are still preliminary, since the error of $`100MeV`$ in the final-state mass spectrum of means that we do not know what resonant state has been observed. This year, CLEOII$`\&`$II.V collaborations have also reported the inclusive branching ratio
$$BR(BX_s\gamma )=3.15\pm 0.35\pm 0.41\times 10^4.$$
(6)
Using the latest results for inclusive and exclusive branching ratios (BR) , we have obtained the following central value for the so-called hadronization rate: $`R_K^{}^{exp}13\%`$. Note here that Skwarnicki has also reported preliminary results for exclusive modes based on the quark $`bd\gamma `$ electroweak transition:
$`BR(B^0\rho ^0\gamma )<4\times 10^5;BR(B^{0()}\omega (\rho ^{})\gamma )<1\times 10^5`$.
## 2 NNL QCD Corrected $`bs\gamma `$ Transition
The $`bs\gamma `$ process is a one-loop electroweak process, where the extra gluon exchange (QCD corrections) changes the nature, i.e. the functional structure of the GIM cancellation: $`(m_t^2m_c^2)/m_w^2\mathrm{ln}(m_t^2/m_c^2)`$.
In the standard model, $`B`$ decays are described by the effective Hamiltonian obtained by integrating out the top-quark and $`W`$-boson fields:
$$H_{\mathrm{\Delta }B=1}^{\mathrm{eff}}(bs\gamma )=\frac{G_F}{\sqrt{2}}V_{tb}V_{ts}^{}\underset{i=1}{\overset{8}{}}c_i(\mu )O_i(\mu ),$$
(7)
where $`c_i`$’s are the well-known Wilson coefficients. The SM theoretical prediction, up to Next-to-Leading Order (NLO) in $`\alpha _sln(m_w/m_b)`$, is considerably larger than the lowest-order result. Buras et al. performed a new analysis by using expansions in powers of $`\alpha _s`$ and reported a slightly higher, short-distance (SD) result: $`BR(BX_s\gamma )_{\mathrm{NLO}}=(3.60\pm 0.33)\times 10^4`$.
## 3 Exclusive Radiative B Decays
Exclusive modes are, in principle, affected by large theoretical uncertainties due to the poor knowledge of nonperturbative dynamics and of a correct treatment of large recoil-momenta, determining form factors.
The Lorentz decomposition of the $`BK^{}`$ matrix element of the operator $`O_7`$, taking into account the gauge condition, the current conservation, the spin symmetry and for the on-shell photon, gives the following hadronization rate $`R_K^{}`$:
$$R_K^{}=\frac{\mathrm{\Gamma }(BK^{}\gamma )}{\mathrm{\Gamma }(bs\gamma )}=[\frac{m_b(m_B^2m_K^{}^2)}{m_B(m_b^2m_s^2)}]^3(1+\frac{m_s^2}{m_b^2})^1|T_1^K^{}(0)|^2.$$
(8)
Since the first calculation of the hadronization rate $`R_K^{}6\%`$ by Deshpande at al. a large number of papers have reported $`R_K^{}`$ from the range of 5% to (unrealistic) 80%. Different methods have been employed, from simple quark models, QCD sum rules, HQET and chiral symmetry, QCD on the lattice, light cone sum rules, to the perturbative QCD type of evaluations of exclusive modes. In any event, the above form factor will be obtained in the future from first-principle calculations on the lattice. Recently, it looks like that the hadronization rate calculations have stabilized around 10%.
If long-distance (LD) and other nonperturbative effects are neglected, two exclusive modes are connected by a simple relation
$$BR(B\rho \gamma )=\xi ^2\left|V_{td}/V_{ts}\right|^2BR(BK^{}\gamma ),$$
(9)
where $`\xi `$ measures the SU(3) breaking effects. They are typically of the order of 30%. Misiak has reported the SD BR’s:
$$BR(bd\gamma )=1.61\times 10^5;BR(B^+\rho ^+\gamma )=[1,4]\times 10^6,$$
(10)
$$BR(B^0\rho ^0\gamma )=BR(B^0\omega \gamma )=[0.5,2]\times 10^6.$$
(11)
## 4 LD Contributions to Inclusive/Exclusive B Decays
First note that LD corrections cannot be computed from first principles; it is possible to estimate them phenomenologically. For instance, the operators $`O_{1,2}`$ contain the $`\overline{c}c`$ current. So one could imagine the $`\overline{c}c`$ pair propagating through a long distance, forming intermediate $`\overline{c}c`$ states (off-shell $`J/\psi `$’s), which turn into a photon via the vector meson dominance (VMD).
The total (SD + LD) amplitude for $`bd(s)\gamma `$ is
$`M(bd\gamma )={\displaystyle \frac{eG_F}{2\sqrt{2}}}[V_{tb}V_{td}^{}({\displaystyle \frac{m_b}{4\pi ^2}}c_7^{eff}(\mu ){\displaystyle \frac{2}{3}}a_2{\displaystyle \underset{i}{}}{\displaystyle \frac{g_{\psi _i}^2(0)}{m_{\psi _i}^2m_b}})`$ (12)
$`{\displaystyle \frac{a_2}{m_b}}V_{ub}V_{ud}^{}({\displaystyle \frac{2}{3}}{\displaystyle \underset{i}{}}{\displaystyle \frac{g_{\psi _i}^2(0)}{m_{\psi _i}^2}}{\displaystyle \frac{1}{2}}{\displaystyle \frac{g_\rho ^2(0)}{m_\rho ^2}}{\displaystyle \frac{1}{6}}{\displaystyle \frac{g_\omega ^2(0)}{m_\omega ^2}})]\overline{d}\sigma ^{\mu \nu }(1+\gamma _5)bF_{\mu \nu }.`$
If in the above equation we replace the d by the s quark and forget the last three terms, then we obtain the total amplitude for the $`bs\gamma `$ decay. Deshpande et al have found a strong suppression when extrapolating $`g_\psi (m_\psi ^2)`$ to $`g_\psi (0)`$: $`g_{\psi (1S)}^2(0)/g_{\psi (1S)}^2(m_\psi ^2)=0.13\pm 0.04`$. The LD contributions to an inclusive mode and to its exclusive modes are all found to be small, typically one to two orders of magnitude below the SD’s.
## 5 Discussion and Conclusions
Note that the so-called spectator-quark contributions and the first calculable nonperturbative, essentialy long distance, correction to the inclusive rate are of the order of a few percent. It has also been proved that the fermionic (quarks and leptons) and photonic loop corrections to $`bs\gamma `$ reduce $`BR(bs\gamma )/BR(bce\overline{\nu })`$ by $`8\pm 2\%`$. Consequently, it is more appropriate to use $`\alpha _{em}=1/137`$ for the real photon emission.
In general, we can conclude that, in theory, more effort is required in calculating quark (inclusive) decays through higher loops. A better understanding of bound states of heavy-light quarks (B-meson ect.) and highly recoiled light-quark bound states ($`K^{}`$, $`\rho `$, …) is desirable. This can be achived by inventing new, more sophisticated (perturbative) methods and apply them to the calculation of radiative B-meson decays, which incorporate the full spectrum of quark bound states ($`K^{}`$, $`\rho `$, $`K_1^{}`$, …). In experiment, with a larger amount of data we might expect a stable but small increase of $`BR(bs\gamma )`$, a decrease of $`BR(BK^{}\gamma )`$, determinations of $`BR(BK_1^{}\gamma )`$ and $`BR(BK_2^{}\gamma )`$ and first measurements of $`BR(bd\gamma )`$ and $`BR(B\rho \gamma )`$.
|
warning/0002/nucl-th0002049.html
|
ar5iv
|
text
|
# 𝑞BCS - the BCS theory of 𝑞-deformed nucleon pairs
## I Zero Coupled $`q`$-deformed Nucleon Pairs
The creation and destruction operators for a zero coupled nucleon pair in a shell model orbit $`j`$ are
$$Z_0=\frac{1}{\sqrt{2}}(A^j\times A^j)^0\text{ and }\overline{Z}_0=\frac{1}{\sqrt{2}}(B^j\times B^j)^0,$$
(1)
where $`A_{jm}=a_{jm}^{};`$ $`B_{jm}=(1)^{j+m}a_{j,m}`$. From the anticommutation relations satisfied by the fermion creation and destruction operators $`a_{jm}^{}`$ and $`a_{j,m}`$, we can verify that with number operator for fermions defined as $`n_{op}^j=_ma_{jm}^{}a_{jm}`$ and $`\mathrm{\Omega }=(2j+1)/2`$,
$$[Z_0,\overline{Z_0}]=\frac{n_{op}\mathrm{\Omega }}{\mathrm{\Omega }}\text{ ; }[n_{op},Z_0]=2Z_0\text{ ; }[n_{op},\overline{Z_0}]=2\overline{Z_0}\text{ .}$$
(2)
These operators are easily related to well known quasi-spin operators by identifying
$$S_+=\sqrt{\mathrm{\Omega }}Z_0\text{ ; }S_{}=\sqrt{\mathrm{\Omega }}\overline{Z_0},\text{ and }S_0=\frac{(n_{op}\mathrm{\Omega })}{2}.$$
(3)
The quasi-spin operators $`S_+`$, $`S_{}`$, and $`S_0`$ are the generators of Lie algebra of SU(2) and satisfy the commutation relations of angular momentum operators, that is
$$[S_+,S_{}]=2S_0\text{ , }[S_0,S_\pm ]=\pm S_\pm \text{ .}$$
(4)
The generators of SU$`{}_{q}{}^{}(2)`$ on the other hand satisfy the $`q`$-commutation relations
$$[S_+(q),S_{}(q)]=\left\{2S_0(q)\right\}_q\text{ , }[S_0(q),S_\pm (q)]=\pm S_\pm (q)\text{ ; }$$
(5)
where $`\{x\}_q=\frac{(q^xq^x)}{(qq^1)}`$. Translated to $`q`$-deformed pair operators $`Z_0(q)`$ and $`\overline{Z_0(q)}`$ the new commutation relations give
$$[Z_0(q),\overline{Z_0(q)}]=\frac{\left\{n_{op}\mathrm{\Omega }\right\}_q}{\mathrm{\Omega }}\text{ ; }[n_{op},Z_0(q)]=2Z_0(q)\text{ ; }[n_{op},\overline{Z_0(q)}]=2\overline{Z_0(q)}\text{ .}$$
(6)
## II The Trial Wave Function
The proposed trial wave function for $`N`$ nucleons distributed over $`m`$ single particle orbits is, $`\mathrm{\Psi }=\mathrm{\Phi }_{j_1}\mathrm{\Phi }_{j_2}\mathrm{}\mathrm{\Phi }_{j_m}`$, where for the orbit $`j,`$
$$\mathrm{\Phi }_j=u_j^{\mathrm{\Omega }_j}\underset{n=0}{\overset{\mathrm{\Omega }_j}{}}\left(\frac{v_j}{u_j}\right)^n\left[\frac{\mathrm{\Omega }_j!}{n!(\mathrm{\Omega }_jn)!}\right]^{\frac{1}{2}}|n\text{ ; }\mathrm{\Omega }_j=\frac{2j+1}{2}\text{ }$$
(7)
and
$`|n=\left[{\displaystyle \frac{\left\{\mathrm{\Omega }_jn\right\}_q!}{\left\{n\right\}_q!\left\{\mathrm{\Omega }_j\right\}_q!}}\right]^{\frac{1}{2}}\left(S_{j+}(q)\right)^n|0`$
is the normalized wave function for $`n`$ zero coupled nucleon pairs with $`q`$-deformation occupying single particle orbit $`j`$. The function $`\mathrm{\Psi }`$ is normalized in case, $`u_j^2+v_j^2=1,`$ for all single particle orbits. The single particle plus pairing Hamiltonian for $`q`$-deformed pairs is given by
$$H=\underset{r}{}\epsilon _rn_{op}^rG\underset{rs}{}S_{r+}(q)S_s(q)\text{ };\text{where }r,sj_1,j_2,\mathrm{}\mathrm{}.j_m\text{.}$$
(8)
The matrix element $`\mathrm{\Psi }\left|GS_{r+}(q)S_r(q)\right|\mathrm{\Psi }`$ , obtained by using the $`q`$-commutation relations given in Eq. (5) and ignoring terms involving products of the type $`v_r^4u_r^m(m=2,4,..,\mathrm{\Omega }_r)`$, is found to be
$`\mathrm{\Psi }\left|GS_{r+}(q)S_r(q)\right|\mathrm{\Psi }=Gv_r^2\mathrm{\Omega }_r\left\{\mathrm{\Omega }_r\right\}_q+Gv_r^4(\mathrm{\Omega }_r1)\left\{\mathrm{\Omega }_r\right\}_q.`$
We also calculate the gap parameter,
$`\mathrm{\Delta }(q)=G\mathrm{\Psi }\left|{\displaystyle \underset{r}{}}S_{r+}(q)\right|\mathrm{\Psi }={\displaystyle \underset{r}{}}\mathrm{\Delta }_r(q)={\displaystyle \underset{r}{}}Gu_rv_r\left\{\mathrm{\Omega }_r\right\}_q.`$
Again the terms involving products of the type $`v_r^3u_r^m`$ have been ignored. After these considerations, we can write the matrix element of the Hamiltonian $`H`$ as
$`\mathrm{\Psi }\left|H\right|\mathrm{\Psi }={\displaystyle \underset{r}{}}\left(2\epsilon _r\mathrm{\Omega }_rv_r^2Gv_r^2\mathrm{\Omega }_r\left\{\mathrm{\Omega }_r\right\}_q+Gv_r^4(\mathrm{\Omega }_r1)\left\{\mathrm{\Omega }_r\right\}_q+{\displaystyle \frac{\mathrm{\Delta }_r^2(q)}{G}}\right){\displaystyle \frac{\left(\mathrm{\Delta }(q)\right)^2}{G}}`$
## III $`q`$BCS Gap equation and the Ground State Energy
In order to evaluate the ground state energy of $`N`$ nucleons, we minimize the expectation value of the Hamiltonian subject to the number constraint by varying $`v_j`$ and obtain $`m`$ equations to be solved self consistently ,
$$4(\epsilon _j^{}\lambda )v_j\mathrm{\Omega }_j2\mathrm{\Delta }(q)\left\{\mathrm{\Omega }_j\right\}_q\left(\frac{12v_j^2}{u_j}\right)4Gv_j^3\left\{\mathrm{\Omega }_j\right\}_q\left(\left\{\mathrm{\Omega }_j\right\}_q\mathrm{\Omega }_j+1\right)=0,$$
(9)
where $`\epsilon _j^{}=\epsilon _j+\frac{G\left\{\mathrm{\Omega }_j\right\}_q\left(\left\{\mathrm{\Omega }_j\right\}_q\mathrm{\Omega }_j\right)}{2\mathrm{\Omega }_j}`$. Leaving out for the time being, the term containing $`u_jv_j^3,`$ we solve these equations to obtain the occupancies,
$$v_j^2=0.5\left(1\frac{\epsilon _j^{}\lambda }{\sqrt{\left(\epsilon _j^{}\lambda \right)^2+\left(\mathrm{\Delta }(q)\frac{\left\{\mathrm{\Omega }_j\right\}_q}{\mathrm{\Omega }_j}\right)^2}}\right),$$
(10)
gap parameter
$$\mathrm{\Delta }(q)=\underset{j}{}G\left\{\mathrm{\Omega }_j\right\}_q0.5\left(1\frac{\left(\epsilon _j^{}\lambda \right)^2}{\left(\epsilon _j^{}\lambda \right)^2+\left(\mathrm{\Delta }(q)\frac{\left\{\mathrm{\Omega }_j\right\}_q}{\mathrm{\Omega }_j}\right)^2}\right)^{\frac{1}{2}}$$
(11)
and consequently the gap equation
$$\frac{G}{2}\underset{j}{}\frac{\left\{\mathrm{\Omega }_j\right\}_q^2}{\sqrt{\left(\epsilon _j^{}\lambda \right)^2\mathrm{\Omega }_j^2+\left(\mathrm{\Delta }(q)\left\{\mathrm{\Omega }_j\right\}_q\right)^2}}=1.$$
(12)
To include the effect of terms containing $`u_jv_j^3`$ left out earlier, we now replace $`\lambda `$ by
$$\lambda (q)=\lambda +\frac{Gv_j^2\left\{\mathrm{\Omega }_j\right\}_q}{\mathrm{\Omega }_j}\left(\left\{\mathrm{\Omega }_j\right\}_q\mathrm{\Omega }_j+1\right).$$
(13)
The ground state BCS energy, $`\mathrm{\Psi }\left|H\right|\mathrm{\Psi }`$ is
$$E_{bcs}(q)=\underset{j=1}{\overset{m}{}}\left(2\epsilon _j^{}\mathrm{\Omega }_jv_j^2Gv_j^4\left\{\mathrm{\Omega }_j\right\}_q\left(\left\{\mathrm{\Omega }_j\right\}_q\mathrm{\Omega }_j+1\right)\right)\frac{\left(\mathrm{\Delta }(q)\right)^2}{G}$$
(14)
We notice that in a very natural way, the SU<sub>q</sub>(2) symmetry introduces in the interaction energy, a $`q`$ dependence which is linked to the $`j`$-value of the orbit occupied by the zero coupled nucleon pairs.
## IV Single orbit with 2$`\mathrm{\Omega }`$ degenerate states
A very special situation arises, when the $`N`$ nucleons occupy a single orbit with an occupancy of $`2\mathrm{\Omega }`$. Using the results of the previous section, the ground state wave function is now $`\mathrm{\Psi }=\mathrm{\Phi }_j`$ and the ground state energy $`E_{bcs}(q)`$ is
$$E_{bcs}(q)=\epsilon _jNG\left\{\mathrm{\Omega }_j\right\}_q\frac{N}{4\mathrm{\Omega }}\left(2\left\{\mathrm{\Omega }_j\right\}_qN+\frac{N}{\mathrm{\Omega }}\right)$$
(15)
to be compared with the exact energy of the $`N`$ nucleon zero seniority state,
$$E_{exact}=\epsilon _jNG^{}\frac{N}{4}\left(2\mathrm{\Omega }_jN+2\right)$$
(16)
We notice that we can have $`E_{bcs}(q)=E_{exact}`$ by choosing $`q`$ value and pairing strength $`G`$ such that
$`G={\displaystyle \frac{G^{}\mathrm{\Omega }_j\left(2\mathrm{\Omega }_jN+2\right)}{\left\{\mathrm{\Omega }_j\right\}_q\left(2\left\{\mathrm{\Omega }_j\right\}_qN+\frac{N}{\mathrm{\Omega }}\right)}}`$
for the choice $`\epsilon _j=0.0`$. For the special case of nuclear sdg major shell with $`\mathrm{\Omega }=16`$ , and $`4,14,20,30`$ valence nucleons occupying degenerate $`1d_{\frac{5}{2}},0g_{\frac{7}{2}},2s_{\frac{1}{2}},1d_{\frac{3}{2}},`$ and $`0h_{\frac{11}{2}}`$ orbits, we plot $`G`$ versus $`q`$ in Fig.1 such that $`E_{bcs}(q)=E_{exact}(G^{}=0.187`$ MeV, $`\epsilon _j=0.0`$ for all levels$`)`$. The intensity of pairing strength required to reproduce $`E_{exact}`$ is seen to fall with increasing $`q`$ and ultimately $`G0`$ for all cases. From the plot at hand we can say that strongly coupled zero coupled pairs of BCS theory may well be replaced by weakly coupled $`q`$-deformed zero coupled pairs of $`q`$BCS theory. The natural question is, is it possible to replace the pairing interaction by a suitable commutation relation between the pairs determined by a characteristic $`q`$ value for the system at hand? To get some clues to the answer, we next consider real nuclei for which we can get the pairing gap from the experiments.
## V Sn Isotopes
We examine the heavy Sn isotopes with $`N=14,16,18,20,22,`$ and $`24`$ neutrons outside $`{}_{}{}^{100}{}_{50}{}^{}`$Sn<sub>50</sub> core. The model space includes $`1d_{\frac{5}{2}},0g_{\frac{7}{2}},2s_{\frac{1}{2}},1d_{\frac{3}{2}},`$ and $`0h_{\frac{11}{2}}`$, single particle orbits, with excitation energies $`0.0`$, $`0.22`$, $`1.90`$, $`2.20`$, and $`2.80`$ MeV respectively. Fig. 2 is a plot of pairing correlations function $`D=\mathrm{\Delta }(q)/\sqrt{G}`$ versus $`G`$ for $`N=20`$ in the cases where deformation parameter takes some typical successively increasing values varying from $`1.0`$ to $`1.7`$. We notice that in $`{}_{}{}^{120}{}_{50}{}^{}`$Sn<sub>70</sub>, pairing correlations increase as $`q`$ increases if the pairing strength $`G`$ is kept fixed. For $`q=1.0`$ that is conventional BCS theory the pairing correlation vanishes for $`G<G_c(0.065`$ MeV$`)`$ as expected. As the deformation $`q`$ of zero coupled pairs increases we find $`D`$ going to zero for successively lower values of coupling strength, for example $`G_c0.04`$ MeV for $`q=1.3`$ . We may infer that the $`q`$BCS takes us beyond BCS theory.
The sets of $`G,q`$ values that reproduce the empirical $`\mathrm{\Delta }`$ for $`{}_{}{}^{120}{}_{50}{}^{}`$Sn<sub>70</sub>, are used to calculate the gap parameter $`\mathrm{\Delta }`$ and the ground state BCS energy E$`_N,`$ for even isotopes <sup>114-124</sup>Sn displayed in Fig. 3. The experimental values of $`\mathrm{\Delta }`$ (filled triangles up) are also shown. As far as the gap parameter $`\mathrm{\Delta }`$ is concerned all the sets of $`G,q`$ values fair equally in comparison with the experiment. The ground state energies from $`q`$BCS are however in general lower than those calculated by using BCS. The underlying $`q`$-deformed nucleon pairs show increasingly strong binding as the value of $`q`$ is increased. It opens the possibility of obtaining the exact correlation energies by choosing appropriately the combination of $`G,q`$ values.
## VI Conclusions
By looking at the results for $`4,14,20,30`$ valence nucleons in nuclear degenerate sdg major shell, we find that the strongly coupled zero angular momentum nucleon pairs may be replaced by weakly coupled $`q`$-deformed zero angular momentum nucleon pairs. The study of a realistic case i.e. Sn isotopes also indicates that their is a well defined universe of sets of values for pairing strength $`G`$ and deformation parameter $`q`$, for which $`q`$BCS converges and has a non-trivial solution. For $`{}_{}{}^{120}{}_{50}{}^{}`$Sn<sub>70</sub> we observe that by choosing the pairing strength $`G0.217`$ MeV a matching value of deformation parameter $`q`$ can be found such that the experimental pairing gap is reproduced. For the choice $`G=0.07`$ MeV for example a large deformation of $`q=1.7`$ is needed to reproduce the empirical $`\mathrm{\Delta }`$ for $`{}_{}{}^{120}{}_{50}{}^{}`$Sn<sub>70</sub>. The results of $`q`$BCS for Sn isotopes are not much different from BCS as far as the Gap parameter $`\mathrm{\Delta }`$ is concerned. The ground state binding energies are however lowered by the deformation. The pairing correlations, measured by $`D=\mathrm{\Delta }(q)/\sqrt{G}`$, are seen to increase as $`q`$ increases (for $`q`$ real) while the pairing strength $`G`$ is kept fixed, in Sn isotopes. It is immediately seen that $`q`$ parameter is a very good measure of the pairing correlations left out in the conventional BCS theory.
The results of our present study are consistent with our earlier conclusions that the $`q`$-deformed pairs with $`q>1`$ $`(q`$ real) are more strongly bound than the pairs with zero deformation and the binding energy increases with increase in the value of parameter $`q`$. In contrast by using complex $`q`$ values one can construct zero coupled deformed pairs with lower binding energy in comparison with the no deformation zero coupled nucleon pairs. In general the pairing correlations in $`N`$ nucleon system, measured by $`D=\mathrm{\Delta }(q)/\sqrt{G}`$ , increase with increasing $`q`$ (for $`q`$ real) and $`q`$BCS takes us beyond the BCS theory. The formalism can be tested for several other systems, for example metal grains, where cooper pairing plays an important role.
Acknowledgments
S. Shelly Sharma acknowledges support from Universidade Estadual de Londrina.
|
warning/0002/hep-th0002067.html
|
ar5iv
|
text
|
# Evidence for Winding States in Noncommutative Quantum Field Theory
## 1 Introduction
Gauge theories on noncommutative spaces can be obtained by taking the infinite tension limit of string theory in the presence of a very strong B-field . One can thus look upon these field theories as interesting limits of string theory in such an extreme environment. Regardless of their ancestry, these field theories have very interesting properties , making their study fascinating in its own right. Many unusual properties already emerge in scalar field theories. In particular, one sees the emergence of unexpected infrared behavior in correlation functions (we will refer to this paper in the following as MSV) unlike anything seen in conventional field theories. These long distance correlations appear in massive theories and are not related to any massless fields in the Lagrangian. Rather, these infrared effects are due to the Moyal phases. As can be seen in Feynman diagrams, it is the integration over the high ultraviolet modes that is responsible for this peculiar infrared behavior. The UV and IR are intertwined in a manner never seen in local field theory on commutative spaces. MSV have a proposal for what kind of degrees of freedom are responsible for these IR correlations: their claim is that in the zero slope limit and large B field, the closed string states do not decouple.
To gain some more insight into the structure of these noncommutative field theories, we subject them to a heat bath and consider various thermodynamic quantities. We will show that the partition functions of these theories get contributions at high temperature from configurations winding around the temporal direction. This implies that there are extended objects in these theories that have the ability to wind, which seems consistent with the suggestions of MSV. We also test this picture by compactifying the spatial commuting direction to a circle and again find contributions to the partition function, this time from configurations winding around the spatial circle.
It is useful to consider a simple field theory like massive $`\varphi ^4`$ and the supersymmetric Wess-Zumino model. We were motivated to include a supersymmetric example in order to show that the existence of winding states persists in theories where the sensitivity to the UV physics is softer. Because of the special UV properties of the supersymmetric case, the IR sensitivity is milder and allows us to limit ourselves to the lowest order in perturbation theory when calculating thermodynamic quantities.
The paper is organized as follows: in section 2 we will describe in some detail the thermodynamics of noncommutative $`\varphi ^4`$. There we present the calculation of the free energy to $`O(g^2)`$ and discuss the various divergences and also show how the winding states appear. In section 3, we investigate whether supersymmetric systems whose UV behavior are less singular affect the existence of winding states. For this purpose, we consider the Wess-Zumino model and find again the existence of winding states. We calculate the free energy, from which the entropy, internal energy and specific heat can be derived. In section 4, we return to the $`\varphi ^4`$ theory and show how to calculate the free energy in the “mean field” approximation. In section 5, we discuss the implications of these results and briefly repeat the calculations at zero and finite temperature for the case where the sole commuting direction is a circle. We show that at finite temperature, there are no membrane states wrapping around the torus spanned by the temporal and $`x_3`$ cycles. Instead, we find winding states winding each of the cycles independently.
We then conclude with some speculations.
## 2 The thermodynamics of noncommutative $`g^2\varphi ^4`$ in perturbation theory
We present a perturbative calculation of the free energy, F, which assumes that the coupling constant $`g^2`$ is small. As we will see, the appearance of IR divergences invalidates the perturbative expansion. Nevertheless, we believe it is instructive to go through this exercise as it already shows the existence of winding states. Later in section 4 we will have to use a mean field approximation to ensure that long range correlations are properly screened.
In this paper, we will restrict ourselves to the case where the noncommutativity is solely among the spatial directions.. Without loss of generality then in this 4-dimensional case, we can choose the (1,2) plane to be noncommutative and leave time and the third spatial dimension commutative.
At finite temperature, the Feynman rules are the ones that correspond to a theory with “time” being a circle of radius $`\beta `$, where $`\beta `$ is the inverse temperature, $`\beta =1/T`$. This implies that the integrals over frequencies in diagrams are replaced by discrete sums over the so-called Matsubara frequencies. The noncommutativity in the (1,2) plane manifests itself in Feynman diagrams through the appearance of Moyal phases in vertices. These Moyal phases depend on momenta in the (1,2) directions.
When evaluating the contributions to the free energy, one needs to consider both planar and non-planar diagrams. The planar contributions are very much like in the commutative theory except for some combinatoric factors . Inspite of the changed combinatorics, the behavior in the UV is still controlled by zero temperature physics. In order to illustrate the divergences that occur in the perturbative expansion of the free energy in the noncommutative theory, we will first concentrate on the non-planar contributions to lowest order in $`g^2`$. This will also enable us to show the appearance of contributions from winding states to the free energy.
The leading non-planar contribution at order $`g^2`$ to the free energy comes from a two-loop diagram.
Fig.1: Planar and nonplanar two loop contributions to the free energy.
This diagram is divergent, where some of the divergences can be interpreted as originating in the UV and some divergence can be attributed to the IR. The UV divergences will be shown to arise from zero temperature physics and similar UV divergences will appear in higher order terms in the expansion in $`g^2`$. Also, in a manner similar to the IR divergences that appears at lowest order, each successive order of perturbation theory will be plagued by IR divergences. These IR divergences grow more singular with increasing order in the perturbative expansion, invalidating the expansion. We will return to the formal application of this mean field approximation in section 4 after the study of the $`O(g^2)`$ contribution to the free energy in the Wess-Zumino model.
The contribution to the free energy from this two-loop non-planar diagram is:
$$g^2T^2\underset{n=\mathrm{}}{\overset{\mathrm{}}{}}\underset{l=\mathrm{}}{\overset{\mathrm{}}{}}\frac{d^3p}{(2\pi )^3}\frac{d^3k}{(2\pi )^3}\frac{e^{ip\theta k}}{(\frac{4\pi ^2n^2}{\beta ^2}+p^2+M^2)(\frac{4\pi ^2l^2}{\beta ^2}+k^2+M^2)},$$
(1)
where $`p\theta k=\theta (p_1k_2p_2k_1)`$ and $`p^2=p_{1}^{}{}_{}{}^{2}+p_{2}^{}{}_{}{}^{2}+p_{3}^{}{}_{}{}^{2}`$.
We will be interested in temperatures much larger than the mass, so that to leading order in $`\beta M`$, it can be neglected.
The integration over the momenta can be done by introducing Schwinger parameters. In order to show how contributions of winding configurations appear, we will look at equation (1) in some detail. The integral appearing in this expression can after some manipulations be rewritten as:
$$g^2T\underset{n,l}{}\frac{d^3p}{(2\pi )^3}\frac{1}{(\frac{4\pi ^2n^2}{\beta ^2}+p^2)(4\pi ^2(l^2\beta ^2+|\theta p|^2)+\mathrm{\Lambda }^2)},$$
(2)
where $`|\theta p|=\theta (p_{1}^{}{}_{}{}^{2}+p_{2}^{}{}_{}{}^{2})^{1/2}`$ and $`\mathrm{\Lambda }`$ is an ultraviolet cutoff similar to the one used by MSV. We also have neglected the mass in obtaining this expression since we are working in the regime $`\beta M1`$.
The sums over $`n`$ and $`l`$ can be performed and will lead to the following formula involving Bose distributions
$`g^2T{\displaystyle \frac{d^3p}{(2\pi )^3}\frac{1+2n_\beta (|p|)}{2|p|}\frac{1+2n_{1/\beta }(2\pi |\theta p|)}{4\pi |\theta p|}},`$ (3)
where $`|p|=(p_{1}^{}{}_{}{}^{2}+p_{2}^{}{}_{}{}^{2}+p_{3}^{}{}_{}{}^{2})^{1/2}`$ and $`n_\beta (|p|)=\frac{1}{e^{\beta |p|}1}`$ is a Bose distribution at temperature $`T=1/\beta `$.
Note the appearance in the last expression of two Bose distributions, one at temperature $`T`$ and the other at temperature $`1/T`$. This would be a rather bizarre state of affair for a system in thermal equilibrium. One possibility is that the system is not in equilibrium, but why then in this very special way, with one set of the degrees of freedom distributed thermally at temperature $`T`$ and the other at temperature $`1/T`$? We will choose the option that the system is indeed in equilibrium, and that there are configurations, if one thinks in a path integral approach to the partition function, that wind around the temporal circle. If one looks at the integral appearing in equation (3), one can see the contributions of winding states with momentum $`|\theta p|`$ to the free energy.
Another interesting feature of equation (3) is that the $`p_3=0`$ sector is invariant under the interchange of $`\beta `$ and $`2\pi \theta /\beta `$. This transformation looks like T-duality where the string scale $`l_{s}^{}{}_{}{}^{2}`$ is replaced by $`\theta `$. We should remind the reader though that this property was obtained in the $`\beta M1`$ limit. We do not know whether this duality invariance persists beyond this order to the full theory.
The expression in equation (3) has UV divergences and IR divergences. The IR divergence is linear and comes from the small $`(p_1,p_2)`$ region of integration. This divergence, as will be shown later, is cured by the mean field approximation.
The UV divergent terms in the non-planar sector are calculated in the appendix. Apart from a zero temperature contribution to the vacuum energy, there is a linearly divergent term which is also linear in the temperature. This is a rather unusual contribution to the free energy. Indeed, this term does not contribute to the internal energy of the system but does provide a temperature independent contribution to the entropy. This contribution to the entropy is due to the degeneracy of the winding states. As can be seen in the Bose distribution, their dispersion relation does not involve the momentum along the commuting spatial dimension, $`p_3`$. The very large number of winding states with momenta $`p_{1,2}=0`$ are the source of the linear divergence. This divergence disappears when screening of the large distance correlations is taken into account. The supersymmetric case does not have such a divergence at the leading order in $`g^2`$ because of its milder behavior in the IR.
## 3 The thermodynamics of the noncommutative Wess-Zumino model
The Lagrangian for the noncommutative Wess-Zumino model is:
$$=i_\mu \overline{\psi }\overline{\sigma }^\mu \psi +A^{}A\frac{1}{2}M\psi \psi \frac{1}{2}M\overline{\psi }\overline{\psi }g\psi \psi Ag\overline{\psi }\overline{\psi }A^{}F^{}F,$$
where $`F`$ is given by
$$F=MA^{}gA^{}A^{}.$$
The calculation of the free energy to $`O(g^2)`$ proceeds along the same lines as in the bosonic case. In this case there are additional contributions due to new interactions among the bosons and bosons with fermions. The fermions, as usual, have frequencies that are odd multiples of the temperature.
Three pairs of diagrams contribute at order $`g^2`$, each pair consisting of a planar and a non-planar piece. They sum to give
$`F/V`$ $`=`$ $`g^2{\displaystyle \frac{d^3p}{(2\pi )^3}\frac{d^3k}{(2\pi )^3}\frac{1+e^{ip\theta k}}{\omega _p\omega _k}\left(n_B(\omega _p)+n_F(\omega _p)\right)\left(n_B(\omega _k)+n_F(\omega _k)\right)}`$ (4)
$`+`$ $`g^2T^4𝒪({\displaystyle \frac{M^2}{T^2}})`$
with $`\omega _p=\sqrt{p^2+M^2}`$ and $`n_{B,F}(\omega _p)=1/(e^{\beta \omega _p}\mathrm{\hspace{0.17em}1})`$. Notice that this expression for the free energy at high temperature ($`\beta M1`$) is “T- duality invariant” as in the bosonic case. This property applies to the sector with zero momentum along the commuting direction. This result can be seen by expanding one of the Bose distributions in a series of exponentials,
$$\frac{1}{e^{\beta (p^2+M^2)^{1/2}}1}=\underset{n=1}{\overset{\mathrm{}}{}}e^{\beta n(p^2+M^2)^{1/2}}.$$
Then performing the integral over $`d^3p`$, neglecting the mass, produces the series,
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{1}{n^2\beta ^2+|\theta k|^2}.$$
A similar expression can be obtained for the integration over the Fermi distributions, except that in the fermion case the series is an alternating one. The presence of winding states is then clear as well as the “T- duality invariance” of the free energy.
Fig.2: Planar and non-planar two loop fermionic contributions to the free energy.
Fig.3: Planar and non-planar two loop bosonic contributions to the free energy.
The expression (4) for the non-planar contribution to the free energy can be further evaluated to yield
$$\frac{(2\pi )^3}{\sqrt{2}}\frac{g^2}{\theta \beta ^2}_0^\pi 𝑑\eta _0^{\mathrm{}}𝑑q\frac{\mathrm{tanh}(\frac{\pi sin(\eta )}{2}\frac{\theta q}{\beta ^2})}{\mathrm{sinh}q}.$$
(5)
At high temperature, $`\beta M1`$, we consider two limiting cases, $`T^21/\theta `$ and $`T^21/\theta .`$ <sup>1</sup><sup>1</sup>1We also assume that $`\theta M^21`$, i.e. the Compton wavelength of the scalars is larger than the length scale associated to the noncommutativity. The first limit corresponds to the case where the thermal wavelength is bigger than the length scale associated to the noncommutativity. In this limit, one might expect to recover conventional high temperature behavior. Indeed, the leading non-planar contribution to the free energy here is
$$\frac{F}{V}|_{\mathrm{np}}g^2T^4.$$
(6)
In the other limit, $`T^21/\theta `$, the thermal wavelength is smaller than the noncommutativity scale. We therefore expect novel behavior. We find the free energy in this regime to be
$$\frac{F}{V}|_{\mathrm{np}}g^2\frac{T^2}{\theta }\mathrm{log}T^2\theta ,$$
(7)
where the logarithm was obtained by numerical methods. In this case, the contribution to the free energy resembles the contribution of a massless gas in 1+1 dimension, modified by a logarithm.
One is tempted to conclude from these results that there is a drastic reduction of degrees of freedom in the non-planar sector! For thermal wavelengths larger than the noncommutativity scale, the system behaves as a relativitic $`3+1`$ dimensional gas. For wavelengths smaller than this scale, the system behaves roughly like a $`1+1`$ dimensional gas. As the temperature is raised past $`1/\theta ^{1/2}`$ it looks like the winding states do behave as strings. Up to the logarithm, one is reminded by the temperature dependence in equation (7) of the result obtained by Atick and Witten for the high temperature behavior of strings. We do not know whether this is just a coincidence or whether there is some deeper significance to this behavior. A test of this high temperature dependence of the free energy due to the non planar sector would be to study higher dimensional examples where there are more noncommuting spatial dimensions. It is amusing to note that the non-planar sector of this system exhibits, to leading order at very high temperature, the equation of state $`p=\rho `$, $`p`$ being pressure and $`\rho `$ energy density.
## 4 The thermodynamics of noncommutative $`g^2\varphi ^4`$ in the “mean field” approximation
We now go back to the noncommutative $`g^2\varphi ^4`$ theory of section 2. There, the appearance of IR divergences ruins the validity of the perturbative expansion (in $`g^2`$) for the free energy. This is a familiar phenomenon in thermal physics. For example, in the case of a plasma of electric charges in QED, one finds IR divergences in perturbation theory for the free energy. This can be traced to the large distance behavior of the correlation function, $`\rho (x)\rho (0).`$ This IR divergence disappears after summing a series of diagrams contributing to the correlation function between charge densities, $`\rho (x)\rho (0).`$ This series of diagrams is a geometric series in the one-particle irreducible charge density correlation function. When this series is summed, one finds that the charge densities are screened. The screening mass or Debye mass is the IPI charge density correlation function evaluated at zero momentum.
In the case of noncommutative $`\varphi ^4`$, the IR divergence already appears at zero temperature and is again screened by summing a geometric series in the one-particle irreducible two-point function (MSV).
In the finite temperature case, we will approach the problem of IR divergences by treating the gas in a mean field approximation. In this approximation, each quantum of the scalar field is moving in the average field produced by the other scalars. Unlike more traditional cases where the mean field approximation is used, this case has the particular feature that the mean field will depend on the momentum of the particle that moves in this background. This momentum lies in the noncommutative plane. Using this approximation, we will show how the long range correlations are screened.
The Hamiltonian H is:
$$H=\frac{1}{2}d^3x(\mathrm{\Pi }^2(x)+(\varphi (x))^2+M^2\varphi ^2(x))+g^2d^3x\varphi (x)\varphi (x)\varphi (x)\varphi (x).$$
The free energy F is obtained from the logarithm of the partition function $`Z`$, where $`Z=\mathrm{tr}e^{\beta H}`$, as $`\beta F=\mathrm{log}Z`$. When $`g^2`$ is small, it can be formally written as $`F_0`$, the free gas contribution to F, corrected by $`O(g^2)`$ contributions. We should remind the reader that this procedure fails because of IR divergences. We present this expression for F here because it will be useful when we get to the mean field approximation:
$$\frac{\beta F}{V}=\frac{d^3p}{(2\pi )^3}[\frac{\beta }{2}(p^2+M^2)^{1/2}+\mathrm{log}(1\mathrm{exp}(\beta (p^2+M^2)^{1/2}]+O(g^2).$$
The contribution from the free gas to the previous formula could have been obtained by calculating a one-loop diagram in the path integral formulation.
In the mean field approximation, a picture emerges where the system looks like a free gas for which the dispersion formula for the energy of a quantum is modified by the average effect of the other quanta.
Formally, this will be implemented in the path integral by using the following action:
$`S={\displaystyle _0^\beta }𝑑\tau {\displaystyle \frac{d^3k}{(2\pi )^3}\left[(_t\varphi )^2(k)k^2\varphi ^2(k)M^2\varphi ^2(k)g^2\varphi ^2_{\theta k}\varphi ^2\left(k\right)\right]}.`$ (8)
where $`\varphi ^2_{\theta k}`$ satisfies an integral equation:
$$\varphi ^2_{\theta k}=1/\beta \underset{n}{}\frac{d^3p}{(2\pi )^3}\frac{2+e^{ip\theta k}}{\frac{4\pi ^2n^2}{\beta ^2}+p^2+M^2+g^2\varphi ^2_{\theta p}}.$$
We will not solve this integral equation exactly but rather find the solution to it as a power series expansion in $`g^2`$. In this paper we will confine ourselves to work with the solution to leading order.
In the limit of high temperature, $`M\beta 1`$, we neglect to first order the mass M and we sum over $`n`$, and obtain to leading order $`\varphi ^2_{\theta k}`$.
$`\varphi ^2_{\theta k}`$ $`=`$ $`{\displaystyle \underset{n}{}}{\displaystyle \frac{1}{4\pi ^2n^2\beta ^2+(2\pi |\theta k|)^2+\mathrm{\Lambda }^2}}`$ (9)
$`=`$ $`{\displaystyle \frac{1}{2\beta \sqrt{(2\pi |\theta k|)^2+\mathrm{\Lambda }^2}}}(1+2n_{1/\beta }(\sqrt{4\pi ^2|\theta k|^2+\mathrm{\Lambda }^2}).`$ (10)
Notice that the Bose distribution in the expression for $`\varphi ^2`$ is unusual as it distributes $`k_{1,2}`$ according to a “temperature” $`\tau `$, where $`\tau =1/\theta T`$, i.e. proportional to the inverse temperature. One can interpret this property, as was argued earlier, to be evidence for winding states. Indeed, the summation that appears in $`\varphi ^2`$ is over an integer multiple of the length of the temporal circle rather than the inverse radius as would be the case for momenta. The free energy can then be written as a “tree” contribution plus a one-loop contribution.
$$\frac{\beta F}{V}=M^2d^4x\varphi ^2(x)+\frac{\beta g^2}{V}d^4x\varphi (x)\varphi (x)\varphi (x)\varphi (x)+$$
$$\frac{d^3p}{(2\pi )^3}[\frac{\beta }{2}\sqrt{p^2+M^2+g^2\varphi ^2_{\theta p}}+\mathrm{log}(1\mathrm{exp}(\beta (p^2+M^2+g^2\varphi ^2_{\theta p})^{1/2})].$$
(11)
In this last equation, the average value of
$$M^2d^4x\varphi ^2(x)+\frac{\beta g^2}{V}d^4x\varphi (x)\varphi (x)\varphi (x)\varphi (x)$$
is evaluated by using the action S, equation (8), in a path integral calculation.
As expected and demonstrated explicitly in the appendix, the IR divergences disappear.
The one-loop contribution
$$\frac{d^3p}{(2\pi )^3}[\frac{\beta }{2}\sqrt{p^2+M^2+g^2\varphi ^2_{\theta p}}+\mathrm{log}(1\mathrm{exp}(\beta (p^2+M^2+g^2\varphi ^2_{\theta p})^{1/2})]$$
can be visualized as a single quantum of the scalar field propagating around a loop with a momentum that at $`O(g^0)`$ is thermally distributed at temperature $`T`$. This particle is moving in the presence of a background, $`\varphi ^2`$, which it interacts with at $`O(g^2)`$ and that consists of states winding around the temporal circle of radius $`1/T`$. These winding states have momenta solely in the noncommutative plane. It is quite tempting to think of them as closed strings reminiscent of the suggestions made in MSV.
To be complete, one should study the UV divergences and see whether they are associated to zero temperature effects. We will not attempt this study here. It should be kept in mind that it is not completely clear yet whether this theory is renormalizable, even though some evidence exists.
## 5 Winding states around spatial cycles
The presence of winding states, which in the finite temperature case wind around the temporal circle, can also be noticed when one compactifies the commuting spatial dimension, $`x_3`$. The states winding around $`x_3`$ can be detected for example if we calculate the free energy for this case.
The calculation of the previous section can be repeated, with each $`\frac{dp_3}{2\pi }`$ replaced by $`\frac{1}{2\pi L}_l`$, where l is an integer.
To show the existence of winding states, it is sufficient to again consider the integral equation for the two point function:
$$\varphi ^2_{\theta k}=\frac{1}{\beta L}\underset{n,l}{}\frac{d^2p}{(2\pi )^2}\frac{e^{ip\theta k}}{\frac{4\pi ^2n^2}{\beta ^2}+\frac{4\pi ^2l^2}{L^2}+p^2+M^2+g^2\varphi ^2_{\theta p}}.$$
In the limit of high temperature, $`\beta M1`$, and small radius, $`LM1`$, we find for $`\varphi ^2_{\theta k}`$, upon performing the $`d^2p`$,
$$\varphi ^2_{\theta k}=\underset{n,l}{}\frac{1}{4\pi ^2n^2\beta ^2+4\pi ^2l^2L^2+(2\pi |\theta k|)^2+\mathrm{\Lambda }^2}.$$
This last expression clearly shows the presence of winding states around both circles. Again these states become prevalent in the small radii limit, which is to be expected since in these limits these states are very “light”.
This formula for $`\varphi ^2_{\theta k}`$ also suggests that there are no wrapping states around the 2-torus spanning the temporal and $`x_3`$ dimensions, which would look like an additional piece $`(nl\beta L)^2`$ in the denominator of this equation.
The same result holds in the supersymmetric case. In some ways, this could have been expected. Imagine subjecting this system to a thermal environment. In this case, the bosons and fermions have different distributions and we should not therefore expect a drastically different behavior from the non-supersymmetric cases.
One might speculate about the existence of multidimensional extended objects in higher dimensional theories where several $`\theta _{ij}0`$. To test this, we briefly consider the example of the six-dimensional $`\varphi ^3`$ theory, with various $`\theta _{ij}0`$.
At high temperature and after compactifying the commuting $`x_5`$ on a small circle, it should be possible to see whether or not wrapping states emerge. This can be tested, as we learned above, simply by calculating $`\varphi ^2_{\theta p,\theta ^{}p^{}}`$. We obtain
$$\varphi ^2_{\theta p,\theta ^{}p^{}}=\underset{l,n}{}\frac{1}{(4\pi ^2p_i\theta ^{ij}\theta _{jk}p^k+4\pi ^2\beta ^2n^2+4\pi ^2L^2l^2)^2}.$$
We thus conclude that there are no wrapping states, just winding states.
## 6 Conclusions
What is the physics that is responsible for the existence of these winding states?
Clearly the UV-IR connection that exists in the noncommutative plane is crucial. The expressions which reveal the existence of the winding states show that this peculiar IR behavior originates in the UV region of integration. One might then conclude that the uncertainty relationship between $`x_1`$ and $`x_2`$ in the case where $`\theta _{12}0`$ is implemented by or implies the existence of one-dimensional structures. In other words, measuring $`x_1`$ sharply renders the other direction $`x_2`$ totally uncertain and maybe this is intimately related to the existence of extended one-dimensional objects in a rather mysterious way. These “strings” can then wind around various cycles. However, we did not find multidimensional objects wrapping multicycles in higher dimension.
We have discovered from studying the high temperature regime of the supersymmetric theory the existence of two different behaviors. We believe that these results are valid in general, but we have only been able to calculate reliably to leading order in the supersymmetric case. When the thermal wavelength is bigger than the noncommutativity length scale, we find that the non-planar sector behaves as a conventional $`3+1`$ dimensional relativistic gas. For the thermal wavelength smaller than the noncommutativity scale, we observe a reduction of the degrees of freedom in the non-planar sector. The non-planar sector now behaves as a $`1+1`$ (not, as might have been expected, $`2+1`$) dimensional gas. This seems to indicate a drastic reduction of the degrees of freedom in that sector. Equivalently, we find the equation of state of this sector is approximately that of an incompressible gas with equation of state $`p=\rho `$.
It would be interesting to see whether this behavior persists in higher dimensional theories.
In the theories we have studied, the fascinating behavior of the non-planar sector is overshadowed by the planar sector. Thus, the challenging question remains of what theory captures exclusively the physics of the non-planar sector.
###### Acknowledgments.
The work of WF, EG, JG, AK-P, SP, PP is supported in part by the Robert Welch Foundation and the NSF under grant number PHY-9511632, JG is partially supported by AEN 98-0431 (CICYT), GC 1998SGR (CIRIT), SP is also supported by NSF grant PHY-9973543.
## Appendix A UV divergences in $`g^2\varphi ^4`$
Of the four terms in equation (3), the two damped by the Bose distribution $`n_\beta (|p|)`$ are obviously UV convergent. The remaining two give rise to a linear and a logarithmic UV divergence (we remind the reader that the argument $`|\theta p|`$ of the second Bose distribution $`n_{1/\beta }`$ appearing in equation (3) is independent of $`p_3`$).
The term
$$g^2T\frac{d^3p}{(2\pi )^3}\frac{n_{1/\beta }(|\theta p|)}{|p|\mathrm{\hspace{0.17em}4}\pi |\theta p|}$$
is divergent in the region $`p_3\sqrt{p_1^2+p_2^2}`$. In this region, we obtain
$$\frac{1}{2(2\pi )^3}\frac{g^2T}{\theta }\mathrm{log}\mathrm{\Lambda }𝑑p_{1,2}n_{1/\beta }(|\theta p|)=\frac{1}{2(2\pi )^3}\frac{g^2}{\theta ^2}\mathrm{log}\mathrm{\Lambda }𝑑x\frac{1}{e^{2\pi x}1}.$$
Notice that the $`x`$ integral is IR divergent. If we introduce an IR cutoff to regulate this divergence, the substitution we use to extract the temperature dependence of the expression renders this cutoff temperature dependent. We ignore this problem at this point, as we expect it to be an IR artifact that will be cured upon resummation of the appropriate diagrams.
The linear divergence arises as:
$$g^2T\frac{d^3p}{(2\pi )^3}\frac{1}{2|p|\mathrm{\hspace{0.17em}4}\pi |\theta p|}=\frac{1}{4(2\pi )^2}\frac{g^2T}{\theta }\mathrm{\Lambda }.$$
As discussed above, this divergence contributes to the entropy of the system, but not to the internal energy. We therefore attribute it to the degeneracy of the winding states as discussed in the text.
## Appendix B IR divergences in $`g^2\varphi ^4`$
The only potential infrared divergence in the mean field expression (11) for the free energy stems from the term
$$\beta Vd^4x\varphi (x)\varphi (x)\varphi (x)\varphi (x).$$
The non-planar contribution to this term is obtained from equation (1) by replacing the bare by the dressed propagators:
$$g^2T^2\underset{n,l}{}\frac{d^3p}{(2\pi )^3}\frac{d^3k}{(2\pi )^3}\frac{e^{ip\theta k}}{(\frac{4\pi ^2n^2}{\beta ^2}+p^2+M^2+g^2\varphi ^2_{\theta p})(\frac{4\pi ^2l^2}{\beta ^2}+k^2+M^2+g^2\varphi ^2_{\theta k})},$$
(12)
where $`\varphi ^2`$ is given to leading order by equation (10). For small $`p`$ and small $`k`$, both terms in the denominator of the above expression blow up and no infrared divergence can occur. For small $`p`$ and large $`k`$ (or vice versa), we can ignore $`\varphi ^2_{\theta k}`$ and perform the $`d^3k`$ integral as in (2). Upon performing the sums, we obtain for small $`p`$
$`g^2{\displaystyle \frac{d^3p}{(2\pi )^3}\frac{1+2n_\beta (|p|)}{2\sqrt{p^2+M^2+g^2\varphi ^2_{\theta p}}}\frac{2}{8\pi ^2|\theta p|^2}}.`$ (13)
As one factor of $`|\theta p|^1`$ is absorbed by the $`\varphi ^2_{\theta p}`$ under the square root, this expression is finite in the infrared.
|
warning/0002/astro-ph0002290.html
|
ar5iv
|
text
|
# Evolution and surface abundances of red giants experiencing deep mixingAccepted for publication in Astron. & Astrophys.
## 1 Introduction
The observed anomalies in CNO-, NeNa- and MgAl- elements in globular cluster red giants (see Kraft (1994) and Da Costa (1998) for reviews) are unexplained in canonical low-mass star evolution theory and indicate effects beyond the standard picture. At least for anomalies in those isotopes participating in the CNO-cycle models relying on the assumption of an additional, non-standard mixing process inside the stars have been presented, which explain convincingly the observations, including the evolution of the carbon abundance along the RGB, i.e. with time (see Charbonnel (1995); Denissenkov & Weiss (1996); Cavallo et al. (1998)). This mixing is supposed to set in after the hydrogen burning shell has reached the composition discontinuity left behind by the first dredge-up on the red giant branch (RGB), i.e., after the so-called RGB bump (e.g. Sweigart & Mengel (1979); Charbonnel (1995); Charbonnel et al. (1998)) where the molecular weight barrier between hydrogen burning shell and envelope is at a minimum. It is usually described in terms of a diffusion process of certain efficiency and penetration depth. The CNO anomalies and their correlations with brightness and metallicity (which are expected qualitatively from nucleosynthesis arguments in standard models, as discussed by Cavallo et al. (1998)) are thus explained by a purely evolutionary picture (Smith & Tout (1992); Denissenkov & Weiss (1996); Boothroyd & Sackmann (1999a)). The physical origin of the mixing process is believed to be found in differential rotation of the star and the parameters used in some of the presently available calculations have been derived from existing theories (e.g. Zahn (1992); Maeder & Zahn (1998)), which are, however, far from being complete.
A similar situation holds for oxygen and sodium, which are found to be anti-correlated in globular cluster red giants (Kraft et al. (1993)). Denissenkov & Denissenkova (1990) and Langer et al. (1993) showed that this could result from the mixing of elements participating in the ONeNa-cycle, which operates at higher temperatures than the CNO-cycle and therefore requires deeper mixing. Denissenkov & Weiss (1996) demonstrated how all anomalies of the mentioned elements known at that time can be explained by the deep mixing scenario. Their calculations were done by using canonical red giant models, which evolved along the RGB without any mixing, and performing the mixing and nuclear reactions in a post-processing way. For this approach to be correct it is necessary that the evolution of the background models is not affected by the mixing process. In fact, the mixing necessary to reproduce the observed anomalies was always so shallow that only very small amounts of hydrogen/helium were mixed between envelope and shell, even in the case of the O-Na-anti-correlation. This was taken as sufficient justification of the underlying basic assumption. Very similar conclusions were obtained by Cavallo et al. (1998).
Within this approach, one cannot investigate the possibility or necessity for even deeper mixing, which would affect the hydrogen/helium structure of the models. Sweigart (1997a) has renewed the interest in such deep mixing by connecting the problem of horizontal branch morphology with that of observed anomalies, as previously suggested by Langer & Hoffman (1995). If the mixing leads to severe helium enhancement in the envelope, increased luminosities and stellar winds result, such that the star will populate the blue horizontal branch (HB), while it remains a red HB star without the additional mixing.
While Denissenkov & Weiss (1996) did not investigate the effect of helium transport, Sweigart (1997a) did not follow the evolution of the participating isotopes to compare with observations. In the present paper, we therefore attempt to close this gap by computing full evolutionary sequences which include deep diffusive mixing and by investigating abundance anomalies using these self-consistent models as background models. In particular, we want to answer the question how much helium enrichment of the envelope is necessary or allowed to achieve or to keep consistency with observations.
In Sect. 2 we will discuss the nucleosynthesis aspects of the problem and review the observational status of the global O-Na anticorrelation, which is a powerful tracer of the transport processes in the red giants. In Sect. 3 we will present and discuss the evolutionary models. After that, the predictions for the abundances based on our mixed models and post-processing nucleosynthesis will follow, before the conclusions close the paper.
## 2 The O-Na anticorrelation: observational status and nucleosynthesis arguments
Among the chemical anomalies observed in red giant atmospheres over the past two decades, the variations in oxygen and sodium have a special status. Indeed, the so-called global oxygen-sodium anticorrelation appears to be a common feature to all the globular clusters in a wide range of metallicity (-2.5$``$\[Fe/H\]$``$-1), for which detailed abundance analysis of the brightest giants have been carried out (Gratton & Ortolani (1989); Sneden et al. (1991); Drake et al. (1992); Brown & Wallerstein (1992); Kraft et al. (1993); Armosky et al. (1994); Minniti et al. (1996); Pilachowski et al. (1996); Kraft et al. (1997); Kraft et al. (1998)). This pattern shows up in “normal” monometallic clusters (M3, M4, M5, M10, M13, M15, M92, NGC 7002) as well as in the multi-metallicity cluster $`\omega `$ Cen (Paltoglou & Norris (1989); Norris & Costa (1995)). Fig. 1 summarizes the present observational status concerning the O-Na-anticorrelation. But more importantly, this feature, and the dependence of the Na enhancement and O depletion on the red giant evolutionary state (Pilachowski et al. (1996); Kraft et al. (1997)) can be explained straightforwardly in the deep mixing scenario (Denissenkov & Weiss (1996)).
Regarding the Mg and Al anomalies, the situation is very different, and the observed Mg-Al anticorrelation (Shetrone (1996a)) requires a combination of the deep mixing and primordial scenarios (Denissenkov et al. (1997); Denissenkov et al. (1998)). Ad hoc assumptions would be needed to obtain the observed <sup>24</sup>Mg depletion within the low mass Al-rich giants themselves (Shetrone (1996b)). Since a low energy resonance in the <sup>24</sup>Mg(p,$`\gamma )^{25}`$Al reaction remains undetected (Angulo et al. (1999)) or even can be excluded (Powell (1999)), “exotic” models are required to episodically increase the temperature of the hydrogen-burning shell up to values as high as $``$ 70-85 MK (while canonical models reach a maximum temperature of only 55 MK) in order to deplete Mg at the expense of <sup>24</sup>Mg (Langer et al. (1997); Zaidins & Langer (1997); Fujimoto et al. (1999)) inside the low mass giants as apparently being the consequence of the results by Shetrone (1996a) (Shetrone (1996a) and Shetrone (1996b)) for M13. Lately, however, Ivans et al. (1999) found that Mg and Al abundance variations in M4 can be explained completely by the idea that the Al-enhancement is due to a destruction of the Mg-isotopes $`{}_{}{}^{25}\mathrm{Mg}`$ and $`{}_{}{}^{26}\mathrm{Mg}`$ (cf. Fig. 2). Even in this case, a significant increase of the initial abundance of <sup>25</sup>Mg is required (Denissenkov et al. (1998)). Since the Mg-Al anticorrelation cannot be explained by the deep mixing scenario alone, we do not consider it further in this paper. In fact, all explanations brought forward up to now being rather exotic and complicated, one should also wait for additional observational support for its existence (as well as for the isotopic ratios) and relation to other stellar properties before advocating any explanations.
Last but not least, the morphology of the global O-Na anticorrelation (Fig. 1) bears crucial clues on the mixing process. First, its extension to very low oxygen abundances (\[O/Fe\] $`0.45`$) is entirely due to the contribution of M13 red giants. It is this “second parameter” globular cluster that has the fastest rotating blue horizontal branch (HB) stars. Peterson et al. (1995) found six stars having $`v\mathrm{sin}i30\mathrm{km}\mathrm{s}^1`$. In the same paper red HB stars in M3 (the cluster forming a classical “second parameter”-pair with M13) have been found to possess smaller projected rotational velocities, from 2 to 20 $`\mathrm{km}\mathrm{s}^1`$. This indicates a relation between rotation and mixing.
Concerning this extension along the horizontal axis, our experience in modelling the global anticorrelation shows that it depends primarily on the mixing rate $`D_{\mathrm{mix}}`$ or, more precisely, on the product “mixing rate $`\times `$ mixing time”. We will come back to these arguments in Sect. 4, where our nucleosynthesis predictions in deeply mixed stars will be compared to the observations of \[O/Fe\] versus \[Na/Fe\].
Secondly, the extension of the global anticorrelation along the vertical axis ($`0.4`$ \[Na/Fe\] $`0.6`$ in all clusters except $`\omega `$ Cen) tells us mostly about the depth of the additional mixing. In Denissenkov & Weiss (1996) (Denissenkov & Weiss (1996), Fig. 4) and in Denissenkov et al. (1998) (Denissenkov et al. (1998), Fig. 1) it has been shown that on approaching the hydrogen burning shell the Na abundance displays two successive rises (see Fig. 2). The first rise results from the reaction <sup>22</sup>Ne(p,$`\gamma `$)<sup>23</sup>Na, whereas the deeper one is produced in the NeNa-cycle by the partial consumption of <sup>20</sup>Ne, which is much more abundant than both <sup>22</sup>Ne and <sup>23</sup>Na. The size of the vertical extension of the global anticorrelation implies that additional mixing, whatever it is, does not penetrate the second step (rise) in the Na abundance profile where H starts to decrease. Otherwise, observed \[Na/Fe\] values would be much larger than they are<sup>1</sup><sup>1</sup>1Actually some metal-rich giants in $`\omega `$ Cen show extremely high \[Na/Fe\] (up to 1 dex), probably indicating a penetration of the mixing to the second rise of Na. This cluster is the only one where some giants exhibit surface enrichment of Na produced from both <sup>20</sup>Ne and <sup>22</sup>Ne. It also has one of the bluest horizontal branches (Whitney et al. (1994)).. Cavallo et al. (1998) have investigated the dependence of the abundance profiles on metallicity (and mass). Their results confirm our arguments completely. Only for stars of near-solar metallicity very deep mixing with significant helium enrichment but without strong sodium enhancement could be possible (see also Fig. 5 of Denissenkov & Weiss (1996)). However, the clusters under discussion (e.g. M13) are metal-poor. Let us note in Fig. 1 the different behavior of field stars (Shetrone (1996a), Shetrone (1996b)). In this population, the O-Na-anticorrelation is not present (Gratton et al. (2000)). This indicates that the deep mixing does not penetrate the region where ON-burning occurs, and may reveal possible environmental effects on its efficiency.
To prepare Fig. 2 we applied a nucleosynthesis code to a red giant model with surface luminosity $`\mathrm{log}L/L_{}=2.21`$ from which our nucleosynthesis with additional deep mixing calculations started (Sect. 3). The considerable growth of the <sup>27</sup>Al abundance with depth is due to our “non-standard” assumption of an enhanced initial abundance of the $`{}_{}{}^{25}\mathrm{Mg}`$ isotope ($`[^{25}\mathrm{Mg}/\mathrm{Fe}]=1.1`$) and of the thousandfold enhanced rate of the reaction $`{}_{}{}^{26}\mathrm{Al}_{}^{\mathrm{g}}(p,\gamma )^{27}\mathrm{Si}`$ (for details see Denissenkov et al. (1998)). These ad hoc modifications were needed to explain the observed Al enhancements (Denissenkov et al. (1998)), but have no influence on the results of the present paper. From Fig. 2, we see immediately that if extra mixing penetrates down to layers, say, at $`\delta m0.06`$, this will result in an enrichment of the red giant’s envelope in N, Na and Al and in its impoverishment in C, O, and $`{}_{}{}^{25}\mathrm{Mg}`$. Again, $`{}_{}{}^{24}\mathrm{Mg}`$ remains unchanged due to the relatively low temperatures reached in such a star. These results were recently confirmed by Palacios et al. (1999) in models using the reaction rates recommended by NACRE (Angulo et al. (1999)).
## 3 Red giant evolution with deep mixing
To produce background models for the nucleosynthesis post-processing we have evolved stellar models under the assumption of additional deep mixing after the RGB bump. All sequences were started at the same initial model, which consisted of an $`0.8M_{}`$ star of initial composition $`Y=0.25`$ and $`Z=0.0003`$, and which had been evolved (canonically) up to the luminosity of the bump, i.e., $`\mathrm{log}L/L_{}=2.21`$ (at this point, its mass is $`0.798M_{}`$). The envelope helium content has increased to 0.256 (in mass fraction) due to the first dredge-up. The input physics of the Garching stellar evolution code (used here) is up-to-date (for a summary, see Denissenkov et al. (1998)), but atomic diffusion has not been included in the computations.
The additional deep mixing between the convective envelope and some point inside the hydrogen shell has been implemented in the same general line as in our previous papers on this subject, that is, as a diffusive process with parameterized values for the diffusive constant $`D_{\mathrm{mix}}`$, which is the same for all elements, and for the penetration depth. The values used for $`D_{\mathrm{mix}}`$ are guided by the results of our earlier papers, and agree with estimates based on rotationally induced mixing theories. We refer the reader to Denissenkov & Weiss (1996) for details of this approach. We use the normalized mass coordinate $`\delta m`$ introduced therein, which is 0 at the bottom of the hydrogen shell (usually, where $`X=10^4`$) and 1 at the bottom of the convective envelope. The choice of this mass coordinate allows accurate interpolation between a small number of background models in the nucleosynthesis calculations (see Cavallo et al. (1998) for a similar approach). The shell, in this coordinate, is located below $`\delta m0.10`$. The depth, down to which the diffusive mixing should occur, we denote as $`\delta m_{\mathrm{mix}}`$. Obviously, due to the lack of solid theories, one could also choose, for example, purely geometrical scales to define the penetration depth (Boothroyd & Sackmann (1999b)). Contrary to our earlier papers, the criterion for penetration is not determined by a fixed value for $`\delta m_{\mathrm{mix}}`$ chosen before the calculations, but is related to the decrease in hydrogen content within the shell (relative to the surface or convective envelope abundance $`X_{\mathrm{env}}`$), expressed as a free parameter $`\mathrm{}X`$. We have investigated several different prescriptions for the penetration criteria and found a great sensitivity of the mixing on these prescriptions, which are
1. find that $`\delta m_{\mathrm{mix}}`$ in the initial model, where $`X=X_{\mathrm{env}}\mathrm{}X`$, and mix to the same $`\delta m_{\mathrm{mix}}`$ in all subsequent models;
2. as 1., but $`\delta m=0`$ is defined as the point where $`X=X_{\mathrm{env}}/2`$ (instead of $`X=10^4`$);
3. as 1., but the diffusion constant $`D_{\mathrm{mix}}`$ is decreasing exponentially from the maximum value $`D_0`$ for $`\delta m>0.10`$ to $`D_{\mathrm{mix}}510^5D_0`$ at $`\delta m_{\mathrm{mix}}`$
4. always mix to the point, where $`X=X_{\mathrm{env}}\mathrm{}X`$
Except for method 3, these schemes were chosen to be as simple as possible and to be similar to the one by Sweigart (1997a).
Keeping the relative mass coordinate fixed, up to which mixing should occur, implies that changes in the hydrogen profile in the shell (usually steepening in the course of evolution) or in the extend of the convective envelope influence the mixing. As an illustration of the “movement” of the profile in this coordinate we display in Fig. 3 an example ($`D_{\mathrm{mix}}=10^9\mathrm{cm}^2\mathrm{s}^1`$; $`\mathrm{}X=0.20`$; method 1). The smooth solid line is the initial model. The next model (at $`\mathrm{log}L/L_{}=2.288`$) is the left-most line. From there, the model profiles shift to the right again. The solid line with clearly reduced hydrogen abundance throughout the envelope corresponds to a model close to the RGB-tip ($`\mathrm{log}L/L_{}=3.539`$). In this phase the burning time at the bottom of the mixed region becomes short enough to lead to “bottom-burning” of the envelope. An extended RGB-phase with luminosities drastically increased above the canonical RGB-tip luminosity of $`\mathrm{log}L_{\mathrm{tip}}/L_{}=3.33`$ is the result; the final value being $`\mathrm{log}L_{\mathrm{tip}}/L_{}=3.8`$. Since mass loss (Reimers (1975), with $`\eta =0.3`$) was taken into account, the total mass at the tip decreases in such cases below $`0.6M_{}`$. $`\delta m_{\mathrm{mix}}`$ is indicated by the vertical line and is $`0.047`$.
Using method 2 instead, $`\delta m=0`$ is defined as the point where the abundance of hydrogen has dropped to half the surface value of the same model ($`\delta m_{\mathrm{mix}}=0.0066`$ in this case). This prevents, obviously, any penetration to shell regions with lower hydrogen content. As in the previous case, the bottom of the envelope is burnt at the end of the RGB evolution. However, as soon as the hydrogen abundance approaches $`X_{\mathrm{env}}/2`$ (guaranteed as long as mixing is not quasi-instantaneous), the mixing criterion inhibits further mixing. For this reason $`X`$ remains constant for $`\delta m<0`$; in the final models $`X_{\mathrm{env}}=0.6402`$ and the hydrogen profile always has a finite step in the shell. In this case, the luminosity rises to $`\mathrm{log}L_{\mathrm{tip}}/L_{}=3.53`$.
The steps visible in the chemical profiles of Fig. 3 are due to the diffusion criterion applied to the numerical grid, because no interpolation to the exact value of $`\delta m_{\mathrm{mix}}`$ had been done. We verified that the results do not depend on the grid resolution, which was increased by a factor of 10 in the shell in part of the calculations. Only the steps got smaller and more numerous. To avoid such steps, we introduced a varying diffusion constant (method 3) motivated by recent results of Denissenkov & Tout (2000). They have proposed a physical mechanism for extra mixing in red giants which quantitatively interprets all the known star-to-star abundance variations in globular clusters. This is Zahn’s mechanism (Zahn (1992); Maeder & Zahn (1998)) which considers extra mixing in a radiative zone of a rotating star as a result of the joint operation of meridional circulation and turbulent diffusion. This process was already advocated by Charbonnel (1995) to explain the low carbon isotopic ratios and lithium abundances in field Population II giants and to lower the <sup>3</sup>He yields by low mass stars. Denissenkov & Tout report that the mixing rate does not vanish abruptly at a particular depth but instead it dies out gradually on a rather short depth range approximately between $`\delta m=0.10`$ and $`\delta m=0.060.07`$. This explains the following choice of an exponential decline approach for $`D_{\mathrm{mix}}`$:
$`D_{\mathrm{mix}}`$ $`=`$ $`D_0;\delta m>\delta m_0`$ (1)
$`=`$ $`D_0\mathrm{exp}\left[c_D\left({\displaystyle \frac{\delta m_0\delta m}{\delta m_{\mathrm{mix}}\delta m_0}}\right)\right];\mathrm{\hspace{0.25em}0}\delta m\delta m_0`$
where $`\delta m_0=0.10`$ was used for the beginning of the decline and $`\delta m_{\mathrm{mix}}`$ is the mixing depth coordinate as defined in method 1. Using $`c_D=10`$ ensures that $`D_{\mathrm{mix}}(\delta m_{\mathrm{mix}})510^5D_0`$. The resulting evolution (Fig. 4) is similar to that of the case shown in Fig. 3, but the profiles are smooth; mixing parameters are identical.
We finally note that method 4, applied straightforwardly, leads to a complete burning of the entire envelope for $`D_{\mathrm{mix}}510^8`$ and/or $`\mathrm{}X0.10`$. However, this we consider to be an artefact, which is easy to understand: Since due to the mixing the hydrogen abundance in the envelope is reduced, the critical point down to which mixing should occur, is moving inwards. Therefore layers of even lower hydrogen content are mixed with the envelope, and the critical point moves to even deeper regions. Only at the point where the burning time-scale is shorter than the mixing time-scale and therefore the surface hydrogen content no longer is able to adjust to the burning, this process is stopped. A way out of this situation is to ensure that the mixing does not lead to a sharp step in the shell’s hydrogen profile. This way, the critical point can be kept within the mixed layers. While the mixing procedure described in Sweigart (1997a) appears to follow the straightforward approach, Sweigart (private communication, 1999) in fact used a more complicated method to keep the hydrogen profile even in the presence of mixing. Our method 3 qualitatively has the same effect.
To summarize, method 4 appears to be unphysical; method 3 is based on the physical picture by Denissenkov & Tout (2000), but makes additional parameters necessary. Methods 1 and 2 are the most straightforward choices leading to composition profiles similar to method 3; method 2 differs from 1 in that it avoids complete mixing and burning of the envelope during the very last phases of RGB evolution.
Obviously the details of the mixing procedure influence the resulting evolution to quite a significant extent. Since at present we are far from providing a solid physical approach (which could allow, for example, for a diffusion speed varying both in space and time), we cannot predict the true evolution of a star experiencing deep mixing. However, the calculations are needed only to provide background models with varying degrees of helium mixing into the envelope, and it is of no importance how this is achieved in detail. We have calculated 25 different sequences, varying method and parameters. The cases selected for Tab. 1 are representative for the range of results we obtained, which are summarized in Figs. 5, 6, and 7. Case D of Tab. 1 is the one also shown in Fig. 4. We add that the amount of mass loss has no influence on the mixing properties. Neither does a gradual switching-on of the additional mixing during the first few models. We have performed some comparison calculations with a completely different code (the Toulouse-Geneva code; Charbonnel et al. (1992)). While the results differ in details, the gross properties are the same. The differences we ascribe to details in the mixing procedures and the implementation of diffusion.
The effects on the evolution, displayed in Figs. 57, are qualitatively as expected from the work by Sweigart (1997a). The increase in the surface helium content is quite dramatic in cases C and D (fast and very deep mixing). However, it is not as large as in Sweigart (1997a), shown, for example, in his Fig. 2, where for $`\mathrm{}X=0.20`$ a value of $`Y0.42`$ was reached. Also, in contrast to Sweigart’s result, in all our calculations the helium enrichment of the outer envelope tends to level off with progressing evolution. This might be ascribed again to differences in the mixing scheme details, as we find these differences also in the post-processing models presented in the next section (cf. Figs. 10 and 12). We also find that the luminosities can get extremely high (cases A and D) with high mass loss as the consequence and a beginning turn-away from the RGB before the He-flash sets in. The beginning of such an evolution might be recognised, too, in Fig. 3 of Sweigart (1997b) in the case of deepest mixing. We also note that mixing method 2 (cases B and C) results in loops in the HRD, which depend on the occurrence of mixing episodes. This is, for example, visible in the non-monotonic luminosity evolution in Fig. 6.
The lifetime on the RGB is in all cases prolonged (Fig. 7) by 1 or 2 Myr, in some cases up to 4 Myr (5–10% of the lifetime after the bump). The evolutionary speed is influenced mainly immediately after the onset of the additional mixing (after the bump). From the initial model to one at $`\mathrm{log}L/L_{}2.5`$, the time increases from 7 Myr (case S) to about 10 Myr (A & B) and 13 Myr (C & D). Thereafter, it is becoming comparable again in all cases. Such an effect could possibly be seen in luminosity functions, but it should be most prominent only in a limited luminosity range. VandenBerg et al. (1998) have argued that the luminosity function of M30 could be evidence for rapidly rotating cores of giants.
In some calculations, in particular those employing the exponentially declining diffusive speed (method 3), an interesting effect appeared. Although the mixing for moderate penetration depth and mixing speed remains small, the luminosity of the models rises to extreme values. As an example, in case A ($`\mathrm{}X=0.05`$, $`D_{\mathrm{mix}}=510^8\mathrm{cm}^2\mathrm{s}^1`$), $`Y_{\mathrm{env}}=0.270`$ (an enrichment of only $`+0.014`$) and $`\mathrm{log}L/L_{}=3.79`$ were reached at the tip of the RGB (see also Fig. 5). An inspection of the models shows that part of the extended region of almost homogeneous composition, which is achieved due to the effect of the additional diffusion, becomes hot enough for significant hydrogen burning. In Fig. 8 we display the H-profiles of selected models in this phase. The first one (top line) is at $`\mathrm{log}L/L_{}=2.73`$ ($`Y_{\mathrm{env}}=0.265`$). In this model, the point were the energy production due to hydrogen burning exceeds $`10^3\mathrm{erg}\mathrm{g}^1\mathrm{sec}^1`$ for the first time, is very close to the composition step. In the next model ($`\mathrm{log}L/L_{}=3.12`$; $`Y_{\mathrm{env}}=0.269`$) this point has shifted to $`\delta m=0.274`$ and is constantly progressing outward until it reaches $`\delta m=0.80`$ in the most advanced models at the bottom of the figure ($`\mathrm{log}L/L_{}=3.68`$; $`Y_{\mathrm{env}}=0.270`$). This burning of the plateau constitutes a broadening of the hydrogen shell and delivers extra luminosities. In fact, 50% of the total luminosity of the models with the plateau value around $`X=0.40`$ are generated at $`\delta m>0.06`$, that is outside the inner composition step. Therefore, the luminosity in excess of that of an ordinary star at the tip of the RGB (around $`\mathrm{log}L/L_{}=3.30`$) can completely be ascribed to the plateau burning (note that diffusion cannot keep the burning plateau homogeneous with the outer regions). The effect we observe here is probably due to our mixing description, which sets $`D_{\mathrm{mix}}`$ to the maximum value outside $`\delta m=0.10`$. If our mixing description is realistic, this would imply that one could get very high luminosities at the RGB-tip without extreme helium mixing. The time spent at luminosities above the standard TRGB brightness is only $`10^6`$ yrs and therefore observation of such a superluminous star is rather unlikely. Due to the extreme overluminosities, the Reimers mass loss formula leads to stellar winds of order $`10^7M_{}`$/yr and a final mass of $`0.58M_{}`$ ($`M_c=0.517M_{}`$). After the helium flash, this star will populate the blue part of the horizontal branch.
## 4 Nucleosynthesis in deeply mixed stars
The detailed nucleosynthesis calculations we present now have been performed in a post-processing way as in our previous works (Denissenkov & Weiss (1996); Denissenkov et al. (1998)). From an evolutionary sequence three red giant models were selected. The starting one was the same one as for the full evolutionary calculations discussed in the previous section, i.e. a model at the bump ($`\mathrm{log}L/L_{}2.2`$) in which the hydrogen burning shell had recently crossed the H-He discontinuity left by the base of the convective envelope on the first dredge-up phase. The final one was a model near the RGB tip ($`\mathrm{log}L/L_{}3.3`$), the second one having a luminosity intermediate to those of the starting and finishing models. Distributions of $`T`$, $`\rho `$ and $`r`$ with $`\delta m`$ in these three “background” models were used for interpolations in $`\mathrm{log}L`$ during the nucleosynthesis calculations. Further details about our post-processing procedure can be found in Denissenkov & Weiss (1996). The network of nuclear kinetics equations was the smallest one of those considered in Denissenkov et al. (1998). It takes into account 26 particles coupled by 30 nuclear reactions from the pp-chains, CNO-, NeNa- and MgAl-cycle. The additional mixing is modelled by diffusion with a constant coefficient $`D_{\mathrm{mix}}`$. We recall that we allow for mixing prescriptions and parameters in these calculations different from those for which the background models have been obtained.
For the comparison with observations we have preferred the “global anticorrelation” of \[O/Fe\] versus \[Na/Fe\] for the reasons detailed in Sect. 2. In Fig. 9 and Fig. 11 it is plotted for globular clusters according to the latest observational data.
In Fig. 9 theoretical dependences of \[Na/Fe\] on \[O/Fe\] obtained in the post-processing way for three values of the diffusion coefficient are shown. This first set of calculations was performed with unmixed background red giant models like in our previous papers but with mass loss taken into account. The depth of additional mixing was determined according to the penetration criterion 1 (Sect. 2) with $`\mathrm{}X=0.2`$. The mass loss rate $`\dot{M}`$ was estimated with Reimers (1975) formula in which the parameter value $`\eta =0.3`$ was adopted. In Fig. 10 the resulting envelope He abundances are shown as functions of $`\mathrm{log}L/L_{}`$. The mixing depth in the starting model which was kept constant during the nucleosynthesis calculations was $`\delta m_{\mathrm{mix}}=0.047`$. Such a value of $`\delta m_{\mathrm{mix}}`$ allows some (modest) penetration of the second Na step (see Sect. 2) by the mixing which results in an upward steepening of the theoretical dependences of \[Na/Fe\] on \[O/Fe\] by the end of the RGB evolution (Fig. 9). The maximum He enrichment achieved in the envelope in this set of calculations is $`\mathrm{}Y_{\mathrm{env}}0.10`$ (Fig. 10).
In the next nucleosynthesis calculations, the results of which are presented in Figs. 11 and 12, we have used mixed background models from sequence C’ of Tab. 1; for this sequence the same mixing parameters as in case C (Tab. 1) but a reduced mass loss rate has been used: the parameter $`\eta `$ was divided by 20 in order to take into account a reduction of $`\dot{M}`$ for low Z (Maeder (1992)). The reduced mass loss rate affects only the final mass, but not the helium enrichment. These models were evolved along the RGB with effects of the He mixing on stellar structure parameter distributions fully taken into account in the stellar evolution code (Sect. 3). Sequence C’ was chosen because out of the sample cases listed in Tab. 1 it has the highest degree of helium enrichment, in contrast to the previous, unmixed, background models. In this second set of nucleosynthesis calculations we repeat the case of the first set (Fig. 9), i.e. mixing down to $`\mathrm{}X=0.2`$ (solid lines), but also add two computer runs (dashed lines) in which the mixing was chosen to be so deep that a high enrichment in He of the envelope was guaranteed. We label these calculations by $`\delta m_{\mathrm{mix}}=0.04`$, which is the penetration depth needed to mix down to $`\mathrm{}X=0.37`$.
Comparison of the results obtained in the two sets of calculations allows to draw the following conclusions:
* mass loss is practically unimportant for this study (at least within the prescriptions and variations used in the various calculations);
* making use of mixed background models instead of unmixed ones does not seriously affect the theoretical dependences of \[Na/Fe\] on \[O/Fe\] (compare the solid curves in Figs. 9 and 11); therefore the details of the mixing prescription used for the background models are not significant for the nucleosynthesis results.
* the total He enrichment of the convective envelope calculated in the post-processing way agrees very well with the final envelope He abundance obtained in the full evolutionary calculations with additional deep mixing;
* values $`\mathrm{}Y_{\mathrm{env}}>0.15`$ were obtained only in the two computer runs with the depth $`\delta m_{\mathrm{mix}}=0.04`$, but in these cases additional mixing penetrated so deeply that it resulted in the \[Na/Fe\] on \[O/Fe\] dependences evidently inconsistent with the observations (dashed curves in Figs. 11 and 12).
A simple inspection of Figs. 9-12 allows the conclusion that the global anticorrelation of \[O/Fe\] vs. \[Na/Fe\] as a whole and especially the \[Na/Fe\] values in its low oxygen abundance tail certainly rule out any hypothesis about an increase of more than $`\mathrm{}Y_{\mathrm{env}}0.10`$ in the envelope He abundance of globular-cluster red giants. A physical reason for this constraint is the above-mentioned inability of additional mixing to penetrate the second Na abundance rise lying at $`\delta m0.06÷0.07`$ as hinted by the observed global anticorrelation. In the starting models $`\delta m0.07`$ corresponds to $`\mathrm{}X0.05`$ and, therefore, the envelope He enrichment is not expected to be much larger than $`\mathrm{}Y_{\mathrm{env}}0.05`$.
## 5 Discussion
If the O-Na-anticorrelation observed in many globular cluster red giants is indeed due to a deep mixing process beyond the standard effects taken into account in canonical stellar evolution theory, the question is justified whether this deep mixing might affect the H-He-profile as well. In this case, consequences for the red giant evolution including phases of enhanced luminosities and mass loss could result. We have, therefore, discussed both the evolutionary and nucleosynthetic effects quantitatively by performing stellar evolution calculations including deep mixing and post-processing nucleosynthesis models (the latter as we did in our earlier papers Denissenkov & Weiss (1996); Denissenkov et al. (1998)).
From arguments depending only on nucleosynthesis we could already infer that for temperature profiles typical of hydrogen-burning shells mixing of appreciable amounts of helium can only be achieved if the second Na rise is penetrated. This, however, leads to oxygen and sodium anomalies exceeding those observed (with the exception of a few stars in $`\omega `$ Cen, a multi-metallicity, untypical cluster). In terms of our normalized mass coordinate (defined such that $`\delta m=0`$ at $`X=10^4`$ at the bottom of the shell) this puts an observationally constrained limit for the maximum mixing depth of $`\delta m_{\mathrm{mix}}>0.06÷0.07`$. Our complete models confirm this argument: helium enrichment in excess of $`\mathrm{}Y_{\mathrm{env}}0.05`$ due to deep mixing can be ruled out for those stars with Na-O-anomalies as observed in clusters such as M15, M92, M3, and even M13 which presents one of the most extended blue horizontal branch.
The details of the surface abundance history along the RGB depend on the details of the deep mixing process and thus on its nature which we did not attempt to specify here. In particular, the field-to-cluster differences must be understood (especially the fact that field giants do not present the O-Na anticorrelation (Gratton et al. (2000)), indicating a deeper and more efficient mixing in their globular cluster counterparts) and seem to point out a non-negligible impact of environmental effects on the extra mixing efficiency.
However, we can compare the observations with the histories of Na and O abundance anomalies predicted by our simple mixing prescriptions in order to get constraints for a solid physical model. This is done in Fig. 13, which shows surface abundances for several clusters as a function of brightness, i.e. progressing evolution. Also shown is the theoretical prediction of the five calculations displayed in Fig. 11. Although the observational data are very few (the uncertainty in the abundances is of order 0.2 dex), and information for stars before or at the bump is available only for M13, some effects can be recognized, nevertheless.
All clusters show the whole spread of Na abundances between the canonical case without extra mixing (they would lie on a horizontal line) and that as obtained from the calculations with less deep mixing (solid lines). As already mentioned, very low O abundances are only found in M13 giants, and the observed spread for this element is a signature of the mixing rate. For both elements the abundance anomalies are limited to values predicted by models with extra mixing not penetrating the second <sup>23</sup>Na rise in the hydrogen shell. Stars with intermediate anomalies we interpret as being due to mixing penetrating less deeply into the hydrogen shell. They could be reproduced with properly adjusted mixing parameters.
Over the small brightness interval for which we have data (for all the clusters except M13 only the brightest giants are accessible), no significant abundance evolution is recognizable. The increase in $`[\mathrm{Na}/\mathrm{Fe}]`$ at the RGB tip obtained in all calculations labeled $`\mathrm{}X=0.20`$ is not visible in the observations, which may be taken as indication that the physical reasons (e.g. rotation) for the additional mixing have lost their importance or even vanished. Such a possible time dependence of the extra mixing has not been taken into account in our simple mixing prescriptions, but should result from more physically motivated models. (We recall that in Denissenkov & Weiss (1996) the carbon evolution could be reproduced with constant mixing parameters, however.) We will therefore, in a forthcoming paper, use the model by Denissenkov & Tout (2000), which includes temporal changes in the diffusion constant due to angular momentum transport.
In our models, the largest abundance changes take place at the onset of the additional mixing, i.e., early on the RGB, after the bump (around $`M_V0.5`$). This might be visible in the M13 data. The low brightness group has normal abundances, but above $`M_V0.4`$ strong anomalies already appear, reminiscent of the steep increase in the calculations mixing deeper (dashed lines). In the case of Na no further enhancement is visible (the range of values does not varies along the RGB), while O seems to get depleted further. In spite of the incomplete data at hand, the abundance anomalies seem to develop rather early and within a narrow brightness range, but then do not increase any more. This general behaviour is consistent with our theoretical predictions – though not exactly reproduced – and again points to not too deep mixing at moderate speed. The calculations with extreme helium mixing would predict the largest O depletion and Na enhancement all along the RGB; the absence of such stars can therefore not be explained with a selection effect working against the shortest lived stars at the tip of the RGB. The fact that M13 seems to show anomalies already before the bump has to be taken with care, because we compare here only with one stellar model which has a metallicity almost a factor of 10 smaller than that appropriate for M13. At that metallicity, the bump would occur about 1 mag earlier (and our initial model is about 0.2 mag brighter than the end of the bump). We stress the fact that the present observational data do not allow a detailed comparison with the abundance evolution, but only rough qualitative statements.
Although Fig. 13 again demonstrates how the abundance evolution depends on the assumption about the additional mixing process, our calculations also show that the nucleosynthesis argument given above is valid for all mixing descriptions tested and therefore model-independent. From this we predict that red giants exhibiting the observed Na-O-anomalies do not have envelopes enriched in helium by much more than $`\mathrm{}Y_{\mathrm{env}}0.05`$, which is comparable to the general uncertainty in our knowledge about the helium abundances in such stars.
Concerning the brightness reached during the RGB evolution, we showed that for specific mixing prescriptions large excess luminosities can be achieved at the end of the evolution without simultaneous mixing of large amounts of helium. As the reason we could identify the burning of the outer parts of the hydrogen shell where some H-He-mixing had happened in earlier phases. These regions become hot enough for significant hydrogen burning on such short time-scales that the deep mixing process is not able to mix the products any longer into the convective envelope. Such models experience very strong mass loss under the assumption of the continuing validity of a Reimers-type stellar wind and finish the RGB phase with total masses below $`0.6M_{}`$ and envelope masses of only 10% of this. They could be candidates for blue HB stars and would link observed abundance anomalies, deep mixing and the second parameter problem, as suggested by Langer & Hoffman (1995) and Sweigart (1997a). They would avoid the problem of overproducing O-Na-anomalies as would result from the helium mixing investigated by Sweigart (1997a).
However, we repeat our warning that the details of the evolutionary consequences of deep mixing depend crucially on the mixing process and history. Mixing speed and depth are both important for the amount of helium mixed and for the result of the competing processes “mixing” and “burning” and their time-scales. The mixing depth is also linked to the criterion for deep mixing, for which we have investigated several simple recipes.
To conclude, we need a solid physical picture for the deep mixing process in order to be able to investigate its effect on red giant evolution further. Presently, we can only point out some interesting possibilities – such as the overluminosities – and derive model-independent features, such as our main conclusion that the observed anomalies of oxygen and sodium rule out strong helium enhancement and therefore very deep mixing.
###### Acknowledgements.
We are grateful to A. Sweigart for helpful discussions. This study was partly done while CC and PAD visited the Max-Planck-Institut für Astrophysik in Garching. They express their gratitude to the staff for hospitality and support. We appreciate the very careful work of an anonymous referee, whose detailed and constructive comments helped to improve this paper.
|
warning/0002/hep-ph0002128.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The presence of an additional $`Z^{}`$ boson is predicted in a certain class of grand unified theories (GUT) with a gauge group whose rank is higher than that of the Standard Model (SM). The supersymmetric (SUSY) $`E_6`$ models are the promising candidates which predict the additional $`Z^{}`$-boson at the weak scale <sup>?</sup>. Because $`E_6`$ is a rank-six group, it can have two extra $`\mathrm{U}(1)`$ factors besides the SM gauge group. A superposition of the two extra $`\mathrm{U}(1)`$ groups may survive as the $`\mathrm{U}(1)^{}`$ gauge symmetry at the GUT scale. The $`\mathrm{U}(1)^{}`$ symmetry may break spontaneously at the weak scale through the radiative corrections to the mass term of the SM singlet scalar field <sup>?</sup>.
In general, the additional $`\mathrm{U}(1)^{}`$ gauge boson $`Z^{}`$ can mix with the hypercharge $`\mathrm{U}(1)_Y`$ gauge boson through the kinetic term at above the electroweak scale, and also it can mix with the SM $`Z`$ boson after the electroweak symmetry is spontaneously broken. Through those mixings, the $`Z^{}`$ boson can affect the electroweak observables at the $`Z`$-pole and the $`W`$-boson mass $`m_W`$. Both the $`Z`$-$`Z^{}`$ mixing and the direct $`Z^{}`$ contribution can affect neutral current experiments off the $`Z`$ pole. The presence of an additional $`Z^{}`$ boson can be explored directly at $`p\overline{p}`$ collider experiments.
In this review article, we report constraints on $`Z^{}`$ bosons in the SUSY $`E_6`$ models from electroweak experiments based on the formalism in Refs. ReferencesReferences. Constraints on the $`Z^{}`$ bosons from electroweak experiments have been studied by several authors <sup>?,?,?,?,?</sup>. Especially, a special attention has been paid to this subject <sup>?</sup> after the new analysis of parity violation in cesium atom has led to the improved data of the weak charge $`Q_W(_{55}^{133}Cs)`$ <sup>?</sup>, which is 2.2-$`\sigma `$ away from the SM prediction (see Table 3). The analysis given in this article updates their studies by allowing for an arbitrary kinetic mixing <sup>?,?,?</sup> between the $`Z^{}`$ boson and the hypercharge $`B`$ boson. The constraints on the $`Z^{}`$ bosons can be found by using the results of $`Z`$-pole experiments at LEP1 and SLC, and the $`m_W`$ measurements at Tevatron and LEP2. Also the low-energy neutral current (LENC) experiments – lepton-quark, lepton-lepton scattering experiments and atomic parity violation (APV) measurements – constrain the direct exchange of $`Z^{}`$ boson.
It has been found <sup>?,?</sup> that the lower mass limit of the heavier mass eigenstate $`Z_2`$ is obtained as a function of the effective $`Z`$-$`Z^{}`$ mixing term $`\zeta `$, which is a combination of the mass and kinetic mixings. In principle, $`\zeta `$ is calculable, together with the gauge coupling $`g_E`$, once the particle spectrum of the $`E_6`$ model is specified. We show the theoretical prediction for $`\zeta `$ and $`g_E`$ in the SUSY $`E_6`$ models by assuming the minimal particle content which satisfies the anomaly free condition and the gauge coupling unification$`^\text{a}`$footnotetext: $`^\text{a}`$Consequence in the case of the maximal particle content which preserve the perturbative unification of the gauge couplings has been studied in Ref. References..
This paper is organized as follows. In the next section, we review the additional $`Z^{}`$ boson in the SUSY $`E_6`$ models and the generic feature of $`Z`$-$`Z^{}`$ mixing in order to fix our notation. We show that the effects of $`Z`$-$`Z^{}`$ mixing and direct $`Z^{}`$ boson contribution are parametrized by the following three terms: (i) a tree-level contribution to the $`T`$ parameter <sup>?</sup>, $`T_{\mathrm{new}}`$, (ii) the effective $`Z`$-$`Z^{}`$ mass mixing angle $`\overline{\xi }`$ and (iii) a contact term $`g_E^2/c_\chi ^2m_{Z_2}^2`$ which appears in the low-energy processes. In Sec. 3, we collect the data of electroweak experiments which will be used in our analysis. We also present the theoretical framework to calculate the electroweak observables. In Sec. 4, we show constraints on the $`Z^{}`$ bosons from the electroweak data. The presence of non-zero kinetic mixing between the $`\mathrm{U}(1)_Y`$ and $`\mathrm{U}(1)^{}`$ gauge bosons modifies the couplings between the $`Z^{}`$ boson and the SM fermions. We discuss impacts of the kinetic mixing term on the $`\chi ^2`$-analysis. The 95% CL lower mass limit of the heavier mass eigenstate $`Z_2`$ in four representative models – $`\chi ,\psi ,\eta ,\nu `$ models – is given as a function of the effective $`Z`$-$`Z^{}`$ mixing parameter $`\zeta `$. The $`\zeta `$-independent constraints from the low-energy experiments and those from the direct search experiments at Tevatron are also discussed. In Sec. 5, we find the theoretical prediction for $`\zeta `$ in $`\chi ,\psi ,\eta ,\nu `$ models by assuming the minimal particle contents. Stringent $`Z_2`$ boson mass bounds are found for most models. Sec. 6 is devoted to summarize this paper.
## 2 $`Z`$-$`Z^{}`$ mixing in supersymmetric $`E_6`$ model
### 2.1 $`Z^{}`$ boson in supersymmetric $`E_6`$ model
Since the rank of $`E_6`$ is six, it has two $`\mathrm{U}(1)`$ factors besides the SM gauge group which arise from the following decompositions:
$`\begin{array}{cc}\hfill E_6& \mathrm{SO}(10)\times \mathrm{U}(1)_\psi \hfill \\ & \mathrm{SU}(5)\times \mathrm{U}(1)_\chi \times \mathrm{U}(1)_\psi .\hfill \end{array}`$ (3)
An additional $`Z^{}`$ boson in the electroweak scale can be parametrized as a linear combination of the $`\mathrm{U}(1)_\psi `$ gauge boson $`Z_\psi `$ and the $`\mathrm{U}(1)_\chi `$ gauge boson $`Z_\chi `$ as <sup>?</sup>
$$Z^{}=Z_\chi \mathrm{cos}\beta _E+Z_\psi \mathrm{sin}\beta _E.$$
(4)
In this paper, the following $`Z^{}`$ models are studied in some detail:
$`\begin{array}{ccccc}& & & & \\ \beta _E& 0& \pi /2& \mathrm{tan}^1(\sqrt{5/3})& \mathrm{tan}^1(\sqrt{15})\\ & & & & \\ \mathrm{model}& \chi & \psi & \eta & \nu \end{array}`$ (7)
In the SUSY-$`E_6`$ models, each generation of the SM quarks and leptons is embedded into a 27 representation. In Table 1, we show all the matter fields contained in a 27 and their classification in SO(10) and SU(5). The $`\mathrm{U}(1)^{}`$ charge assignment on the matter fields for each model is also given in the same table. The normalization of the $`\mathrm{U}(1)^{}`$ charge follows that of the hypercharge.
Besides the SM quarks and leptons, there are two SM singlets $`\nu ^c`$ and $`S`$, a pair of weak doublets $`H_u`$ and $`H_d`$, a pair of color triplets $`D`$ and $`\overline{D}`$ in each generation. The $`\eta `$ model arises when $`E_6`$ breaks into a rank-5 group directly in a specific compactification of the heterotic string theory <sup>?</sup>. In the $`\nu `$ model, the right-handed neutrinos $`\nu ^c`$ are gauge singlet <sup>?</sup> and can have large Majorana masses to realize the see-saw mechanism <sup>?</sup>.
The $`\mathrm{U}(1)^{}`$ symmetry breaking occurs if the scalar component of the SM singlet field develops the vacuum expectation value (VEV). It can be achieved at near the weak scale via radiative corrections to the mass term of the SM singlet scalar field. Recent studies of the radiative $`\mathrm{U}(1)^{}`$ symmetry breaking can be found, e.g., in Ref. References.
Several problems may arise in the $`E_6`$ models from view of low-energy phenomenology. For example, the presence of the baryon number violating operators give rise to too fast proton decay, or the absence of the Majorana neutrino mass terms (except for the $`\nu `$ model) requires a fine-tuning of the Dirac neutrino mass in order to satisfy experimentally observed neutrino mass relations. Some approaches to these problems are summarized in Ref. References. In the following, we assume that these problems are solved by an unknown mechanism. Moreover we assume that all the super-partners of the SM particles and the exotic matters do not affect the radiative corrections to the electroweak observables significantly, i.e., they are assumed to be heavy enough to decouple from the weak boson mass scale.
### 2.2 Phenomenological consequences of $`Z`$-$`Z^{}`$ mixing
If the SM Higgs field carries a non-trivial $`\mathrm{U}(1)^{}`$ charge, its VEV induces the $`Z`$-$`Z^{}`$ mass mixing. On the other hand, the kinetic mixing between the hypercharge gauge boson $`B`$ and the $`\mathrm{U}(1)^{}`$ gauge boson $`Z^{}`$ can occur through the quantum effects below the GUT scale. After the electroweak symmetry is broken, the effective Lagrangian for the neutral gauge bosons in the $`\mathrm{SU}(2)_L\times \mathrm{U}(1)_Y\times \mathrm{U}(1)^{}`$ theory is given by <sup>?</sup>
$`_{gauge}`$ $`=`$ $`{\displaystyle \frac{1}{4}}Z^{\mu \nu }Z_{\mu \nu }{\displaystyle \frac{1}{4}}Z^{\mu \nu }Z_{\mu \nu }^{}{\displaystyle \frac{\mathrm{sin}\chi }{2}}B^{\mu \nu }Z_{\mu \nu }^{}{\displaystyle \frac{1}{4}}A^{0\mu \nu }A_{\mu \nu }^0`$ (8)
$`+m_{ZZ^{}}^2Z^\mu Z_\mu ^{}+{\displaystyle \frac{1}{2}}m_Z^2Z^\mu Z_\mu +{\displaystyle \frac{1}{2}}m_Z^{}^2Z^\mu Z_\mu ^{},`$
where $`F^{\mu \nu }(F=Z,Z^{},A^0,B)`$ represents the gauge field strength. The $`Z`$-$`Z^{}`$ mass mixing and the kinetic mixing are characterized by $`m_{ZZ^{}}^2`$ and $`\mathrm{sin}\chi `$, respectively. In this basis, the interaction Lagrangian for the neutral current process is given as
$`_{NC}`$ $`=`$ $`{\displaystyle \underset{f,\alpha }{}}\{eQ_{f_\alpha }\overline{f_\alpha }\gamma ^\mu f_\alpha A_\mu ^0+g_Z\overline{f_\alpha }\gamma ^\mu (I_{f_L}^3Q_{f_\alpha }\mathrm{sin}^2\theta _W)f_\alpha Z_\mu `$ (9)
$`+g_EQ_E^{f_\alpha }\overline{f_\alpha }\gamma ^\mu f_\alpha Z_\mu ^{}\},`$
where $`g_Z=g/\mathrm{cos}\theta _W=g_Y/\mathrm{sin}\theta _W`$. The $`\mathrm{U}(1)^{}`$ gauge coupling constant is denoted by $`g_E`$ in the hypercharge normalization. The symbol $`f_\alpha `$ denotes the quarks or leptons with the chirality $`\alpha `$ ($`\alpha =L`$ or $`R`$). The third component of the weak isospin, the electric charge and the $`\mathrm{U}(1)^{}`$ charge of $`f_\alpha `$ are given by $`I_{f_\alpha }^3`$, $`Q_{f_\alpha }`$ and $`Q_E^{f_\alpha }`$, respectively. The $`\mathrm{U}(1)^{}`$ charge of the quarks and leptons listed in Table 1 should be read as
$`Q_E^Q=Q_E^{u_L}=Q_E^{d_L},Q_E^L=Q_E^{\nu _L}=Q_E^{e_L},Q_E^{f^c}=Q_E^{f_R}(f=e,u,d).`$ (10)
The mass eigenstates $`(Z_1,Z_2,A)`$ is obtained by the following transformation;
$$\left(\begin{array}{c}Z\\ Z^{}\\ A^0\end{array}\right)=\left(\begin{array}{ccc}\mathrm{cos}\xi +\mathrm{sin}\xi \mathrm{sin}\theta _W\mathrm{tan}\chi & \mathrm{sin}\xi +\mathrm{cos}\xi \mathrm{sin}\theta _W\mathrm{tan}\chi & 0\\ \mathrm{sin}\xi /\mathrm{cos}\chi & \mathrm{cos}\xi /\mathrm{cos}\chi & 0\\ \mathrm{sin}\xi \mathrm{cos}\theta _W\mathrm{tan}\chi & \mathrm{cos}\xi \mathrm{cos}\theta _W\mathrm{tan}\chi & 1\end{array}\right)\left(\begin{array}{c}Z_1\\ Z_2\\ A\end{array}\right).$$
(11)
Here the mixing angle $`\xi `$ is given by
$$\mathrm{tan}2\xi =\frac{2c_\chi (m_{ZZ^{}}^2+s_Ws_\chi m_Z^2)}{m_Z^{}^2(c_\chi ^2s_W^2s_\chi ^2)m_Z^2+2s_Ws_\chi m_{ZZ^{}}^2},$$
(12)
with the short-hand notation, $`c_\chi =\mathrm{cos}\chi `$, $`s_\chi =\mathrm{sin}\chi `$ and $`s_W=\mathrm{sin}\theta _W`$. The physical masses $`m_{Z_1}`$ and $`m_{Z_2}`$ ($`m_{Z_1}<m_{Z_2}`$) are given as follows;
$`m_{Z_1}^2`$ $`=`$ $`m_Z^2(c_\xi +s_\xi s_Wt_\chi )^2+m_Z^{}^2\left({\displaystyle \frac{s_\xi }{c_\chi }}\right)^2+2m_{ZZ^{}}^2{\displaystyle \frac{s_\xi }{c_\chi }}(c_\xi +s_\xi s_Wt_\chi ),`$ (13a)
$`m_{Z_2}^2`$ $`=`$ $`m_Z^2(c_\xi s_Wt_\chi s_\xi )^2+m_Z^{}^2\left({\displaystyle \frac{c_\xi }{c_\chi }}\right)^2+2m_{ZZ^{}}^2{\displaystyle \frac{c_\xi }{c_\chi }}(c_\xi s_Wt_\chi s_\xi ),`$ (13b)
where $`c_\xi =\mathrm{cos}\xi `$, $`s_\xi =\mathrm{sin}\xi `$ and $`t_\chi =\mathrm{tan}\chi `$. The lighter mass eigenstate $`Z_1`$ should be identified with the observed $`Z`$ boson at LEP1 or SLC. The excellent agreement between the current experimental results and the SM predictions at the quantum level implies that the mixing angle $`\xi `$ has to be small. In the limit of small $`\xi `$, the interaction Lagrangians for the processes $`Z_{1,2}f_\alpha \overline{f_\alpha }`$ are expressed as
$`_{Z_1}`$ $`=`$ $`{\displaystyle \underset{f,\alpha }{}}g_Z\overline{f_\alpha }\gamma ^\mu \left[\left(I_{f_L}^3Q_{f_\alpha }\mathrm{sin}^2\theta _W\right)+\stackrel{~}{Q}_E^{f_\alpha }\overline{\xi }\right]f_\alpha Z_{1\mu },`$ (14a)
$`_{Z_2}`$ $`=`$ $`{\displaystyle \underset{f,\alpha }{}}{\displaystyle \frac{g_E}{c_\chi }}\overline{f_\alpha }\gamma ^\mu \left[\stackrel{~}{Q}_E^{f_\alpha }\left(I_{f_\alpha }^3Q_{f_\alpha }\mathrm{sin}^2\theta _W\right){\displaystyle \frac{g_Zc_\chi }{g_E}}\xi \right]f_\alpha Z_{2\mu },`$ (14b)
where the effective mixing angle $`\overline{\xi }`$ in Eq. (14a) is given as
$$\overline{\xi }=\frac{g_E}{g_Z\mathrm{cos}\chi }\xi .$$
(15)
In Eq. (2.2), the effective $`\mathrm{U}(1)^{}`$ charge $`\stackrel{~}{Q}_E^{f_\alpha }`$ is introduced as a combination of $`Q_E^{f_\alpha }`$ and the hypercharge $`Y_{f_\alpha }`$:
$`\stackrel{~}{Q}_E^{f_\alpha }`$ $``$ $`Q_E^{f_\alpha }+Y_{f_\alpha }\delta ,`$ (16a)
$`\delta `$ $``$ $`{\displaystyle \frac{g_Z}{g_E}}s_Ws_\chi ,`$ (16b)
where the hypercharge $`Y_{f_\alpha }`$ should be read from Table 1 in the same manner with $`Q_E^{f_\alpha }`$ (see, Eq. (10)). As a notable example, one can see from Table 1 that the effective charge $`\stackrel{~}{Q}_E^{f_\alpha }`$ of the leptons ($`L`$ and $`e^c`$) disappears in the $`\eta `$ model if $`\delta `$ is taken to be $`1/3`$ <sup>?</sup>.
Now, due to the $`Z`$-$`Z^{}`$ mixing, the observed $`Z`$ boson mass $`m_{Z_1}`$ at LEP1 or SLC is shifted from the SM $`Z`$ boson mass $`m_Z`$:
$$\mathrm{\Delta }m^2m_{Z_1}^2m_Z^20.$$
(17)
The presence of the mass shift affects the $`T`$-parameter <sup>?</sup> at tree level. Following the notation of Ref. References, the $`T`$-parameter is expressed in terms of the effective form factors $`\overline{g}_Z^2(0),\overline{g}_W^2(0)`$ and the fine structure constant $`\alpha `$:
$`\alpha T`$ $``$ $`1{\displaystyle \frac{\overline{g}_W^2(0)}{m_W^2}}{\displaystyle \frac{m_{Z_1}^2}{\overline{g}_Z^2(0)}}`$ (18a)
$`=`$ $`\alpha \left(T_{\mathrm{SM}}+T_{\mathrm{new}}\right),`$ (18b)
where $`T_{\mathrm{SM}}`$ and the new physics contribution $`T_{\mathrm{new}}`$ are given by:
$`\alpha T_{\mathrm{SM}}`$ $`=`$ $`1{\displaystyle \frac{\overline{g}_W^2(0)}{m_W^2}}{\displaystyle \frac{m_Z^2}{\overline{g}_Z^2(0)}},`$ (19a)
$`\alpha T_{\mathrm{new}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Delta }m^2}{m_{Z_1}^2}}0.`$ (19b)
It is worth noting that the sign of $`T_{\mathrm{new}}`$ is always positive. The effects of the $`Z`$-$`Z^{}`$ mixing in the $`Z`$-pole experiments have hence been parametrized by the effective mixing angle $`\overline{\xi }`$ and the positive parameter $`T_{\mathrm{new}}`$.
We note here that we retain the kinetic mixing term $`\delta `$ as a part of the effective $`Z_1`$ coupling $`\stackrel{~}{Q}_E^{f_\alpha }`$ in Eq. (16a). As shown in Refs. References,References, References, the kinetic mixing term $`\delta `$ can be absorbed into a further redefinition of $`S`$ and $`T`$. Such re-parametrization may be useful if the term $`Y_{f_\alpha }\delta `$ in Eq. (16a) is much larger than the $`Z^{}`$ charge $`Q_E^{f_\alpha }`$. In the $`E_6`$ models studied in this paper, we find no merit in absorbing the $`Y_f\delta `$ term because, the remaining $`Q_E^{f_\alpha }`$ term is always significant. We therefore adopt $`\stackrel{~}{Q}_E^{f_\alpha }`$ as the effective $`Z_1`$ couplings and $`T_{\mathrm{new}}`$ accounts only for the mass shift (17). All physical consequences such as the bounds on $`\overline{\xi }`$ and $`m_{Z_2}`$ are of course independent of our choice of the parametrization.
The two parameters $`T_{\mathrm{new}}`$ and $`\overline{\xi }`$ are complicated functions of the parameters of the effective Lagrangian (8). In the small mixing limit, we find the following useful expressions
$`\overline{\xi }`$ $`=`$ $`\left({\displaystyle \frac{g_E}{g_Z}}{\displaystyle \frac{m_Z}{m_Z^{}}}\right)^2\zeta \left[1+O({\displaystyle \frac{m_Z^2}{m_Z^{}^2}})\right],`$ (20a)
$`\alpha T_{\mathrm{new}}`$ $`=`$ $`\left({\displaystyle \frac{g_E}{g_Z}}{\displaystyle \frac{m_Z}{m_Z^{}}}\right)^2\zeta ^2\left[1+O({\displaystyle \frac{m_Z^2}{m_Z^{}^2}})\right],`$ (20b)
where we introduced an effective mixing parameter $`\zeta `$
$$\zeta =\frac{g_Z}{g_E}\frac{m_{ZZ^{}}^2}{m_Z^2}\delta .$$
(21)
The $`Z`$-$`Z^{}`$ mixing effect disappears at $`\zeta =0`$. Stringent limits on $`m_Z^{}`$ and hence on $`m_{Z_2}`$ can be obtained through the mixing effect if $`\zeta `$ is $`O(1)`$. We will show in Sec. 5 that $`\zeta `$ is calculable once the particle spectrum of the model is specified. The parameter $`\zeta `$ plays an essential role in the analysis of $`Z^{}`$ models.
In the low-energy neutral current processes, effects of the exchange of the heavier mass eigenstate $`Z_2`$ can be detected. In the small $`\overline{\xi }`$ limit, they constrain the contact term $`g_E^2/c_\chi ^2m_{Z_2}^2`$.
## 3 Electroweak observables in the $`Z^{}`$ model
In this section, we briefly discuss the theoretical framework <sup>?,?</sup> to calculate the electroweak observables which are used in our analysis. The experimental data of the $`Z`$-pole experiments, the $`W`$-boson mass measurement and the low-energy experiments used in this paper are summarized in Table 3.
The pseudo-observables of the $`Z`$-pole experiments are expressed in terms of the effective coupling $`g_\alpha ^f`$ <sup>?</sup>, where $`f`$ denotes all the SM fermions except for the top-quark, and $`\alpha `$ being their chirality, $`L`$ or $`R`$. Following our parametrization of the $`Z`$-$`Z^{}`$ mixing (14a), the effective coupling $`g_\alpha ^f`$ in the $`Z^{}`$ models can be expressed as
$`g_\alpha ^f=(g_\alpha ^f)_{\mathrm{SM}}+\stackrel{~}{Q}_E^{f_\alpha }\overline{\xi }.`$ (22)
The SM prediction for the effective coupling $`(g_\alpha ^f)_{\mathrm{SM}}`$ can be expanded in terms of the gauge boson propagator corrections $`\mathrm{\Delta }\overline{g}_Z^2`$ and $`\mathrm{\Delta }\overline{s}^2`$:
$`(g_\alpha ^f)_{\mathrm{SM}}=a+b\mathrm{\Delta }\overline{g}_Z^2+c\mathrm{\Delta }\overline{s}^2,`$ (23)
where the numerical coefficients $`a,b`$ and $`c`$ are given in Refs. References,References. Two parameters $`\mathrm{\Delta }\overline{g}_Z^2`$ and $`\mathrm{\Delta }\overline{s}^2`$ in Eq. (23) are defined as the shift in the effective couplings $`\overline{g}_Z^2(m_{Z_1}^2)`$ and $`\overline{s}^2(m_{Z_1}^2)`$ <sup>?</sup> from their SM reference values at $`m_t=175\mathrm{GeV}`$ and $`m_H=100\mathrm{GeV}`$. They can be expressed in terms of the $`S`$ and $`T`$ parameters <sup>?</sup> as
$`\mathrm{\Delta }\overline{g}_Z^2`$ $`=`$ $`\overline{g}_Z^2(m_{Z_1}^2)0.55635=0.00412\mathrm{\Delta }T+0.00005[1(100\mathrm{GeV}/m_H)^2],`$ (24a)
$`\mathrm{\Delta }\overline{s}^2`$ $`=`$ $`\overline{s}^2(m_{Z_1}^2)0.23035=0.00360\mathrm{\Delta }S0.00241\mathrm{\Delta }T0.00023x_\alpha ,`$ (24b)
where the expansion parameter $`x_\alpha `$ is introduced to estimate the uncertainty of the hadronic contribution to the QED coupling $`1/\overline{\alpha }(m_{Z_1}^2)=128.75\pm 0.09`$ <sup>?</sup>:
$$x_\alpha \frac{1/\overline{\alpha }(m_{Z_1}^2)128.75}{0.09}.$$
(25)
Here, $`\mathrm{\Delta }S,\mathrm{\Delta }T,\mathrm{\Delta }U`$ parameters are also measured from their SM reference values and they are given as the sum of the SM and the new physics contributions
$$\mathrm{\Delta }S=\mathrm{\Delta }S_{\mathrm{SM}}+S_{\mathrm{new}},\mathrm{\Delta }T=\mathrm{\Delta }T_{\mathrm{SM}}+T_{\mathrm{new}},\mathrm{\Delta }U=\mathrm{\Delta }U_{\mathrm{SM}}+U_{\mathrm{new}}.$$
(26)
A convenient parametrization of $`\mathrm{\Delta }S_{\mathrm{SM}},\mathrm{\Delta }T_{\mathrm{SM}}`$ and $`\mathrm{\Delta }U_{\mathrm{SM}}`$ in terms of $`m_t`$ and $`m_H`$ has been given in Ref. References. The formulae of the $`Z`$-pole observables listed in Table 3 in terms of $`g_\alpha ^f`$ can be found in Refs. References,References.
The theoretical prediction of $`m_W`$ is given as <sup>?,?</sup>
$$m_W(\mathrm{GeV})=80.4020.288\mathrm{\Delta }S+0.418\mathrm{\Delta }T+0.337\mathrm{\Delta }U+0.012x_\alpha ,$$
(27)
by using the same parameters, $`\mathrm{\Delta }S,\mathrm{\Delta }T,\mathrm{\Delta }U`$ (26) and $`x_\alpha `$ (25).
The observables in the LENC experiments which are used in our analysis are as follows – (i) polarization asymmetry of the charged lepton scattering off nucleus target, (ii) parity violation in cesium atom, (iii) inelastic $`\nu _\mu `$-scattering off nucleus target and (iv) neutrino-electron scattering. Theoretical expressions for the observables of (i) and (ii) are conveniently given in terms of the model-independent parameters $`C_{1q},C_{2q}`$ <sup>?</sup> and $`C_{3q}`$ <sup>?</sup>. The $`\nu _\mu `$-scattering data (iii) and (iv) are expressed in terms of the parameters $`g_{L\alpha }^{\nu _\mu f}`$. In the $`Z^{}`$ models, these model-independent parameters can be written as follows:
$`C_{iq}`$ $`=`$ $`(C_{iq})_{\mathrm{SM}}+\mathrm{\Delta }C_{iq},`$ (28a)
$`g_{L\alpha }^{\nu _\mu f}`$ $`=`$ $`(g_{L\alpha }^{\nu _\mu f})_{\mathrm{SM}}+\mathrm{\Delta }g_{L\alpha }^{\nu _\mu f},`$ (28b)
where the first term in each equation is the SM contribution which is parametrized conveniently by $`\mathrm{\Delta }S`$ and $`\mathrm{\Delta }T`$ <sup>?</sup>. The second terms $`\mathrm{\Delta }C_{iq}`$ and $`\mathrm{\Delta }g_{L\alpha }^{\nu _\mu f}`$ represent the additional contributions from the $`Z`$-$`Z^{}`$ mixing and the direct $`Z_2`$ exchange, which are proportional to $`\overline{\xi }`$ and $`g_E^2/c_\chi ^2m_{Z_2}^2`$, respectively. The theoretical prediction of the LENC observables in terms of $`\mathrm{\Delta }C_{iq}`$ and $`\mathrm{\Delta }g_{L\alpha }^{\nu \mu f}`$ can be found in Ref. References.
## 4 Constraints on $`Z^{}`$ bosons from electroweak experiments
### 4.1 $`\chi ^2`$-analysis on the $`Z^{}`$ models
There are six free parameters in the $`Z^{}`$ models – the tree level contribution to the $`T`$ parameter $`T_{\mathrm{new}}`$, the $`Z`$-$`Z^{}`$ mass mixing angle $`\overline{\xi }`$, the direct $`Z_2`$-boson contribution to the low-energy processes $`g_E^2/c_\chi ^2m_{Z_2}^2`$, and the three SM parameters, $`m_t,\alpha _s(m_{Z_1})`$ and $`\overline{\alpha }(m_{Z_1}^2)`$. Throughout our analysis, we use $`m_t=173.8\pm 5.2(\mathrm{GeV})`$ <sup>?</sup>, $`\alpha _s(m_{Z_1})=0.119\pm 0.002`$ <sup>?</sup>, and $`1/\overline{\alpha }(m_{Z_1}^2)=128.75\pm 0.09`$ <sup>?</sup> as constraints on the SM parameters. The Higgs mass dependence of the results are parametrized by $`x_H\mathrm{ln}(m_H/100\mathrm{GeV})`$ in the range $`90\mathrm{GeV}<m_H\mathrm{\Gamma }<\mathrm{\hspace{0.17em}150}\mathrm{GeV}`$. The lower bound is obtained at the LEP2 experiment <sup>?</sup>. The upper bound is the theoretical limit on the lightest Higgs boson mass in any supersymmetric models that accommodate perturbative unification of the gauge couplings <sup>?</sup>.
We summarize the results of the fit for the $`\psi ,\chi ,\eta `$ and $`\nu `$ models:
$`(1)\chi \mathrm{model}(\delta =0)`$
$`\begin{array}{cc}T_{\mathrm{new}}\hfill & =0.059+0.14x_H\pm 0.098\hfill \\ \overline{\xi }(10^4)\hfill & =0.029x_H\pm 4.05\hfill \\ g_E^2/c_\chi ^2m_{Z_2}^2\hfill & =0.237+0.0032x_H\pm 0.107\hfill \end{array}\},\chi _{\mathrm{min}}^2=17.8+1.2x_H,`$ (29d)
$`(2)\psi \mathrm{model}(\delta =0)`$
$`\begin{array}{cc}T_{\mathrm{new}}\hfill & =0.075+0.14x_H\pm 0.097\hfill \\ \overline{\xi }(10^4)\hfill & =1.21.3x_H\pm 4.8\hfill \\ g_E^2/c_\chi ^2m_{Z_2}^2\hfill & =1.31+0.16x_H\pm 2.97\hfill \end{array}\},\chi _{\mathrm{min}}^2=22.9+1.4x_H,`$ (29h)
$`(3)\eta \mathrm{model}(\delta =0)`$
$`\begin{array}{cc}T_{\mathrm{new}}\hfill & =0.062+0.14x_H\pm 0.097\hfill \\ \overline{\xi }(10^4)\hfill & =2.76.7x_H\pm 9.4\hfill \\ g_E^2/c_\chi ^2m_{Z_2}^2\hfill & =0.814+0.089x_H\pm 0.449\hfill \end{array}\},\chi _{\mathrm{min}}^2=18.7+0.7x_H,`$ (29l)
$`(4)\nu \mathrm{model}(\delta =0)`$
$`\begin{array}{cc}T_{\mathrm{new}}\hfill & =0.057+0.14x_H\pm 0.098\hfill \\ \overline{\xi }(10^4)\hfill & =0.42+0.6x_H\pm 3.9\hfill \\ g_E^2/c_\chi ^2m_{Z_2}^2\hfill & =0.619+0.024x_H\pm 0.275\hfill \end{array}\},\chi _{\mathrm{min}}^2=18.0+1.1x_H,`$ (29p)
where $`\mathrm{d}.\mathrm{o}.\mathrm{f}.=20`$.
The mixing angle $`\overline{\xi }`$ and the contact term $`g_E^2/c_\chi ^2m_{Z_2}^2`$ are given in units of $`10^4`$ and $`\mathrm{TeV}^2`$, respectively. The best fit value of $`T_{\mathrm{new}}`$ falls into the unphysical region ($`T_{\mathrm{new}}<0`$) for all $`Z^{}`$ models even if the Higgs boson mass is its upper limit ($`150\mathrm{GeV}`$). It should be noticed that $`T_{\mathrm{new}}`$ and $`\overline{\xi }`$ are consistent with zero in all models while $`g_E^2/c_\chi ^2m_{Z_2}^2`$ shows the deviation from zero in the 1-$`\sigma `$ level for $`\chi ,\eta ,\nu `$ models.
The best fit values of the six-parameters for $`m_H=100\mathrm{GeV}`$ under the condition $`T_{\mathrm{new}}0`$ are shown in Table 4.1, together with the SM best fit result at $`m_H=100\mathrm{GeV}`$. Only the best fit value of $`g_E^2/c_\chi ^2m_{Z_2}^2`$ in the $`\eta `$ model is found in the unphysical region ($`g_E^2/c_\chi ^2m_{Z_2}^2<0`$). The pull factors of the electroweak observables at the best fit point are also shown in Table 3. We learn from the table that almost no improvement of the fit over the SM is found for the $`Z`$-pole and $`m_W`$ measurements. However, the $`\chi ,\eta ,\nu `$ models show the excellent fit to the weak charge of cesium atom $`Q_W(_{55}^{133}Cs)`$: the pull factor is reduced from 2.2 (SM) to less than 0.1. This may imply that more than 2-$`\sigma `$ deviation of the APV data from the SM prediction could be explained by the direct exchange of the $`Z_2`$ boson in the low-energy processes <sup>?</sup>. On the other hand, the $`\psi `$ model does not show the reduction of the pull factor in $`Q_W(_{55}^{133}Cs)`$. Since all the SM matter fields in the $`\psi `$ model have the same $`\mathrm{U}(1)^{}`$ charge (see Table 2.1), the couplings of contact interactions are parity conserving, which makes the contact term useless in the fit to the APV.
We introduce a parameter
$`\mathrm{\Delta }\chi ^2\chi _{\mathrm{min}}^2(\beta _E,\delta )\chi _{\mathrm{min}}^2(\mathrm{SM}),`$ (30)
to measure the goodness of the fit in the $`Z^{}`$ models compared to the SM. We can see from Table 3 that the $`\chi ,\eta `$ and $`\nu `$ models lead to $`\mathrm{\Delta }\chi ^2=5.5(\chi `$), $`4.6(\eta )`$ and $`5.4(\nu )`$, respectively while the $`\psi `$ model shows no improvement of the fit, $`\mathrm{\Delta }\chi ^2=0.3`$.
In order to see the impact of kinetic mixing on the fit, we show the contour plot of $`\mathrm{\Delta }\chi ^2`$ from the electroweak data under the conditions $`T_{\mathrm{new}},g_E^2/c_\chi ^2m_{Z_2}^20`$ on the $`(\beta _E,\delta )`$ plane in Fig. 1. We can see from the figure that the fit of the $`\eta `$ model at $`\delta =0`$ is rather worsen ($`\mathrm{\Delta }\chi ^23`$) as compared to that given in Table 3 ($`\mathrm{\Delta }\chi ^2=4.6`$) because the fit in the table has been found without imposing the constraint $`g_E^2/c_\chi ^2m_{Z_2}^20`$. The leptophobic $`\eta `$ model ($`\delta =1/3`$) does not improve the fit because, due to the leptophobity, the model does not have the contact term $`g_E^2/c_\chi ^2m_{Z_2}^2`$ which is used to make the fit to the LENC (essentially the APV) data better. We find that the $`Z^{}`$ model with $`(\beta _E,\delta )(\pi /4,0.2)`$ shows the most excellent fit over the SM where $`\mathrm{\Delta }\chi ^2\mathrm{\Gamma }<7`$.
### 4.2 Lower mass bound on $`Z^{}`$ bosons
As we expected from the formulae for $`T_{\mathrm{new}}`$ and $`\overline{\xi }`$ in the small mixing limit (2.2), the $`Z_2`$ mass is unbounded from the $`Z`$-pole data at $`\zeta =0`$. For models with very small $`\zeta `$, the lower bound of the heavier mass eigenstate $`Z_2`$ in the $`Z^{}`$ models, therefore, comes from the LENC experiments. In Fig. 2, we show the contour plot of the 95% CL lower mass limit of $`Z_2`$ boson from the LENC experiments on the $`(\beta _E,\delta )`$ plane by setting $`g_E=g_Y`$ and $`m_H=100\mathrm{GeV}`$ under the condition $`m_{Z_2}0`$. In practice, we obtain the 95% CL lower limit of the $`Z_2`$ boson mass $`m_{95}`$ in the following way:
$`0.05={\displaystyle \frac{{\displaystyle _{g_E^2/m_{95}^2}^{\mathrm{}}}d\left({\displaystyle \frac{g_E^2}{m_{Z_2}^2}}\right)P\left({\displaystyle \frac{g_E^2}{m_{Z_2}^2}}\right)}{{\displaystyle _0^{\mathrm{}}}d\left({\displaystyle \frac{g_E^2}{m_{Z_2}^2}}\right)P\left({\displaystyle \frac{g_E^2}{m_{Z_2}^2}}\right)}},`$ (31)
where we assume that the probability density function $`P(g_E^2/m_{Z_2}^2)`$ is proportional to $`\mathrm{exp}(\chi ^2(g_E^2/m_{Z_2}^2)/2)`$.
We can read off from Fig. 2 that the lower mass bound of the $`Z_2`$ boson in the $`\psi `$ model at $`\delta =0`$ is much weaker than those of the other $`Z^{}`$ models. This is because, as we mentioned before, the U(1) charge assignment on the SM matter fields in the model makes the constraint from the APV measurement useless. We also find in Fig. 2 that the lower mass bound of the $`Z_2`$ boson disappears near the leptophobic $`\eta `$-model ($`\beta _E=\mathrm{tan}^1(\sqrt{5/3})`$ and $`\delta =1/3`$<sup>?</sup>. Furthermore the lower mass bound tend to be small at the “best fit” point which we found in Fig. 1, $`(\beta _E,\delta )=(\pi /4,0.2)`$.
We summarize the 95% CL lower bound on $`m_{Z_2}`$ for the $`\chi ,\psi ,\eta `$ and $`\nu `$ models ($`\delta =0`$) in Table 4.2. For comparison, those in Ref. References are given in the same table. It should be noticed that the bounds on $`Z_\chi ,Z_\eta `$ and $`Z_\nu `$ masses are more severely constrained as compared to Ref. References due to the improved value of $`Q_W(_{55}^{133}Cs)`$ while the bound on the $`Z_\psi `$ mass is almost unchanged.
We have found that the $`Z`$-pole, $`m_W`$ and the LENC data constrain ($`T_{\mathrm{new}},\overline{\xi }`$), $`T_{\mathrm{new}}`$ and $`g_E^2/c_\chi ^2m_{Z_2}^2`$, respectively. We can see from Eq. (2.2) that, for a given $`\zeta `$, constraints on $`T_{\mathrm{new}},\overline{\xi }`$ and $`g_E^2/c_\chi ^2m_{Z_2}^2`$ can be interpreted as the bound on $`m_{Z_2}`$. We show the 95% CL lower mass bound of the $`Z_2`$ boson for $`m_H=100\mathrm{GeV}`$ in four $`Z^{}`$ models as a function of $`\zeta `$. The bound is again found under the condition $`m_{Z_2}0`$. Results are shown in Fig. 4.2(a) $``$ 4.2(d) for the $`\chi ,\psi ,\eta ,\nu `$ models, respectively. The lower bound from the $`Z`$-pole and $`m_W`$ data, and that from the LENC data are separately plotted in the same figure. Shown in the figure is the lower bound of $`m_{Z_2}g_Y/g_E`$ for $`g_E=g_Y`$. The bound on $`m_{Z_2}g_Y/g_E`$ is approximately independent of $`g_E`$ for $`g_E/g_Y=0.52.0`$ in each model <sup>?</sup>.
The $`Z_2`$ mass is unbounded from the $`Z`$-pole data at $`\zeta =0`$ because the data constrain $`T_{\mathrm{new}}`$ and $`\overline{\xi }`$ which are proportional to $`\zeta ^2`$ and $`\zeta `$, respectively. Then, the lower bound on $`m_{Z_2}`$ at very small $`\zeta `$ is obtained from the LENC experiments and the direct search experiment at Tevatron. For comparison, we plot the 95% CL lower bound on $`m_{Z_2}`$ obtained from the direct search experiment <sup>?</sup> in Fig. 4.2. In the direct search experiment, the $`Z^{}`$ decays into the exotic particles, e.g., the decays into the light right-handed neutrinos which are expected for some models, are not taken into account. We summarize the 95% CL lower bound on $`m_{Z_2}`$ for the $`\chi ,\psi ,\eta `$ and $`\nu `$ models ($`\delta =0`$) obtained from the low-energy data and the direct search experiment <sup>?</sup> in Table 4.2. The lower bound of $`m_{Z_2}`$ in the $`\eta `$ model from the LENC experiments is competitive the bound from the direct search experiment.
The lower bound of $`m_{Z_2}`$ is affected by the Higgs boson mass through the $`T`$ parameter. As we mentioned previously, $`T_{\mathrm{new}}`$ tends to be in the physical region ($`T_{\mathrm{new}}0`$) for large $`m_H`$ $`(x_H)`$. Then, we find that the large Higgs boson mass decreases the lower bound of $`m_{Z_2}`$. For $`\zeta =1`$, the lower $`m_{Z_2}`$ bound in the $`\chi ,\psi ,\nu `$ ($`\eta `$) models for $`m_H=150\mathrm{GeV}`$ is weaker than that for $`m_H=100\mathrm{GeV}`$ about 7% (11%). On the other hand, the Higgs boson with $`m_H=80\mathrm{GeV}`$ makes the lower $`m_{Z_2}`$ bound in all the $`Z^{}`$ models severe about 5% as compared to the case for $`m_H=100\mathrm{GeV}`$. Because $`T_{\mathrm{new}}`$ and $`\overline{\xi }`$ are proportional to $`\zeta ^2`$ and $`\zeta `$, respectively (see Eq. (2.2)), and it is unbounded at $`|\zeta |0`$, the lower bound of $`m_{Z_2}`$ may be independent of $`m_H`$ in the small $`|\zeta |`$ region. The $`m_H`$-dependence of the lower mass bound obtained from the LENC data is safely negligible.
It has been discussed that the presence of $`Z_2`$ boson whose mass is much heavier than the SM $`Z`$ boson mass, say 1 TeV, may lead to a find-tuning problem to stabilize the electroweak scale against the $`\mathrm{U}(1)^{}`$ scale <sup>?</sup>. The $`Z_2`$ boson with $`m_{Z_2}1\mathrm{TeV}`$ for $`g_E=g_Y`$ is allowed by the electroweak data only if $`\zeta `$ satisfies the following condition:
$`\begin{array}{cc}0.5\mathrm{\Gamma }<\zeta \mathrm{\Gamma }<+0.4\hfill & \mathrm{for}\mathrm{the}\chi ,\psi ,\nu \mathrm{models},\hfill \\ 0.6\mathrm{\Gamma }<\zeta \mathrm{\Gamma }<+0.6\hfill & \mathrm{for}\mathrm{the}\eta \mathrm{model}.\hfill \end{array}`$ (34)
## 5 Light $`Z^{}`$ boson in minimal SUSY $`E_6`$-models
It is known that the gauge couplings are not unified in the $`E_6`$ models with three generations of 27. In order to guarantee the gauge coupling unification, a pair of weak-doublets, $`H^{}`$ and $`\overline{H^{}}`$, should be added into the particle spectrum at the electroweak scale <sup>?</sup>. They could be taken from $`\mathrm{𝟐𝟕}+\overline{\mathrm{𝟐𝟕}}`$ or the adjoint representation 78. The $`\mathrm{U}(1)^{}`$ charges of the additional weak doublets should have the same magnitude and opposite sign, $`a`$ and $`a`$, to cancel the $`\mathrm{U}(1)^{}`$ anomaly. In addition, a pair of the complete SU(5) multiplet such as $`\mathrm{𝟓}+\overline{\mathrm{𝟓}}`$ can be added without spoiling the unification of the gauge couplings <sup>?,?</sup>.
The minimal $`E_6`$ model which have three generations of 27 and a pair $`\mathrm{𝟐}+\overline{\mathrm{𝟐}}`$ depends in principle on the three cases; $`H^{}`$ has the same quantum number as $`L`$ or $`H_d`$ of 27, or $`\overline{H_u}`$ of $`\overline{\mathrm{𝟐𝟕}}`$. In the following we represent the hypercharge and the U(1) quantum numbers of the additional pair as $`(1/2,a)`$ for $`H^{}`$ and $`(+1/2,a)`$ for $`\overline{H^{}}`$, where the U(1) quantum number $`a`$ in each $`Z^{}`$ model follows the same normalization in Table 2.1.
In the minimal model, the following eight scalar-doublets can develop VEV to give the mass terms $`m_Z^2`$ and $`m_{ZZ^{}}^2`$ in Eq. (8): three generations of $`H_u,H_d`$, and an extra pair, $`H^{}`$ and $`\overline{H^{}}`$. We express their VEVs as follows:
$`{\displaystyle \underset{i=1}{\overset{3}{}}}H_u^i^2={\displaystyle \frac{v_u^2}{2}},{\displaystyle \underset{i=1}{\overset{3}{}}}H_d^i^2={\displaystyle \frac{v_d^2}{2}},H^{}^2={\displaystyle \frac{v_H^{}^2}{2}},\overline{H^{}}^2={\displaystyle \frac{v_{\overline{H^{}}}^2}{2}},`$ (35)
where $`i`$ is the generation index. The sum of these VEVs gives the observed $`\mu `$-decay constant: $`v_u^2+v_d^2+v_H^{}^2+v_{\overline{H^{}}}^2v^2=\frac{1}{\sqrt{2}G_F}(246\mathrm{GeV})^2`$. By further introducing the notation
$`\mathrm{tan}\beta ={\displaystyle \frac{v_u}{v_d}},x^2={\displaystyle \frac{v_H^{}^2+v_{\overline{H^{}}}^2}{v^2}},`$ (36)
we can express $`\zeta `$ in Eq. (21) as <sup>?</sup>
$`\zeta `$ $`=`$ $`2\left\{Q_E^{H_u}(1x^2)\mathrm{sin}^2\beta +Q_E^{H_d}(1x^2)\mathrm{cos}^2\beta +Q_E^H^{}x^2\right\}\delta .`$ (37)
Because $`H^{}`$ and $`\overline{H^{}}`$ are taken from 27 \+ $`\overline{\mathrm{𝟐𝟕}}`$, the $`\mathrm{U}(1)^{}`$ charge of $`H^{}`$, $`Q_E^H^{}`$, is identified with that of $`L`$, $`H_d`$ or $`\overline{H_u}`$.
Let us remind the reader that, in the $`\chi `$ model, three generations of the matter fields 16 and a pair of Higgs doublets make the model anomaly free. In this case, $`\zeta `$ is found to be independent of $`\mathrm{tan}\beta `$:
$`\zeta `$ $`=`$ $`2Q_E^{H_d}\delta .`$ (38)
We can now examine the kinetic mixing parameter $`\delta `$ in each model. The boundary condition of $`\delta `$ at the GUT scale is $`\delta =0`$. The non-zero kinetic mixing term can arise at low-energy scale through the following RGEs:
$`{\displaystyle \frac{d}{dt}}\alpha _i`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}b_i\alpha _i^2,`$ (39a)
$`{\displaystyle \frac{d}{dt}}\alpha _4`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}(b_E+2b_{1E}\delta +b_1\delta ^2)\alpha _4^2,`$ (39b)
$`{\displaystyle \frac{d}{dt}}\delta `$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}(b_{1E}+b_1\delta )\alpha _1,`$ (39c)
where $`i=1,2,3`$ and $`t=\mathrm{ln}\mu `$. We define $`\alpha _1`$ and $`\alpha _4`$ as
$`\alpha _1{\displaystyle \frac{5}{3}}{\displaystyle \frac{g_Y^2}{4\pi }},\alpha _4{\displaystyle \frac{5}{3}}{\displaystyle \frac{g_E^2}{4\pi }}.`$ (40)
The coefficients of the $`\beta `$-functions for $`\alpha _1,\alpha _4`$ and $`\delta `$ are:
$`b_1={\displaystyle \frac{3}{5}}\mathrm{Tr}(Y^2),b_E={\displaystyle \frac{3}{5}}\mathrm{Tr}(Q_E^2),b_{1E}={\displaystyle \frac{3}{5}}\mathrm{Tr}(YQ_E).`$ (41)
From Eq. (39c), we can clearly see that the non-zero $`\delta `$ is generated at the weak scale if $`b_{1E}0`$ holds. In Table 5, we list $`b_1,b_E`$ and $`b_{1E}`$ in the minimal $`\chi ,\psi ,\eta `$ and $`\nu `$ models. As explained above, the $`\chi (16)`$ model has three generations of 16, and the $`\chi (27)`$ model has three generations of 27. We can see from Table 5 that the magnitudes of the differences $`b_1b_2`$ and $`b_2b_3`$ are common among all the models including the minimal supersymmetric SM (MSSM). This guarantees the gauge coupling unification at $`\mu =m_{GUT}10^{16}\mathrm{GeV}`$. It is straightforward to obtain $`g_E(m_{Z_1})`$ and $`\delta (m_{Z_1})`$ for each model. The analytical solutions of Eqs. (39a)$``$ (39c) are given in Ref. References. In our calculation, $`\alpha _3(m_{Z_1})=0.118`$ and $`\alpha (m_{Z_1})=e^2(m_{Z_1})/4\pi =1/128`$ are used as example. These numbers give $`g_Y(m_{Z_1})=0.357`$. We summarize the predictions for $`g_E`$ and $`\delta `$ at $`\mu =m_{Z_1}`$ in the minimal $`E_6`$ models in Table 5. In all the minimal models, the ratio $`g_E/g_Y`$ is approximately unity and $`|\delta |`$ is smaller than about 0.07. Some further extra fields, therefore, may be needed to give $`\delta =0.2`$ which leads to the “minimal $`\mathrm{\Delta }\chi ^2`$” when $`\beta _E=\pi /4`$, which we found in Fig. 1. We also show the result of the quasi leptophobic $`\eta `$ model ($`\eta _{\mathrm{BKM}}`$) proposed by Babu et al. <sup>?</sup> in the same table. The $`\eta _{\mathrm{BKM}}`$ has, besides three generation of 27, two pairs of $`\mathrm{𝟐}+\overline{\mathrm{𝟐}}`$ from $`\mathrm{𝟕𝟖}`$ and a pair of $`\mathrm{𝟑}+\overline{\mathrm{𝟑}}`$ from $`\mathrm{𝟐𝟕}+\overline{\mathrm{𝟐𝟕}}`$ in order to achieve the leptophobity ($`\delta 1/3)`$ at the weak scale through the quantum corrections. We find that the $`\eta _{\mathrm{BKM}}`$ model predicts $`g_E/g_Y0.86`$ and $`\delta 0.29`$, which is rather close to the leptophobity, $`\delta =1/3`$.
Next we estimate the parameter $`\zeta `$ for several sets of $`\mathrm{tan}\beta `$ and $`x`$. In Table 7, we show the predictions for $`\zeta `$ in the minimal $`\chi ,\psi ,\eta `$ and $`\nu `$ models. The results are shown for $`\mathrm{tan}\beta =2`$ and $`30`$, and $`x^2=0`$ and $`0.5`$. We find from the table that the parameter $`\zeta `$ is in the range $`|\zeta |\mathrm{\Gamma }<\mathrm{\hspace{0.17em}1.35}`$. It is shown in Fig. 4.2 that $`m_{Z_2}g_Y/g_E`$ is approximately independent of $`g_E/g_Y`$. Actually, we find in Table 5 and Table 7 that the predicted $`|\delta |`$ is smaller than about 0.1 and $`g_E/g_Y`$ is quite close to unity in all the minimal models. We can, therefore, read off from Fig. 4.2 the lower bound of $`m_{Z_2}`$ in the minimal models at $`g_E=g_Y`$. In Table 8, we summarize the 95% CL lower $`m_{Z_2}`$ bound for the minimal $`\chi ,\psi ,\eta `$ and $`\nu `$ models which correspond to the predicted $`\zeta `$ in Table 7.
Most of the lower mass bounds in Table 8 exceed 1 TeV. The $`Z_2`$ boson with $`m_{Z_2}O(1\mathrm{TeV})`$ should be explored at the future collider such as LHC. The discovery limit of the $`Z^{}`$ boson in the $`E_6`$ models at LHC is expected as (in unit of $`\mathrm{GeV}`$<sup>?</sup>
$`\begin{array}{cccc}& & & \\ \chi & \psi & \eta & \nu \\ & & & \\ 3040& 2910& 2980& \end{array}`$ (44)
All the lower bounds of $`m_{Z_2}`$ listed in Table 8 are smaller than 2 TeV and they are, therefore, in the detectable range of LHC. But, it should be noticed that most of them ($`1\mathrm{TeV}\mathrm{\Gamma }<m_{Z_2}`$) may require the fine-tuning to stabilize the electroweak scale against the $`\mathrm{U}(1)^{}`$ scale <sup>?</sup>.
## 6 Summary
In this review article, we have studied constraints on $`Z^{}`$ bosons in the SUSY $`E_6`$ models. Four $`Z^{}`$ models — the $`\chi ,\psi ,\eta `$ and $`\nu `$ models are studied in detail. The presence of the $`Z^{}`$ boson affects the electroweak processes through the effective $`Z`$-$`Z^{}`$ mass mixing angle $`\overline{\xi }`$, a tree level contribution $`T_{\mathrm{new}}`$ and the contact term $`g_E^2/c_\chi ^2m_{Z_2}^2`$, where the latter two parameters are positive definite quantities. The $`Z`$-pole, $`m_W`$ and LENC data constrain ($`T_{\mathrm{new}},\overline{\xi }`$), $`T_{\mathrm{new}}`$ and $`g_E^2/c_\chi ^2m_{Z_2}^2`$, respectively. From the updated electroweak data, we find that three $`Z^{}`$ models ($`\chi ,\eta ,\nu `$) improve the fit over the SM where the total $`\chi ^2`$ decrease about five units, owing to the excellent fit mainly to the improved data of parity violation in cesium atom which is expressed by the weak charge $`Q_W(_{55}^{133}Cs)`$. The more than 2-$`\sigma `$ deviation of $`Q_W(_{55}^{133}Cs)`$ from the SM prediction could be explained in these three $`Z^{}`$ models. Due to its parity conserving property of the U(1) charge assignment on the SM matter fields, the $`\psi `$ model does not improve the fit to the $`Q_W(_{55}^{133}Cs)`$ data. The impact of the kinetic mixing $`(\delta 0)`$ on the fit is also examined on the $`(\beta _E,\delta )`$ plane. The $`Z^{}`$ model with ($`\beta _E,\delta )=(\pi /4,0.2`$) shows the most excellent fit to the data among the SUSY $`E_6`$ models where the total $`\chi ^2`$ decreases by about seven units as compared to the SM best fit. The 95% CL lower mass bound of the heavier mass eigenstate $`Z_2`$ is shown as a function of the effective $`Z`$-$`Z^{}`$ mixing parameter $`\zeta `$ together with the result of direct search experiment. By assuming the minimal particle content of the $`E_6`$ model, we have found the theoretical predictions for $`\zeta `$. It is shown that the $`E_6`$ models with minimal particle content which is consistent with the gauge coupling unification predict the non-zero kinetic mixing term $`\delta `$ and the effective mixing parameter $`\zeta `$ of order one. The present electroweak experiments lead to the lower mass bound of order 1 TeV or larger for those models.
Acknowledgements
The author would like to thank K. Hagiwara and Y. Umeda for fruitful collaborations which this report is based upon. He is also grateful to R. Barbieri for reading manuscript.
References
|
warning/0002/hep-th0002159.html
|
ar5iv
|
text
|
# Untitled Document
PUPT-1919 OHSTPY-HEP-T-00-002 hep-th/0002159
Gravity Duals of Supersymmetric $`SU(N)\times SU(N+M)`$ Gauge Theories
Igor R. Klebanov<sup>1</sup> and Arkady A. Tseytlin<sup>2</sup> Also at Lebedev Physics Institute, Moscow and Imperial College, London.
<sup>1</sup> Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544
<sup>2</sup> Department of Physics, The Ohio State University, Columbus, OH 43210
Abstract
The world volume theory on $`N`$ regular and $`M`$ fractional D3-branes at the conifold singularity is a non-conformal $`𝒩=1`$ supersymmetric $`SU(N+M)\times SU(N)`$ gauge theory. In previous work the Type IIB supergravity dual of this theory was constructed to leading non-trivial order in $`M/N`$: it is the $`AdS_5\times T^{1,1}`$ background with NS-NS and R-R 2-form fields turned on. Far in the UV this dual description was shown to reproduce the logarithmic flow of couplings found in the field theory. In this paper we study the supersymmetric RG flow at all scales. We introduce an ansatz for the 10-d metric and other fields and show that the equations of motion may be derived in first order form from a simple superpotential. This allows us to explicitly solve for the gravity dual of the RG trajectory.
02/00
1. Introduction
The AdS/CFT correspondence \[1,,2,,3\] is usually motivated by comparing stacks of elementary branes with corresponding gravitational backgrounds in string or M-theory. For example, the correspondence between a large number $`N`$ of coincident D3-branes and the 3-brane classical solution leads, after an appropriate low-energy limit is taken, to the duality between $`𝒩=4`$ supersymmetric $`SU(N)`$ gauge theory and Type IIB strings on $`AdS_5\times S^5`$ \[1,,2,,3\]. This construction gives an explicit realization of the gauge theory strings \[5,,6\].
In order to construct the Type IIB duals of other 4-dimensional CFT’s, one may place the D3-branes at appropriate conical singularities \[7,,8,,9,,10\]. Then the background dual to the CFT on the D3-branes is $`AdS_5\times X_5`$ where $`X_5`$ is the Einstein manifold which is the base of the cone. In addition to regular D-branes which can reside on or off the conical singularity, there are also “fractional” D-branes pinned to the singularity \[11,,12\]. In previous work the effect of such fractional branes on the dual supergravity background was considered, and it was shown how they break the conformal invariance. In this paper we continue this line of investigation, and calculate the back-reaction of the fractional branes on the gravitational background. We obtain and solve a system of first-order equations describing renormalization group (RG) flow in the gravity dual of the $`𝒩=1`$ supersymmetric $`SU(N)\times SU(N+M)`$ gauge theory. This theory is realized on D-branes at the conifold singularity and we review it in section 2.
Our study of the gravitational RG flow builds on recent work studying such flows in gauged 5-d supergravity \[14,,15,,16,,17,,18,,19,,20\]. In particular, we will make use of the results on $`𝒩=1`$ supersymmetric flows \[16,,21,,22\] which reduce second-order equations to a much simpler first-order gradient flow induced by a superpotential function (extensions of these methods to non-supersymmetric flows were given in \[17,,23,,18,,24\]). In our example we start with an ansatz for the 10-d background and reduce it to a 5-d gauged supergravity coupled to scalar fields. This gives a clear geometrical meaning to the scalars fields, so that we can follow the RG evolution of the entire 10-d background. This is similar in spirit, although not in detail, to examples of RG flow found in 10-d type 0 string theory \[25,,26,,27,,28\]. An interesting novel feature of the solution we find is that its existence crucially depends on the presence of the Chern-Simons term in the type IIB supergravity action.
2. RG Flow Associated with Fractional Branes on the Conifold
The conifold is a singular Calabi-Yau manifold described in terms of complex variables $`w_1,\mathrm{},w_4`$ by the equation $`_{a=1}^4w_a^2=0.`$ The base of this cone is $`T^{1,1}=(SU(2)\times SU(2))/U(1)`$ whose Einstein metric may be written down explicitly as follows \[30,,29\],
$$ds_{T^{1,1}}^2=\frac{1}{9}\left(d\psi +\underset{i=1}{\overset{2}{}}\mathrm{cos}\theta _id\varphi _i\right)^2+\frac{1}{6}\underset{i=1}{\overset{2}{}}\left(d\theta _i^2+\mathrm{sin}^2\theta _id\varphi _i^2\right).$$
The $`𝒩=1`$ superconformal field theory on $`N`$ regular D3-branes placed at the singularity of the conifold has gauge group $`SU(N)\times SU(N)`$ and global symmetry $`SU(2)\times SU(2)\times U(1)`$ which is a symmetry of the metric (2.1). The theory contains two chiral superfields $`A_i`$ transforming as $`(N,\overline{N})`$ and as a doublet of the first $`SU(2)`$, and two chiral superfields $`B_k`$ transforming as $`(\overline{N},N)`$ and as a doublet of the second $`SU(2)`$. The R-charge of all four chiral superfields is $`1/2`$ and the theory has an exactly marginal superpotential $`𝒲=ϵ^{ij}ϵ^{kl}\mathrm{Tr}A_iB_kA_jB_l`$.
Type IIB supergravity modes on $`AdS_5\times T^{1,1}`$ have been matched in some detail with operators in this gauge theory whose dimensions are of order $`1`$ in the large $`N`$ limit \[31,,32,,33\]. Topologically, $`T^{1,1}`$ is $`S^2\times S^3`$ so that additional objects may be constructed from wrapped branes . In particular, a D5-brane wrapped over the 2-cycle acts as a domain wall in $`AdS_5`$. If this domain wall is located at $`r=r_w`$ then, by studying the behavior of wrapped D3-branes upon crossing it, it was shown in that for $`r>r_w`$ the gauge group changes to $`SU(N+1)\times SU(N)`$. This is precisely the gauge theory expected on $`N`$ regular and one fractional D3-branes. Thus, a D5-brane wrapped over the 2-cycle is nothing but a fractional D3-brane placed at a definite $`r`$. The identification of a fractional D3-brane with a wrapped D5-brane is consistent with the results of \[35,,12,,36,,37\].
Adding $`M`$ fractional D3-branes thus produces $`SU(N+M)\times SU(N)`$ supersymmetric gauge theory coupled to the chiral superfields $`A_i`$ and $`B_k`$. Its supergravity dual carries $`M`$ units of the R-R 3-form, $`H_{RR}`$, flux through the 3-cycle of $`T^{1,1}`$. In it was shown that this flux induces a radial variation of the integral of the NS-NS 2-form potential $`_{S^2}B_2`$. This was interpreted as the stringy dual of the logarithmic RG flow in the field theory. In the next section we set up the gravitational RG flow equations systematically, so that the back-reaction of the 3-form field strengths on other fields may be calculated. This will allow us to follow the flow far into the infrared and address the issues related to singularities.
3. The Supergravity Ansatz and the Effective Action
The type IIB supergravity equations can be obtained from the action
$$S_{10}=\frac{1}{2\kappa _{10}^2}d^{10}x(\sqrt{g_{10}}[R_{10}\frac{1}{2}(\mathrm{\Phi })^2\frac{1}{12}e^\mathrm{\Phi }(B_2)^2$$
$$\frac{1}{2}e^{2\mathrm{\Phi }}(𝒞)^2\frac{1}{12}e^\mathrm{\Phi }(C_2𝒞B_2)^2\frac{1}{45!}F_5^2]\frac{1}{24!(3!)^2}ϵ_{10}C_4C_2B_2+\mathrm{}),$$
$$(B_2)_{MNK}3_{[M}B_{NK]},(C_4)_{MNKLP}5_{[M}C_{NKLP]},$$
$$F_5=C_4+5(B_2C_2C_2B_2),$$
with the additional on-shell constraint $`F_5=F_5`$ . In these equations were solved to leading order in $`M/N`$, and it was shown that the back-reaction on the metric and 5-form fields enters at order $`(M/N)^2`$. To study the back-reaction, we introduce the following ansatz. The 10-d Einstein frame metric will be chosen as a sum of the 5-d space-time metric and the internal 5-manifold metric which has the same symmetries as (2.1):
$$ds_{10}^2=L^2\left[e^{\frac{2}{3}(B+4C)}ds_5^2+ds_5^{}^2\right],$$
$$ds_5^{}^2=\frac{1}{9}e^{2B}\left(d\psi +\underset{i=1}{\overset{2}{}}\mathrm{cos}\theta _id\varphi _i\right)^2+\frac{1}{6}e^{2C}\underset{i=1}{\overset{2}{}}\left(d\theta _i^2+\mathrm{sin}^2\theta _id\varphi _i^2\right).$$
Here $`B,C`$ are in general functions of the 5-d space-time coordinates and the conformal factor of the 5-d metric is chosen to preserve the Einstein frame upon compactification to 5-d. The numerical coefficients $`1/9`$ and $`1/6`$ (which are the same as in (2.1)) are inserted in order to have $`B=C=0`$ for the $`AdS_5\times T^{1,1}`$ solution. $`L`$ is the scale related to the radius of $`AdS_5`$ ($`L^4N`$).
We set the RR scalar $`𝒞`$ to zero (this will be consistent with the ansatz for 3-form fields made below) and study the case where
$$ds_5^2=du^2+e^{2A(u)}dx_ndx_n,$$
and $`B,C`$ and the 10-d dilaton $`\mathrm{\Phi }`$ are functions of $`u`$. Following we shall note that since the fractional D3-brane, i.e. the wrapped D5-brane, creates R-R 3-form flux through $`T^{1,1}`$, $`H_{\mathrm{RR}}=dC_2`$ should be proportional to the closed 3-form. This 3-form was constructed in ,
$$H_{\mathrm{RR}}=Pe^\psi \omega _2,\omega _2\frac{1}{\sqrt{2}}(e^{\theta _1}e^{\varphi _1}e^{\theta _2}e^{\varphi _2}).$$
Here $`P`$ is a constant proportional to the integer number $`M`$ of $`H_{RR}`$ flux units. In the normalization where $`L=1`$ which we shall use below, $`PM/N`$ ($`N`$ is fixed by the boundary conditions at $`u=u_0`$). We have introduced the orthonormal basis of 1-forms
$$e^\psi =\frac{1}{3}(d\psi +\underset{i=1}{\overset{2}{}}\mathrm{cos}\theta _id\varphi _i),e^{\theta _i}=\frac{1}{\sqrt{6}}d\theta _i,e^{\varphi _i}=\frac{1}{\sqrt{6}}\mathrm{sin}\theta _id\varphi _i.$$
The form of the NS-NS 2-form potential is also as in
$$B_2=T(u)\omega _2,H_{\mathrm{NS}}=T^{}(u)du\omega _2.$$
$`T`$ plays the role of a scalar field in the effective 5-d supergravity theory.
The natural ansatz for the self-dual 5-form is
$$F_5=+,$$
$$=K(u)\mathrm{vol}(\mathrm{T}^{1,1})=K(u)e^\psi e^{\theta _1}e^{\varphi _1}e^{\theta _2}e^{\varphi _2},$$
$$=e^{4A\frac{8}{3}(B+4C)}Kdudx^1dx^2dx^3dx^4.$$
Then $`H_{\mathrm{RR}}`$ in (3.1) satisfies the required equation $`dH_{\mathrm{RR}}F_5H_{\mathrm{NS}}`$ since $`F_5H_{\mathrm{NS}}=0`$.
Note that our ansatz preserves the symmetry between the two 2-sphere factors in (3.1). Had we put in different warp factors for the two 2-spheres, there would still be a solution where they are equal. This is due to the special form of the ansatz for the 3-form field strengths. Thus, our ansatz is rigidly constrained and, as we will see, leads to rather simple RG equations.
The $`F_5`$ equation of motion implies a relation between the functions $`K`$ and $`T`$. Indeed, $`dF_5=dF_5H_{\mathrm{NS}}H_{\mathrm{RR}}`$ implies that the scalars $`K`$ and $`T`$ are not independent:
$$K^{}=PT^{},\mathrm{i}.\mathrm{e}.K(u)=Q+PT(u).$$
For $`P=0`$ the constant $`Q`$ plays the role of the 5-brane charge, $`QN`$. For non-zero $`P`$ the constant $`Q`$ can be absorbed into a redefinition of the function $`T`$. This follows from the fact that $`F_5=dC_4+5(B_2dC_2C_2dB_2)=dC_4^{}+10B_2dC_2`$, where $`C_4^{}`$ is related by a field redefinition to $`C_4`$,
$$C_4=C_4^{}+5B_2C_2.$$
Since $`d(dC_4^{})=0`$, $`dC_4^{}`$ must contain the volume 5-form $`\mathrm{vol}(\mathrm{T}^{1,1})`$ part with constant coefficient $`Q`$.
The equation for $`T`$ which follows from $`dH_{\mathrm{NS}}F_5H_{\mathrm{RR}}`$ has the structure $`^\mu (e^\mathrm{\Phi }_\mu T)P(Q+PT)`$.
In general, the dilaton will be running according to
$$^2\mathrm{\Phi }=\frac{1}{12}(e^\mathrm{\Phi }H_{\mathrm{RR}}^2e^\mathrm{\Phi }H_{\mathrm{NS}}^2).$$
However, as we shall explain below, there exists a special class of solutions for which $`\mathrm{\Phi }`$ remains constant. Then $`H_{\mathrm{NS}}^2=e^{2\mathrm{\Phi }}H_{\mathrm{RR}}^2`$, or $`T^{}=Pe^\mathrm{\Phi }e^{\frac{4}{3}(B+C)}`$, i.e. $`T`$ can be expressed in terms of $`B`$ and $`C`$.
The full set of equations for the 5-d metric function $`A`$ and scalars $`B,C,\mathrm{\Phi },T`$ can be found from the following 5-d action which can be obtained from (3.1) by taking into account the solution of the $`F_5`$ equation of motion and integrating by parts
$$S_5=\frac{2}{\kappa _5^2}d^5x\sqrt{g_5}\left[\frac{1}{4}R_5\frac{1}{2}G_{ab}(\phi )\phi ^a\phi ^bV(\phi )\right].$$
Here the scalar fields $`\phi ^a=(q,f,\mathrm{\Phi },T)`$ include the “diagonal” combinations of $`B`$ and $`C`$
$$q=\frac{2}{15}(B+4C),f=\frac{1}{5}(BC),$$
which measure the volume and the ratio of scales of the internal manifold (3.1). In the presence of a 5-d cosmological constant both turn out to be positive (mass)<sup>2</sup> scalars (see below). Explicitly, in (3.1)
$$G_{ab}(\phi )\phi ^a\phi ^b=15(q)^2+10(f)^2+\frac{1}{4}(\mathrm{\Phi })^2+\frac{1}{4}e^{\mathrm{\Phi }4f6q}(T)^2,$$
$$V(\phi )=e^{8q}(e^{12f}6e^{2f})+\frac{1}{8}P^2e^{\mathrm{\Phi }+4f14q}+\frac{1}{8}(Q+PT)^2e^{20q}.$$
The kinetic term for $`T`$ comes from the $`e^\mathrm{\Phi }(B_2)^2`$ term in (3.1). The three terms in the potential have the following origin. The first combination in (3.1) comes from $`R_{10}`$ in (3.1) and reflects the curvature of the internal space (3.1) present in the limit of constant “radii” $`e^B`$ and $`e^C`$. The second term is the $`e^\mathrm{\Phi }(C_2)^2`$ evaluated on the solution (3.1). The third term corresponds to $`F_5^2`$. Note that the contribution of the Chern-Simons term in (3.1) is already effectively accounted for (it should not directly influence the 10-d gravitational part of equations of motion). As discussed above, and as apparent from the structure of the action (3.1),(3.1), for non-zero $`P`$ the constant $`Q`$ can be absorbed into $`T`$. This is an important feature of the system under consideration.
When written in terms of the 5-d metric function $`A(u)`$ and the scalars $`\phi ^a(u)`$ the action takes the following form
$$S_5=\frac{2\mathrm{Vol}_4}{\kappa _5^2}𝑑ue^{4A}\left[3A^2\frac{1}{2}G_{ab}(\phi )\phi ^a\phi ^bV(\phi )\right].$$
The set of equations obtained by varying $`A,q,f,\mathrm{\Phi },T`$ should be supplemented by the “zero-energy” constraint
$$3A^2\frac{1}{2}G_{ab}(\phi )\phi ^a\phi ^b+V(\phi )=0.$$
The simplest “fixed-point” solution found for $`P=0`$ corresponds to the $`AdS_5`$ space: when all scalars are constant, $`V=5+\frac{1}{8}Q^2`$. In the normalization where $`L=1`$ this gives the $`AdS_5`$ space of unit radius, i.e. $`A=u`$, for $`Q=4`$.
The reader may be slightly puzzled by the origin of the rescaling that sends $`NQ`$ and $`MP`$ (recall that $`M`$ is the actual number of $`H_{RR}`$ quanta). If we reinstate the dependence on $`L`$ and $`g_s=e^\mathrm{\Phi }`$ then the kinetic term for $`T`$ in (3.1) has a factor of $`e^\mathrm{\Phi }/L^4`$, while the $`N^2`$ and $`(N+MT)^2`$ terms in the potential have factors $`e^\mathrm{\Phi }/L^4`$ and $`1/L^8`$ respectively. Noting that the scale of the string-frame 10-d metric, which is to be held fixed, is related to the scale of the Einstein-frame metric, $`LN^{1/4}`$, by $`L_s=(g_s)^{1/4}L(g_sN)^{1/4}`$, we find that these three factors become $`1/L_s^4`$, $`L_s^4/N^2`$ and $`1/N^2`$ respectively. That explains why $`M`$ becomes replaced by $`PM/N`$ while $`N`$ by $`Q1`$.
4. The First Order System and its Solution
In general, the action $`S_5`$ (3.1) generates a system of second-order differential equations. However, as was observed in \[16,,21\], in the case of solutions preserving some amount of supersymmetry this system can be replaced by first-order equations in $`u`$:
$$\phi ^a=\frac{1}{2}G^{ab}\frac{W}{\phi ^b},A^{}=\frac{1}{3}W(\phi ),$$
where the superpotential $`W`$ is a function of $`k`$ scalars $`\phi ^a`$ satisfying
$$V=\frac{1}{8}G^{ab}\frac{W}{\phi ^a}\frac{W}{\phi ^b}\frac{1}{3}W^2.$$
It is easy to check directly that (4.1),(4.1) imply the second-order equations following from (3.1),(3.1). This first-order form leads to a crucial simplification in finding the RG flow.<sup>1</sup> It was noted in \[17,,23,,18\] that, even in the absence of supersymmetry, any solution of the second-order system of equations that follows from (3.1) can be obtained as solution of the first-order system (4.1),(4.1). Given $`V(\phi )`$, one may, in principle, solve the non-linear equation (4.1) for $`W`$. The solution is not unique, depending on $`k`$ integration constants. In general, in the absence of supersymmetry, $`W`$ is not guaranteed to be simple (in particular, monotonic).
Since the RG trajectory we are after is dual to $`𝒩=1`$ supersymmetric gauge theory, we expect the equations of motion to be simply expressible in the first order form. We have succeeded in guessing a simple superpotential $`W`$ which governs these equations (for definiteness, we assume $`Q,P0`$):
$$W=e^{4q}(2e^{6f}+3e^{4f})+\frac{1}{2}(Q+PT)e^{10q}.$$
For $`Q=4,P=0`$ we get for small $`f`$ and $`q`$
$$W=360f^2+60q^2+\mathrm{},$$
which shows that $`f`$ and $`q`$ are a “good” (diagonal) choice of fields. The potential $`V`$ (3.1),(4.1) has the expansion
$$V=3+60f^2+240q^2+\mathrm{},$$
which shows that these fields have masses $`m_q^2=32`$ and $`m_f^2=12`$. The combination $`q`$ in (3.1) is the standard fixed scalar degree of freedom corresponding to the overall volume of the compact manifold. For the KK reduction on $`S^5`$, the superpotential for the fixed scalar was derived in .
Although we have not checked explicitly that the first-order flow generated by the superpotential (4.1) preserves $`𝒩=1`$ supersymmetry, we believe that it is the case. One indirect check of the supersymmetry is to set $`P=0`$ and to consider the linearized equation for $`q`$,
$$q^{}=4q.$$
Its solution is $`qe^{4u}`$ which corresponds to adding a source for an operator of dimension 8 \[2,,3\] (schematically, this operator has a $`\mathrm{Tr}F^4`$ structure ). This describes the leading perturbation from the $`AdS_5\times T^{1,1}`$ background toward the metric of $`N`$ D3-branes at the conifold, which is known to preserve $`𝒩=1`$ supersymmetry. In section 6 we exhibit the full BPS 3-brane solution, which serves as a consistency check on the first-order equations we have derived.
Note that, although $`V`$ in (3.1) depends on the dilaton $`\mathrm{\Phi }`$, the superpotential $`W`$ does not depend on it.<sup>2</sup> The $`\mathrm{\Phi }`$ dependence of $`V`$ is reproduced in (4.1) due to the dilaton dependence of the “metric” $`G_{ab}`$ entering the kinetic term of $`T`$ in (3.1). As a result, the dilaton remains constant along the flow! In what follows we shall set $`\mathrm{\Phi }=0`$.
The system of equations for $`T,f,q`$ that follow from (4.1),(4.1) is thus:
$$T^{}=Pe^{4q+4f},$$
$$f^{}=\frac{3}{5}e^{4q+4f}(1e^{10f}),$$
$$q^{}=\frac{2}{15}e^{4q+4f}(3+2e^{10f})\frac{1}{6}(Q+PT)e^{10q}.$$
Note that for $`P=0`$ this system has a stable fixed point solution $`f=0,q=\frac{1}{6}\mathrm{ln}(Q/4),T=\mathrm{const}`$ mentioned above. It corresponds to the $`AdS_5\times T^{1,1}`$ background with unit radius, and $`q=0`$ if $`Q=4`$. Therefore, we will use $`Q=4`$ below.
To fix the boundary conditions, we assume that above a UV cut-off scale $`u=u_0`$ we have the superconformal theory with $`PM=0`$. Physically this corresponds to inserting $`M`$ fractional anti-branes at $`u=u_0`$. Thus we set
$$q=f=0,T=T_0\mathrm{at}u=u_0,$$
and consider the solution for $`u<u_0`$. The constant $`T_0`$ determines the value of $`g_1^2g_2^2`$ at the UV cut-off \[9,,13\] ($`g_1,g_2`$ are the two gauge coupling constants). Since $`W/f`$ (the r.h.s. of (4.1)) vanishes for $`f=0`$, we find that $`f(u)=0`$ along the entire RG flow. This means that the shape of the internal manifold $`T^{1,1}`$ does not change, only its overall size does!
This simplifies our task to finding just two functions, $`T(u)`$ and $`q(u)`$. The first order equations for them are governed by the superpotential
$$W=5e^{4q}+\frac{1}{2}(Q+PT)e^{10q},$$
which is (4.1) with $`f`$ set to zero. Thus, we have
$$T^{}=Pe^{4q},$$
$$q^{}=\frac{2}{3}e^{4q}\frac{1}{6}(Q+PT)e^{10q}.$$
Introducing the variables
$$K(u)=Q+PT(u),Y(u)=e^{6q(u)},$$
we get from (4.1)
$$K^{}=P^2e^{4q}=P^2Y^{2/3}.$$
Using also (4.1) we find that
$$\frac{dY}{dK}=\frac{1}{P^2}(4YK).$$
This has a general solution
$$Y=a_0e^{4K/P^2}+\frac{K}{4}+\frac{P^2}{16}.$$
The constant $`a_0`$ has to be chosen to implement the UV boundary condition that $`Y=1`$ when $`K=K_0=4+PT_0`$:
$$a_0=\left(\frac{P^2}{16}+\frac{PT_0}{4}\right)\mathrm{exp}\left[\frac{16}{P^2}\frac{4T_0}{P}\right].$$
This completely fixes the relation between $`T`$ and $`q`$ along the RG trajectory. In particular, for small $`TT_0`$$`q=\frac{1}{6}T_0(TT_0)+\mathrm{}.`$ This is consistent with the perturbative solution of (4.1),(4.1)
$$T=T_0+P(uu_0)+\mathrm{},q=\frac{1}{6}PT_0(uu_0)+\mathrm{}.$$
Note that $`uu_0`$ translates into $`\mathrm{ln}(\mathrm{\Lambda }/\mathrm{\Lambda }_0)`$ in the field theory. Thus, the variation of $`T`$ translates into a logarithmic flow of $`g_1^1g_2^2`$ in the field theory confirming the result of . Furthermore, we can now calculate higher order corrections to the metric and $`T`$ in powers of $`PM/N`$.
Substituting (4.1) into (4.1) we find
$$K^{}=P^2\left[\frac{P^2}{16}+a_0e^{4K/P^2}+\frac{K}{4}\right]^{2/3},$$
which in turn leads to an implicit equation for $`K(u)`$,
$$u_0u=\frac{1}{P^2}_K^{K_0}𝑑z\left[\frac{P^2}{16}+a_0e^{4z/P^2}+\frac{z}{4}\right]^{2/3}.$$
Using this relation and (4.1) we also have a relation between $`q`$ and $`u`$.
To complete our solution, we need to find $`A(u)`$. The equation for the function $`A`$ in (4.1) has the form
$$A^{}=\frac{1}{3}\left[e^{4q+4f}(3+2e^{10f})\frac{1}{2}(Q+PT)e^{10q}\right]=q^{}+\frac{1}{P}T^{}+\frac{2}{3}f^{},$$
where we have used (4.1)–(4.1) to express the exponents in terms of the derivatives. Thus in general $`A`$ is simply a linear combination
$$A(u)=A_0+q(u)+\frac{2}{3}f(u)+\frac{1}{P}T(u).$$
For our particular trajectory with $`f=0`$, this gives $`A`$ in terms of $`q`$ and $`T`$. The integration constant $`A_0`$ may be shifted by a rescaling of 4-d coordinates $`x_n`$ in (3.1). We can choose it so that the metric (3.1) approaches the canonical $`AdS_5`$ one, i.e. $`A(u)u`$ for $`uu_0`$. The resulting expression for the 10-d metric (3.1),(3.1),(3.1) may be written as ($`L=1`$)
$$ds_{10}^2=e^{5q}du^2+e^{2A5q}dx_ndx_n+e^{3q}ds_{T^{1,1}}^2,$$
where we have used that for $`f=0`$ eq. (3.1) gives $`B=C=\frac{3}{2}q`$. Introducing the coordinate $`y`$ such that $`dy=e^{(AA_0)}du`$, we may write this metric in the form
$$ds_{10}^2=e^{3q+\frac{2}{P}T}(dy^2+dx_ndx_n)+e^{3q}ds_{T^{1,1}}^2.$$
5. Solution in a Special Case
There is a simple choice of the boundary condition for $`T`$ in (4.1), $`T_0=P/4`$, which leads to $`a_0=0`$. Then an explicit solution of the RG equations is straightforward: (4.1) becomes
$$Y=\frac{K}{4}+\frac{P^2}{16},$$
and we find from (4.1) that
$$Y^{}=\frac{P^2}{4}Y^{2/3},$$
i.e.
$$Y(u)=e^{6q(u)}=a_1P^{6/5}(uu_s)^{3/5},a_1=(5/12)^{3/5}.$$
Then we have
$$K=4+PT(u)=\frac{1}{4}P^2+4a_1P^{6/5}(uu_s)^{3/5}.$$
Obviously, $`u_s`$ is the position of the singularity where $`T^{1,1}`$ shrinks to vanishing size. To find the relation of $`u_s`$ to the UV cut-off $`u_0`$ we note that the boundary condition (4.1) implies $`Y(u_0)=1`$, i.e.
$$u_0u_s=\frac{12}{5P^2}.$$
The effective scale factor in 5-d gauged supergravity metric (3.1) is given by (see (4.1))
$$e^{2A}P^{2/5}(uu_s)^{1/5}\mathrm{exp}[8a_1P^{4/5}(uu_s)^{3/5}].$$
The fact that this vanishes at $`u=u_s`$ seems to indicate the presence of a naked singularity in the geometry. However, recall that in the 10-d metric (4.1) the effective scale factor in front of $`dx_ndx_n`$ is
$$e^{2A5q}P^{3/5}(uu_s)^{3/10}\mathrm{exp}[8a_1P^{4/5}(uu_s)^{3/5}],$$
which, in fact, blows up at $`u_s`$. Thus, in order to study the singularity structure, it is essential to know the full 10-d form of the solution.
6. More General Solutions
In this section we go beyond the ‘near-horizon’ approximation and construct asymptotically flat solutions of the first-order system of equations (4.1). For $`P=0`$ our solution describes regular D3-branes at the conifold singularity. For $`P0`$ we find an interesting generalization of this solution with a logarithmically running charge.
Let us look for 10-d metric of the following 4+6 form
$$ds_{10}^2=s^{1/2}(r)dx_ndx_n+h^{1/2}(r)(dr^2+r^2ds_{T^{1,1}}^2).$$
The relation of this to the notation of (4.1) is
$$s^{1/2}(r)=e^{2A(u)5q(u)},e^{3q(u)/2}=rh^{1/4}(r).$$
The new radial coordinate, $`r`$, is related to $`u`$ through
$$e^{4q(u)}du=\frac{dr}{r}.$$
Hence, multiplying the last two equations, we have
$$h^{1/4}(r)dr=e^{5q(u)/2}du,$$
which shows that (6.1) is equivalent to (4.1).
Using (4.1) we get
$$dT=Pe^{4q}du=Pd(\mathrm{ln}r),$$
which has a solution
$$T(r)=\stackrel{~}{T}+P\mathrm{ln}r.$$
We already know the solution (4.1) for $`q(T)`$, hence we find
$$r^4h(r)=e^{6q}=a_0e^{4\stackrel{~}{Q}/P^2+4\mathrm{ln}r}+\frac{1}{4}(\stackrel{~}{Q}+P^2\mathrm{ln}r)+\frac{1}{16}P^2,$$
where
$$\stackrel{~}{Q}Q+P\stackrel{~}{T}.$$
Thus, we have the following explicit solution for $`h(r)`$:
$$h(r)=b_0+\frac{k_0+P^2\mathrm{ln}r}{4r^4},$$
where
$$k_0=\stackrel{~}{Q}+\frac{1}{4}P^2,b_0=a_0e^{4\stackrel{~}{Q}/P^2}.$$
To solve for $`s(r)`$ we note that according to (6.1),(4.1)
$$s(r)=e^{4T(r)/P+6q(r)}=\frac{1}{r^4}e^{6q(r)}=h(r).$$
Remarkably, the 10-d metric thus assumes the usual “D-brane” form
$$ds_{10}^2=h^{1/2}(r)dx_ndx_n+h^{1/2}(r)(dr^2+r^2ds_{T^{1,1}}^2).$$
The function $`h(r)`$, given in (6.1), may be written as
$$h(r)=b_0+\frac{P^2\mathrm{ln}\frac{r}{r_{}}}{4r^4},$$
where $`r_{}`$ is defined by
$$P^2\mathrm{ln}r_{}=Q+P\stackrel{~}{T}+\frac{1}{4}P^2.$$
The Ricci scalar for this 10-d metric turns out to have has the following simple form
$$R=\frac{5r^1h^{}(r)+h^{\prime \prime }(r)}{2h^{3/2}(r)}=\frac{4P^2}{\left(4b_0r^4+P^2\mathrm{ln}\frac{r}{r_{}}\right)^{3/2}}.$$
This makes it clear that the metric becomes singular at $`r=r_s`$ such that $`h(r_s)=0`$.
For $`P=0`$ and $`b_0>0`$ the solution (6.1) is precisely the BPS metric of D3-branes placed at the conifold singularity (note that (6.1) vanishes in this case, as it should). This shows that, at least in this case, the solution to the first order system preserves supersymmetry.
For $`P0`$, the solution is still asymptotically flat. Note in particular that the NS-NS field strength falls off at large $`r`$, $`H_{NS}=\frac{P}{r}dr\omega _2`$. A remarkable property of the solution (6.1) is that it looks like a threebrane metric whose effective charge and mass per unit volume depend on the radius logarithmically. To strengthen this interpretation, let us consider a test D3-brane oriented along the source D3-branes and placed at radial coordinate $`r`$. As in the case where the transverse 6-d CY space in (6.1) is replaced by $`R^6`$, the gravitational force on the test brane is proportional to the derivative of the metric function (6.1),
$$\mu (r)=r^5\frac{dh}{dr}=K(r),K(r)=Q+PT(r)=\stackrel{~}{Q}+P\mathrm{ln}r.$$
For this reason $`K(r)`$ may be thought of as the mass per unit volume enclosed inside radius $`r`$. Note that this is different from the coefficient of the $`\frac{1}{4r^4}`$ term in (6.1) (i.e. $`K(r)+\frac{1}{4}P^2`$), but is the same as the coefficient in the 5-form field strength (3.1). This is in agreement with the expected BPS nature of this configuration. Indeed, the force on the static D3-brane probe oriented parallel to the source brane will vanish as a result of the balance between the electric force proportional to the 5-form component $``$ in (3.1) and the gravitational force proportional to (6.1). The cancellation of forces is also another argument in favor of our solution preserving $`𝒩=1`$ supersymmetry.
To study the solution in more detail, we have to distinguish the cases where $`b_0`$ is positive, zero, or negative. The asymptotically flat region exists only if $`b_0>0`$. In this case, and for $`P=0`$, we find the AdS horizon at $`r=0`$. For $`P0`$, however, the situation is completely different because the enclosed 3-brane charge or mass density at radius $`r`$ is $`K(r)`$. As $`r`$ decreases, so does this effective 3-brane charge density. At $`r=r_e>r_{}`$ where $`K(r_e)=0`$, the gravitational force changes sign, i.e. inside this radius we have antigravity. Thus, $`r=r_e`$ is similar to the enhancon radius found in a different setting in . It is not hard to check that the metric is nonsingular at $`r=r_e`$ (this is obvious from (6.1)). Continuing to $`r<r_e`$ past $`r_{}`$ we eventually reach the singularity where $`h(r_s)=0`$, i.e. $`4b_0r_s^4=P^2\mathrm{ln}\frac{r_{}}{r_s}`$. One may speculate that this singularity should be “excised” because it occurs in the region where the effective 3-brane charge and tension are negative.
The case of $`b_0=a_0=0`$ is the one discussed in the preceding section. The exact metric found there is simply (6.1) expressed in terms of a different radial coordinate. Again, in this case the singularity occurs at $`r`$ smaller than $`r_e`$, the point where the effective 3-brane charge per unit volume vanishes. Indeed, from (5.1) we see that $`K=\frac{1}{4}P^2`$ is negative at the singularity.
Finally, let us consider the case $`b_0<0`$. Now $`r`$ cannot increase indefinitely: as $`r`$ increases we find a singularity at $`r_+`$ where $`h`$ vanishes. For very small $`r`$ there is another curvature singularity located at $`r_{}>r_{}`$. Thus, for $`b_0<0`$ there are two curvature singularities, and we have $`r_{}<r_{}<r_+`$. It is not clear, however, if the case $`b_0<0`$ is physical: if we interpret the metric (6.1) as the geometry around $`N`$ regular and $`M`$ fractional D3-branes placed at the conifold singularity, then this geometry is required to have the asymptotically flat region at large $`r`$.
7. Gauge Theory Interpretation and Comments
In this section we summarize some main points and further discuss the gravitational solution dual to the RG flow in the supersymmetric $`SU(N+M)\times SU(N)`$ gauge theory. In investigating the actual solution, we first consider the case $`PM/N1`$. Then we see from (4.1) that near the UV cut-off $`u_0`$ both derivatives $`T^{}`$ and $`q^{}`$ are small, hence the gravity approximation is valid. As $`u`$ decreases from $`u_0`$, $`K=Q+PT`$ decreases monotonically. At the value $`u_e`$ given by
$$u_e=u_0\frac{1}{P^2}_0^{K_0}𝑑z\left[\frac{P^2}{16}+a_0\mathrm{exp}(4z/P^2)+\frac{z}{4}\right]^{2/3}$$
$`K`$ reaches $`0`$. Since the 5-form field strength vanishes at $`u_e`$, this location is similar to the enhancon radius found in . As already mentioned above, for $`u<u_e`$ we find ‘antigravity,’ and it is plausible to assume that this region has to be excised in a full string theoretic treatment.
If we continue the effective gravity solution to $`u<u_e`$, we find a singularity of the metric: the value $`u_s`$ where $`Y=e^{6q}=0`$. Since $`e^{16/P^2}`$ is negligible for small $`P`$, we can see from (4.1) that $`K(u_s)P^2/4`$. Using (4.1) and (4.1) we can derive the behavior of the metric functions $`A`$, $`q`$ and $`K`$. The leading behavior coincides with that found in the exact solution exhibited in section 5: near the singularity both $`q`$ and $`A`$ diverge to $`\mathrm{}`$ as $`\frac{1}{10}\mathrm{ln}(uu_s)`$. While $`e^{2A}`$ vanishes, we note again that in the 10-dimensional metric (4.1) the conformal factor is not $`A`$, but rather
$$A\frac{5}{2}q\frac{3}{20}\mathrm{ln}(uu_s),$$
i.e. the longitudinal part of the 10-d metric expands rather than contracts as we approach $`u_s`$. The 10-d metric (4.1) is nevertheless singular because of the volume of $`T^{1,1}`$ shrinking to zero. In particular, the 10-d Ricci scalar is $`R_{10}[P^2(uu_s)]^{3/10}`$. This is why the gravity approximation near $`u_s`$ has to be taken with a grain of salt: stringy corrections could alter the conclusions entirely. We find it suggestive, however, that far in the infrared the compact 5-manifold seems to be removed dynamically – this is a desirable feature for understanding the dynamics of realistic gauge theories . Perhaps one day it will be possible to understand the effective 5-d string theory where the singular compact manifold is ‘integrated out.’
The fate of the singularity is an interesting issue. From (4.1) we see that, for all $`a_0>P^2/16`$, the singularity is hidden behind the enhancon-type locus $`K(u)=0`$ where the effective 3-brane tension vanishes. Thus, following we may conjecture that such singularities have to be excised in a string-theoretic treatment. Since negative $`a_0`$ may be unphysical, this protection of singularities may be a general phenomenon in the system we are considering.
Another issue we need to address is the fact that a change of $`T`$ shifts the effective 3-brane charge. Given that $`T`$ is scale dependent, it therefore appears that far in the IR the gauge group is different from that found in the UV. Let us suggest the following qualitative picture. Since $`Q`$ scales as $`N`$, and $`P`$ scales as $`M`$, from the point of view of the dual $`SU(N+M)\times SU(N)`$ gauge theory, we conjecture that $`N`$ starts decreasing dynamically as the theory flows to the IR. At first, the variation of $`T`$ may be interpreted as the variation of $`g_1^2g_2^2`$ in the field theory . But what happens when we reach a value of $`u`$ where one of the couplings diverges? Since a shift of $`T`$ corresponds to a shift of $`Q`$, the natural continuation past this infinite coupling involves the field theory with $`N`$ replaced by $`NM`$. Repeating this reasoning many times we seem to eventually arrive at a theory with $`N`$ comparable to $`M`$ or even at the theory with a single gauge group $`SU(M)`$. Presumably, this is the theory described by the vicinity of $`u_e`$ where $`K`$ is near 0. This is an intriguing conjecture, but of course we need further checks to establish it, even on a qualitative level, because of difficulties with the effective gravity approximation.
In view of the above, it appears that, even if we start with $`MN`$, the theory dynamically drives itself into a regime where $`N`$ and $`M`$ are comparable. A natural question then is: why can’t we start with $`PM/N`$ of order one from the beginning. Then the problem is that, even in the UV, the flow is no longer slow and the supergravity approximation is suspect. Nevertheless, it might provide a useful qualitative picture. Even if $`P`$ is of order 1, the solution typically exhibits a repulson singularity hidden behind the enhancon-type locus, similar to those found in .
Acknowledgements
We are grateful to R. Corrado, N. Nekrasov, K. Pilch and N. Warner for useful discussions. The work of I. R. K. was supported in part by the NSF grant PHY-9802484 and by the James S. McDonnell Foundation Grant No. 91-48. The work of A.T. is supported in part by the DOE grant DOE/ER/01545, PPARC SPG grant PPA/G/S/1998/00613, EC TMR grant ERBFMRX-CT96-0045, INTAS grant No.96-538, and NATO grant PST.CLG 974965.
References
relax J. Maldacena, “The Large N limit of superconformal field theories and supergravity,” Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200 relax S.S. Gubser, I.R. Klebanov, and A.M. Polyakov, “Gauge theory correlators from noncritical string theory,” Phys. Lett. B428 (1998) 105, hep-th/9802109 relax E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150 relax S.S. Gubser and I.R. Klebanov, “Absorption by Branes and Schwinger Theory,” Phys. Lett. B413 (1997) 41, hep-th/9708005; for a review see I.R. Klebanov, “From Threebranes to Large N Gauge Theories,” hep-th/9901018. relax G. ’t Hooft, “A Planar Diagram Theory for Strong Interactions,” Nucl. Phys. B72 (1974) 461. relax A.M. Polyakov, “String Theory and Quark Confinement,” Nucl. Phys. B (Proc. Suppl.) 68 (1998) 1, hep-th/9711002. relax M. Douglas and G. Moore, “D-branes, quivers, and ALE instantons,” hep-th/9603167 relax A. Kehagias, “New Type IIB Vacua and Their F-Theory Interpretation,” Phys. Lett. B435 (1998) 337, hep-th/9805131. relax I.R. Klebanov and E. Witten, “Superconformal Field Theory on Threebranes at a Calabi-Yau Singularity,” Nucl. Phys. B536 (1998) 199, hep-th/9807080. relax D.R. Morrison and M.R. Plesser, “Non-Spherical Horizons, I,” Adv. Theor. Math. Phys. 3 (1999) 1, hep-th/9810201 relax E.G. Gimon and J. Polchinski, “Consistency Conditions for Orientifolds and D Manifolds”, Phys. Rev. D54 (1996) 1667, hep-th/9601038 relax M.R. Douglas, “Enhanced Gauge Symmetry in M(atrix) theory”, JHEP 007(1997) 004, hep-th/9612126 relax I.R. Klebanov and N. Nekrasov, “Gravity Duals of Fractional Branes and Logarithmic RG Flow,” hep-th/9911096. relax L. Girardello, M. Petrini, M. Porrati and A. Zaffaroni, “Novel Local CFT and Exact Results on Perturbations of N=4 Super Yang Mills from AdS Dynamics,” JHEP 9812 (1998) 022, hep-th/9810126; “Confinement and condensates without fine tuning in supergravity duals of gauge theories,” JHEP 9905 (1999) 026, hep-th/9903026 . relax J. Distler and F. Zamora, “Nonsupersymmetric conformal field theories from stable anti-de Sitter spaces,” Adv. Theor. Math. Phys. 2 (1999) 1405, hep-th/9810206; “Chiral Symmetry Breaking in the AdS/CFT Correspondence,” hep-th/9911040. relax D.Z. Freedman, S.S. Gubser, K. Pilch and N.P. Warner, “Renormalization Group Flows from Holography–Supersymmetry and a c-Theorem,” hep-th/9904017 relax O. DeWolfe, D.Z. Freedman, S.S. Gubser and A. Karch, “Modeling the fifth dimension with scalars and gravity,” hep-th/9909134. relax J. de Boer, E. Verlinde and H. Verlinde, “On the Holographic Renormalization Group,” hep-th/9912012 relax M. Cvetic, H. Lu and C.N. Pope, “Domain Walls and Massive Gauged Supergravity Potentials,” hep-th/0001002. relax I. Bakas, A. Brandhuber and K. Sfetsos, “Domain walls of gauged supergravity, M-branes, and algebraic curves,” hep-th/9912132. relax L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, “The Supergravity Dual of N=1 Super Yang-Mills Theory,” hep-th/9909047 relax S.S. Gubser, “Non-conformal examples of AdS/CFT,” hep-th/9910117 relax K. Skenderis and P. K. Townsend, Phys. Lett. B468 (1999) 46, hep-th/9909070; K. Behrndt and M. Cvetic, “Supersymmetric domain wall world from D = 5 simple gauged supergravity,” hep-th/9909058 relax A.M. Polyakov, “A Few Projects in String Theory,” in: Les Houches Summer School on Gravitation and Quantizations, Session 57, Les Houches, France, 5 Jul - 1 Aug 1992, J. Zinn-Justin and B. Julia eds. (North-Holland, 1995) hep-th/9304146 relax I.R. Klebanov and A.A. Tseytlin, “D-Branes and Dual Gauge Theories in Type 0 Strings,” Nucl. Phys. B546 (1999) 155, hep-th/9811035 relax J. Minahan, “Glueball Mass Spectra and Other Issues for Supergravity Duals of QCD Models,” JHEP 9901 (1999) 020, hep-th/9811156; “Asymptotic Freedom and Confinement from Type 0 String Theory,” JHEP 9904 (1999) 007, hep-th/9902074 relax I.R. Klebanov and A.A. Tseytlin, “Asymptotic Freedom and Infrared Behavior in the Type 0 String Approach to Gauge Theory,” Nucl. Phys. B547 (1999) 143, hep-th/9812089 relax C. Angelantonj and A. Armoni, “Non-Tachyonic Type 0B Orientifolds, Non-Supersymmetric Gauge Theories and Cosmological RG Flow,” hep-th/9912257. relax P. Candelas and X. de la Ossa, “Comments on Conifolds,” Nucl. Phys. B342 (1990) 246. relax D. N. Page and C. N. Pope,“Which Compactifications Of D = 11 Supergravity Are Stable?,” Phys. Lett. B144 (1984) 346. relax S.S. Gubser, “Einstein Manifolds and Conformal Field Theories,” Phys. Rev. D59 (1999) 025006, hep-th/9807164. relax D.P. Jatkar and S. Randjbar-Daemi, “Type IIB string theory on $`AdS_5\times T^{nn^{}}`$,” Phys. Lett. B460 (1999) 281, hep-th/9904187. relax A. Ceresole, G. Dall’Agata, R. D’Auria, and S. Ferrara, “Spectrum of Type IIB Supergravity on $`AdS_5\times T^{1,1}`$: Predictions On $`𝒩=1`$ SCFT’s,” Phys. Rev. D61 (2000) 066001, hep-th/9905226 relax S.S. Gubser and I.R. Klebanov, “Baryons and Domain Walls in an N=1 Superconformal Gauge Theory,” Phys. Rev. D58 (1998) 125025, hep-th/9808075. relax C.V. Johnson and R.C. Myers, “Aspects of Type $`II`$B Theory on ALE Spaces”, Phys. Rev. D55 (1997) 6382, hep-th/9610140 relax D.-E. Diaconescu, M. Douglas and J. Gomis, “Fractional Branes and Wrapped Branes,” JHEP 02 (1998) 013, hep-th/9712230 relax K. Dasgupta and S. Mukhi, “Brane Constructions, Fractional Branes and Anti-de Sitter Domain Walls,” JHEP 9907, 008 (1999), hep-th/9904131. relax J. H. Schwarz, “Covariant Field Equations Of Chiral N=2 D = 10 Supergravity,” Nucl. Phys. B226 (1983) 269. relax E. Bergshoeff, C. Hull and T. Ortin, “Duality in the type II superstring effective action,” Nucl. Phys. B451 (1995) 547, hep-th/9504081 relax S. Gubser, A. Hashimoto, I. R. Klebanov and M. Krasnitz, “Scalar Absorption and the Breaking of the World Volume Conformal Invariance,” Nucl. Phys. B526 (1998) 393. relax C. Johnson, A. Peet and J. Polchinski, “Gauge Theory and the Excision of Repulson Singularities,” hep-th/9911161 relax A.M. Polyakov, “The Wall of the Cave,” Int. J. Mod. Phys. A14 (1999) 645, hep-th/9809057
|
warning/0002/hep-ph0002217.html
|
ar5iv
|
text
|
# Implications of the Unitarity Triangle ‘uc’ for J, 𝛿 and |𝑉_{𝐶𝐾𝑀}| elements.
## Abstract
The Jarlskog rephasing invariant parameter $`|J|`$ is evaluated using one of the six Unitarity Triangles involving well known CKM matrix elements $`|V_{ud}|`$, $`|V_{us}|`$$`|\frac{V_{ub}}{V_{cb}}|`$, $`|V_{cd}|`$, $`|V_{cs}|`$ and $`|V_{cb}|`$. With PDG2000 values of $`|V_{ud}|`$ etc. as input, we obtain $`|J|=(2.71\pm 1.12)\times 10^5`$, which in the PDG representation of CKM matrix leads to the range $`21^oto159^o`$ for the CP violating phase $`\delta `$. The CKM matrix elements evaluated using this range of $`\delta `$ are in agreement with the PDG CKM matrix. The implications of refinements in the input on $`|J|`$, $`\delta `$ and CKM matrix elements have also been studied.
Recent discovery of neutrino oscillations in atmospheric neutrinos by the SuperKamiokande Collaboration has not only given the first clear cut signal for physics beyond the standard model (SM), but has also triggered great amount of activity in neutrino mixing phenomena as well as in the related issue of fermion mass matrices. This has also given an impetus to study more deeply the quark mixing phenomena which have been under investigation for the last two decades. In fact, there is a need to closely examine the quark mixing phenomena in the hope that one may decipher some signal, howsoever faint, for physics beyond the SM.
In the context of quark mixing phenomena, over the last two decades, several analyses have been carried out , some of these in the last few years . The basic purpose of these analyses has been the evaluation of Cabibbo, Kobayashi and Maskawa matrix ($`V_{CKM}`$) elements defined as,
$$\left(\begin{array}{c}d^{^{}}\\ s^{^{}}\\ b^{^{}}\end{array}\right)=\left(\begin{array}{ccc}V_{ud}& V_{us}& V_{ub}\\ V_{cd}& V_{cs}& V_{cb}\\ V_{td}& V_{ts}& V_{tb}\end{array}\right)\left(\begin{array}{c}d\\ s\\ b\end{array}\right).$$
(1)
The usual inputs for the analyses are the CP violating parameters, $`ϵ_K`$ and $`ϵ_K^{^{}}`$, as well as $`B_o\overline{B_o}`$ mixing phenomenon besides unitarity of CKM matrix defined as
$$\underset{\alpha =d,s,b}{}V_{i\alpha }V_{j\alpha }^{}=\delta _{ij},$$
(2)
$$\underset{i=u,c,t}{}V_{i\alpha }V_{i\beta }^{}=\delta _{\alpha \beta }.$$
(3)
where Latin subscripts run over the up type quarks $`(u,c,t)`$ and Greek ones run over the down type quarks $`(d,s,b)`$. These analyses have given considerable insight into the dynamics of CKM matrix elements and their consequences, in particular the unitarity triangle (UT) based analyses have considerably sharpened the relationship between the CP violation and $`B`$ \- decays. However, it is to be noted that in these analyses, the effect of unitarity, $`ϵ_K`$ and $`B_o\overline{B_o}`$ mixing etc. on the CKM matrix elements is carried out simultaneously. In other words, the separate implications of unitarity, $`ϵ_K`$ and $`B_o\overline{B_o}`$ mixing have not been studied, in particular, on such important quantities as CP violating phase $`\delta `$ and CKM matrix elements involving $`t`$ quark.
In view of the availability of rephasing the quark fields , the CKM matrix has 36 representations, therefore it has been advocated in the literature that the analysis of CKM phenomenology should be carried out in a rephasing invariant manner . In this context Jarlkog has defined an interesting quantity $`J`$ which is rephasing invariant as well as all CP violating effects within the CKM paradigm are proportional to it. Interestingly $`J`$ is also directly related to the commutator of the quark mass matrices , for example,
$`det[mm^{},m^{^{}}m^{{}_{}{}^{}}]`$ $`=`$ $`2iJ(m_t^2m_c^2)(m_c^2m_u^2)(m_u^2m_t^2)`$ (4)
$`\times (m_b^2m_s^2)(m_s^2m_d^2)(m_d^2m_b^2).`$
Therefore an evaluation of $`|J|`$ based on data is going to have important implications for texture specific mass matrices in the sense that it could provide valuable clues for searching the right texture for fermion mass matrices .
In the very recent analyses, Parodi et al and Mele primarily concentrate on finding CKM parameters in the Wolfenstein parametrization and the angles of unitarity triangle, while J. Swain et al determine CKM parameters with and without unitarity. The PDG analysis evaluates the CKM matrix elements using the measured CKM elements and the unitarity as implied by the nine equations given by 2 and 3. These analyses, however, have not been carried out in the rephasing invariant manner as well as they do not evaluate $`J`$. Therefore a rephasing invariant analysis of the CKM matrix based on unitarity is very much desirable in the hope that this may sharpen the predictions of unitarity for CKM matrix elements.
The purpose of the present Rapid Communication is to evaluate $`|J|`$, based on non zero CP violation and on the unitarity triangle expressed by the relation,
$$V_{ud}V_{cd}^{}+V_{us}V_{cs}^{}+V_{ub}V_{cb}^{}=0$$
(5)
and referred to as $`uc`$ triangle. This is the only unitarity triangle out of the six implied by equations 2 and 3 with $`ij`$ and $`\alpha \beta `$, which involves well determined CKM matrix elements. After evaluating $`|J|`$, we use the PDG representations of $`V_{CKM}`$ to find CP violating phase $`\delta `$ and the elements of CKM matrix involving $`t`$ quark. We also intend to examine the implications of present as well as future refinements in measured $`V_{CKM}`$ elements on fixing $`\delta `$.
To begin with, we evaluate $`|J|`$, defined as
$$Im[V_{\alpha j}V_{\beta k}V_{\alpha k}^{}V_{\beta j}^{}]=J\underset{\gamma ,l}{}ϵ_{\alpha ,\beta ,\gamma }ϵ_{j,k,l}.$$
(6)
In principle one can evaluate $`|J|`$ using the above formula, however, in practice it does not help much as it involves CP violating phase $`\delta `$, the least known CKM parameter. Therefore, for the purpose of our analysis, we exploited the relationship of $`|J|`$ with the unitarity triangle. Out of the six possible unitarity triangles we have used the triangle expressed through the equation 5. As mentioned earlier this triangle involves only those CKM elements which have been directly measured, consequently $`|J|`$ can be evaluated through the relation,
$$|J|=2\times AreaoftheUnitarityTriangle.$$
(7)
With the availabilty of PDG2000 CKM elements , it is natural to use these as input for evaluating $`|J|`$, however, with a view to understand the effect of refinements in CKM matrix elements on $`|J|`$, we have also done our calculations with PDG98 values. In the same vein we have also carried our calculations for “future” values of CKM elements which may be available in future. In table 1 we have given the PDG98 values of $`V_{CKM}`$ elements, $`|V_{ud}|`$, $`|V_{us}|`$, $`|\frac{V_{ub}}{V_{cb}}|`$, $`|V_{cd}|`$, $`|V_{cs}|`$ and $`|V_{cb}|`$, as well as the recent PDG2000 values and “future” values. While listing the “future” values, we have considered only those elements in which the present error is more than $`15\%`$.
Before proceeding further, it is to be noted that the triangle mentioned above is highly squashed. The sides of the triangle represented by $`|V_{ud}^{}V_{cd}|(=a)`$ and $`|V_{us}^{}V_{cs}|(=b)`$ are of comparable lengths while the third side $`|V_{ub}^{}V_{cb}|(=c)`$ is several orders of magnitude smaller compared to $`a`$ and $`b`$. This creates complications for evaluating the area of the triangle without violating unitarity and the existence of CP violation. To avoid these complications we have used the constraints $`|a|+|c|>|b|`$ and $`|b|+|c|>|a|`$ as suggested by Branco and Lavoura . Using these constraints and the experimental data given in the column II of table 1 (PDG98), we have generated a histogram shown in figure 1. In generating the histogram all inputs i.e. $`|V_{ud}|`$$`|V_{us}|`$$`|\frac{V_{ub}}{V_{cb}}|`$, $`|V_{cd}|`$, $`|V_{cs}|`$ and $`|V_{cb}|`$ have been taken at their $`90\%`$ confidence level facilitating comparison with the corresponding PDG analysis. A Gaussian is fitted into the histogram plotted with approximately 30,000 entries as shown in figure 1. The resulting value of $`|J|`$ is given as,
$$|J|=(2.28\pm 0.86)\times 10^5,$$
(8)
which in the $`90\%`$ C.L. leads to the range,
$$|J|=(0.873.69)\times 10^5.$$
(9)
The value of $`|J|`$ can now be used to calculate $`\delta `$ using the PDG representations of CKM matrix, for example,
$$V_{CKM}=\left(\begin{array}{ccc}c_{12}c_{13}\hfill & s_{12}s_{13}\hfill & s_{13}e^{i\delta }\hfill \\ s_{12}c_{23}c_{12}s_{23}s_{13}e^{i\delta }\hfill & c_{12}c_{23}s_{12}s_{23}s_{13}e^{i\delta }\hfill & s_{23}c_{13}\hfill \\ s_{12}s_{23}c_{12}c_{23}s_{13}e^{i\delta }\hfill & c_{12}s_{23}s_{12}c_{23}s_{13}e^{i\delta }\hfill & c_{23}c_{13}\hfill \end{array}\right),$$
(10)
with $`c_{ij}=cos\theta _{ij}`$ and $`s_{ij}=sin\theta _{ij}`$ for the generation labels $`i,j=1,2,3.`$ In the above representation, $`J`$ can be expressed as,
$$J=J^{^{}}sin\delta ,$$
(11)
where,
$$J^{^{}}=sin\theta _{12}sin\theta _{23}sin\theta _{13}cos\theta _{12}cos\theta _{23}cos^2\theta _{13}.$$
(12)
Before evaluating $`J^{^{}}`$, using equation 1 we calculate $`sin\theta _{12},sin\theta _{23}`$ and $`sin\theta _{13}`$ from the experimental values of $`|V_{us}|`$, $`|\frac{V_{ub}}{V_{cb}}|`$, and $`|V_{cb}|`$ given in table 1. The corresponding values of $`sin\theta _{12},sin\theta _{23}`$ $`sin\theta _{13}`$, presented in table 2, are used to evaluate $`J^{^{}}`$. Following the procedure outlined above for evaluating $`|J|`$, with all input values at their $`90\%`$ C.L., $`J^{^{}}`$ comes out to be,
$$J^{^{}}=(2.86\pm 0.76)\times 10^5.$$
(13)
The range corresponding to $`90\%`$ C.L. of $`J^{^{}}`$ can be easily found out and is given as,
$$J^{^{}}=(1.614.11)\times 10^5.$$
(14)
Using equation 11, we can find the range of $`\delta `$ corresponding to various C.Ls. of $`|J|`$ and $`J^{^{}}`$. In this regard in figure 2, we have plotted $`J^{^{}}sin\delta `$ as a function of $`\delta `$. The upper and lower sinusoidal curves correspond to upper and lower limits of $`J^{^{}}`$ given by equation 13. The horizontal lines depict upper and lower limits of $`|J|`$ given by equation 8. Since $`J^{^{}}sin\delta `$ should reproduce $`|J|`$ calculated through the unitarity triangle $`uc`$, therefore from figure 2, by comparing the two one can easily find out the widest limits on $`\delta `$, for example,
$$\delta =23^oto157^o.$$
(15)
The above range of $`\delta `$ corresponds to $`1\sigma `$ C.L. of both $`|J|`$ and $`J^{^{}}`$. Similarly, using 9 and 14, the corresponding range of $`\delta `$ at $`90\%`$ C.L. of $`|J|`$ and $`J^{^{}}`$ can be found out as shown in figure 3 and is given as,
$$\delta =12^oto168^o.$$
(16)
These values of $`\delta `$ apparently look to be the consequence only of the unitarity relationship given by equation 5. However on further investigation, as shown by Branco and Lavoura , one finds that these $`\delta `$ ranges are consequences of all the non trivial unitarity constraints. In this sense the above range could be attributed to as a consequence of unitarity of the CKM matrix.
Alternatively, using equation 11, one can find out $`\delta `$ for each of the 30,000 entries and likewise plot a histogram for $`\delta `$. Again fitting a Gaussian to the histogram, $`\delta `$ comes out to be,
$`\delta `$ $`=`$ $`51^o\pm 21^o(Iquadrant),`$ (17)
$`129^o\pm 21^o(IIquadrant).`$
The corresponding range of $`\delta `$ at $`90\%`$ C.L. is,
$`\delta `$ $`=`$ $`17^oto85^o(Iquadrant),`$ (18)
$`95^oto163^o(IIquadrant).`$
This gives us relatively stronger bounds on $`\delta `$. However, for the subsequent calculations we have used ranges of $`\delta `$ as given by 15 and 16.
As has been mentioned earlier, we intend to compare our results both with PDG98 as well as with PDG2000 CKM matrix. To begin with, as the input for $`|J|`$ has been from PDG98 values, therefore we compare our results with PDG98 CKM matrix. To calculate the CKM elements, we have used the values of sines of mixing angles at their $`90\%`$ C.L. (using column II of table 2) and $`\delta `$ at $`90\%`$ C.L. of $`|J|`$ (equation 16). The calculated $`V_{CKM}`$ matrix is,
$$\left[\begin{array}{ccc}0.97470.9765& 0.2160.223& 0.00190.0045\\ 0.2160.223& 0.97390.9757& 0.0370.042\\ 0.0040.014& 0.0350.042& 0.9992\end{array}\right].$$
(19)
To facilitate the comparison of corresponding CKM matrix elements as well as for easy readability, we present below the $`V_{CKM}`$ from the PDG98 also calculated at $`90\%`$ C.L..
$$\left[\begin{array}{ccc}0.97450.9760& 0.2170.224& 0.00180.0045\\ 0.2170.224& 0.97370.9753& 0.0360.042\\ 0.0040.013& 0.0350.042& 0.99910.9994\end{array}\right].$$
(20)
Comparing 19 and 20, one finds that we have been able to reproduce PDG matrix with minor differences only at fourth decimal places. In addition to the reproduction of CKM matrix elements, it needs to be emphasized that we have calculated $`|J|`$ and $`\delta `$ based entirely on unitarity and data, which to our knowledge has not been calculated earlier. The availabilty of $`|J|`$ and $`\delta `$ simplifies the task of calculating CKM matrix elements as well as giving a deeper insight into the contribution of CKM paradigm to CP violating phenomena.
The above method of evaluating $`|J|`$, $`\delta `$ and CKM matrix elements can be easily repeated for the PDG2000 as well as “future” values of input parameters mentioned in tables 1 and 2. The corresponding $`|J|`$ and $`\delta `$ have been given in table 3. For the sake of completeness, we have repeated the whole analysis with input values at their $`1\sigma `$ and $`3\sigma `$ C.Ls. and the corresponding results for $`|J|`$, $`J^{^{}}`$ and $`\delta `$ are listed in table 3.
A close look at the table 3 leads us to several interesting points. For example, the bounds on $`|J|`$ and $`\delta `$ are weak when the input values are taken at their $`3\sigma `$ C.L., but we get relatively stronger bounds when the analysis is done with the input values being at their $`90\%`$ and $`1\sigma `$ C.L.. Further, it is clear from the same table that there is not much change in the range of $`\delta `$ with PDG98 and PDG2000 values. However, if in future the ranges of $`|V_{cs}|`$ and $`|\frac{V_{ub}}{V_{cb}}|`$ are further constrained, lt may lead to significant narrowing in the range of $`\delta `$ so obtained, as shown in columns V and VI of row IV of table 3.
To give a better insight into the implications of PDG2000 values on our calculations, we present below the CKM matrix evaluated using values of sines of mixing angles at their $`90\%`$ C.L. (from column III of table 2) and $`\delta `$ at $`90\%`$ C.L. of $`|J|`$.
The calculated $`V_{CKM}`$ matrix is,
$$\left[\begin{array}{ccc}0.97560.9765& 0.2160.223& 0.0020.005\\ 0.2160.223& 0.97390.9757& 0.0370.043\\ 0.0050.013& 0.0360.043& 0.9992\end{array}\right].$$
(21)
To facilitate the comparison, we present below the $`V_{CKM}`$ from the PDG2000 also calculated at $`90\%`$ C.L..
$$\left[\begin{array}{ccc}0.97420.9757& 0.2190.226& 0.0020.005\\ 0.2190.225& 0.97340.9749& 0.0370.043\\ 0.0040.014& 0.0350.043& 0.99900.9993\end{array}\right].$$
(22)
Comparing the CKM matrices 21 and 22, we see that we are able to reproduce PDG2000 CKM matrix, once again justifying the procedure followed in carrying out the present analysis. To assess the impact of future refinements in CKM matrix elements $`|\frac{V_{ub}}{V_{cb}}|`$ and $`|V_{cs}|`$, we present below the CKM matrix elements corresponding to $`|J|`$ and $`\delta `$ as mentioned in row III of table 3.
$$\left[\begin{array}{ccc}0.97470.9765& 0.2160.223& 0.00290.0043\\ 0.2160.223& 0.97390.9757& 0.0370.043\\ 0.0060.013& 0.0360.043& 0.9992\end{array}\right].$$
(23)
Comparing 19, 21 and 23 we find that the $`|V_{CKM}|`$ matrix elements do not show much variation when the latest or the “future” values are used. The small changes in the CKM elements mostly at the fourth decimal places are not primarily due to change in the range of $`\delta `$, but due to overall changes in all the input parameters. This probably restricts the use of unitarity in evaluating the $`|V_{CKM}|`$ elements involving $`t`$ quark.
In conclusion, we would like to mention that using only one of the six unitarity triangles, we have evaluated Jarlskog rephasing invariant parameter $`|J|`$ and consequently $`\delta `$. Using this range of $`\delta `$ we have been able to reproduce the PDG matrix at $`90\%`$ C.L. evaluated by PDG group. Our calculations also indicate that improvements and further refinements in $`V_{CKM}`$ elements $`|V_{cs}|`$ and $`|\frac{V_{ub}}{V_{cb}}|`$ result in significant narrowing in the range of $`\delta `$, however there is no appreciable impact on $`V_{CKM}`$ elements involving $`t`$ quark, therefore, their range can be narrowed only by direct measurement of $`\delta `$. It needs to be mentioned that an evaluation of $`|J|`$ based on data is going to have important implications for texture specific mass matrices as the parameter $`J`$ is directly related to the mass matrices. Our conclusions in this regard would be published elsewhere.
ACKNOWLEDGMENTS
M.R. would like to thank CSIR, Govt. of India, for financial support. M.R. and P.S.G. would like to thank the chairman, Department of Physics, for providing facilities to work in the department. P.S.G. acknowledges the financial support received for his UGC project.
|
warning/0002/hep-th0002051.html
|
ar5iv
|
text
|
# References
Sparked in part by the discovery of the top quark, there has recently been a great deal of interest in the inter-quark potential, see, for example, . It has long been known that the pure QCD corrections to the Coulombic potential in 3$`+`$1 dimensions are of two types: a dominant anti-screening contribution and a lesser interaction which corresponds to screening by physical, transverse gluons. In this letter we will continue our programme to study the structure of the forces between quarks. We will demonstrate that in $`2+1`$ dimensions, which for Euclidean metrics is by dimensional reduction related to the high temperature limit of QCD, the inter-quark potential has an unexpectedly rich structure and that the relative weights of the attractive and repulsive interactions are almost identical to those of the $`3+1`$ case.
In $`SU(N)`$, for static quarks without additional light fermions, the inter-quark potential to order $`g^4`$ in $`d+1`$ space-time dimensions is given by
$$V(q)=\frac{g^2C_F}{𝒒^2}\left\{1+g^2\mu ^{2ϵ}C_A(4d1)\frac{|𝒒|^{d3}}{(16\pi )^{\frac{d}{2}}}\frac{\mathrm{\Gamma }(\frac{3d}{2})\mathrm{\Gamma }(\frac{d+1}{2})}{\mathrm{\Gamma }(\frac{d}{2}+1)}\right\},$$
(1)
where $`q=|𝒒|`$, $`C_A=N`$, $`C_F=(N^21)/(2N)`$ and $`d+1=42ϵ`$. In $`3+1`$ dimensions this reduces to the familiar form
$$V(q)=\frac{g^2C_F}{𝒒^2}\left\{1\frac{g^2}{(4\pi )^2}C_A\frac{11}{3}\mathrm{ln}\left(\frac{𝒒^2}{\mu ^2}\right)\right\},$$
(2)
while in $`2+1`$ dimensions this becomes
$$V(q)=\frac{g^2C_F}{𝒒^2}\left\{1+g^2C_A\frac{7}{32|𝒒|}\right\}.$$
(3)
Note that in $`2+1`$ dimensions the result is finite and no renormalisation is needed . In $`3+1`$ dimensions the order $`g^4`$ correction to the Coulombic potential is related to the universal beta function of QCD, but in $`2+1`$ dimensions no such identification is possible since the beta function vanishes. Finally, we recall that in $`2+1`$ dimensions the coupling constant is a dimensionful quantity.
These corrections to the potential have been understood in 3$`+`$1 dimensions as the sum of two distinct physical effects: a dominant anti-screening interaction which arises from the Coulombic potential, and a smaller screening interaction which arises from the virtual production of physical, i.e., gauge invariant, gluon pairs. The dominance of anti-screening over screening is the origin of QCD’s asymptotic freedom. Concretely the coefficient of the logarithmic correction can be decomposed as:
$$V(q)=\frac{g^2C_F}{𝒒^2}\left\{1\frac{g^2}{(4\pi )^2}C_A\left[4\frac{1}{3}\right]\mathrm{ln}\left(\frac{𝒒^2}{\mu ^2}\right)\right\},$$
(4)
where the factor of $`4`$ comes from the anti-screening interaction and the $`\frac{1}{3}`$ from the smaller screening forces. The relative strength of the screening part of the potential is, we note, only $`8.33\%`$ of the anti-screening contribution. Due to the universality of the beta function, this decomposition can be calculated in many different ways , although it cannot be obtained from the Wilson loop approach to the potential. This structure of the inter-quark potential was previously unknown in $`2+1`$ dimensions and we shall now calculate it.
We will follow the method of Ref. . The lowest energy states corresponding to two heavy quarks a distance $`r=|𝒓|`$ apart is $`|\overline{\psi }(𝒓)h(𝒓)h^1(0)\psi (0)`$, where the quarks are in the same time slice. We call $`h^1`$ a dressing for the matter field, $`\psi `$. This field dependent term is the lowest energy gluonic configuration around an individual fermion which maintains gauge invariance for the composite charged quark. The kinematics of the heavy quark determines the form of the dressing, and we have shown elsewhere that it factors into a product of two terms: a gauge dependent term which makes the dressed quark gauge invariant, and a gauge invariant structure.
The first part of the dressing is the minimal gluonic configuration which renders the quark gauge invariant. This first term originates from Gauss’ law and hence from longitudinal degrees of freedom. It produces the spreading of the colour charge, *anti-screening*, which underlies asymptotic freedom in non-abelian gauge theories, and will raise the energy of the quark-antiquark state. It is the non-abelian extension of the Coulomb interaction. Since the overall dressed quark has to correspond to the lowest energy state, the additional gauge invariant glue must lower the energy. As such, it can only correspond to a *screening* contribution. This physical decomposition into structures which necessarily raise and lower the energy is the correct identification of anti-screening and screening effects even in 2$`+`$1 dimensions where the coupling does not run.
Generalising the construction in to $`d`$-spatial dimensions, the anti-screening part of the potential at order $`g^4`$ is
$`V_{\mathrm{anti}}^{(4)}(r)`$ $`=`$ $`3g^4C_AC_F{\displaystyle \frac{\mathrm{\Gamma }^3(\frac{d}{2}1)}{64\pi ^{\frac{3d}{2}}}}{\displaystyle }\mathrm{d}^dz\mathrm{d}^dw{\displaystyle \frac{1}{|𝒛𝒘|^{d2}}}\times `$ (5)
$`\left(_j^z{\displaystyle \frac{1}{|𝒛𝒓|^{d2}}}\right)\left(_k^w{\displaystyle \frac{1}{|𝒘|^{d2}}}\right)0|A_j^\mathrm{T}(𝒛)A_k^\mathrm{T}(𝒘)|0,`$
where we have used the $`d`$-dimensional result
$$\left(\frac{1}{^2}f\right)(x^0,𝒙)=\frac{\mathrm{\Gamma }(\frac{d}{2}1)}{4\pi ^{\frac{d}{2}}}\mathrm{d}^dz\frac{f(x^0,𝒛)}{|𝒛𝒙|^{d2}}.$$
(6)
This reduces to Eq. 16 of for $`d=3`$. The gauge invariant, equal-time, free propagator in (5), in $`d+1`$ dimensions, is given by
$$0|A_j^\mathrm{T}(𝒛)A_k^\mathrm{T}(𝒘)|0=\frac{\mathrm{\Gamma }(\frac{d+1}{2})}{2\pi ^{\frac{d+1}{2}}}\frac{(zw)_j(zw)_k}{|𝒛𝒘|^{d+1}},$$
(7)
which can be understood as the Coulomb propagator for the spatial components or, in the Lorentz class of gauges, as the $`\xi =1`$ propagator (where $`\xi =1`$ corresponds to Feynman gauge). This last identification, based upon the requirement of the propagator being transverse to $`_i^x`$, allows us to use the computational power of covariant gauges. The dimension-independence of this gauge should be contrasted with the Yennie gauge, where the propagator is transverse to momentum derivatives *only* in $`d=3`$.
Combining these expressions results in a standard, finite integral which may be straightforwardly evaluated in $`d`$ dimensions. In $`2+1`$ dimensions, after Fourier transforming, we have
$$V_{\mathrm{anti}}^{(4)}(q)=\frac{3}{2}g^4C_AC_F\frac{1}{𝒒^2}\frac{\mathrm{d}^2l}{(2\pi )^2}\frac{1}{|𝒍|(𝒍𝒒)^2}\left(1\frac{(𝒒𝒍)^2}{𝒒^2𝒍^2}\right).$$
(8)
This is, of course, a *finite* integral. From this we can rapidly show that the anti-screening contribution to the potential at order $`g^4`$ is given by
$$V_{\mathrm{anti}}^{(4)}(q)=g^4C_FC_A\frac{3}{4\pi }\frac{1}{|𝒒|^3}.$$
(9)
Comparing this with the full potential in $`2+1`$ dimensions at this order (3) we see first of all that the factors of $`\pi `$ do not agree. However, the energy is still higher than the total result of (3) so it is still a physically acceptable result. We now want to show how the screening contribution supplies the different $`\pi `$ factors needed to lower the energy to the final physical result. We will, therefore, now independently calculate the screening contribution.
We shall follow an approach to the potential which has been presented by Gribov and by Drell . Working now in Coulomb gauge, the Hamiltonian is
$$H=\frac{1}{2}\mathrm{d}^2x\left(\left(𝑬_\mathrm{T}^a\right)^2+\left(𝑩^a\right)^2\varphi ^a^2\varphi ^a\right),$$
(10)
where we have decomposed the chromoelectric field into transverse and longitudinal components, $`𝑬^a=𝑬_T^a\mathbf{}\varphi ^a`$ and summation over colour is understood. Gauss’ law tells us that $`\varphi `$ is related to the static matter sources, $`\rho `$, and the gluonic fields by
$$^2\varphi ^a=g\left(\rho ^af_{abc}𝑨^b𝑬^c\right),$$
(11)
from which we can obtain the following equation up to order $`g^3`$, which is all we shall require in this letter:
$$^2\varphi ^a=\left\{g\delta ^{ae}+g^2f_{abe}𝑨^b\mathbf{}\frac{1}{^2}+g^3f_{abc}f_{cde}𝑨^b\mathbf{}\frac{1}{^2}𝑨^d\mathbf{}\frac{1}{^2}\right\}\left(\rho ^ef_{egh}𝑨^g𝑬_\mathrm{T}^h\right),$$
(12)
where $`^2`$ acts on whatever is on its right. We take the sources for simplicity to have the form $`\rho ^a=\rho _q^a+\rho _{\overline{q}}^a`$ where $`\rho _q^a(𝒙)=t_q^a\delta ^3(𝒙)`$, $`\rho _{\overline{q}}^a(𝒙)=t_{\overline{q}}^a\delta ^3(𝒙𝒓)`$. Here we assume that $`t_q^a`$ and $`t_{\overline{q}}^a`$ are the colour charges of a heavy (classical) quark $`q_i`$ and antiquark $`\overline{q}_j`$ in a normalized colour singlet state $`|\mathrm{\Psi }=N^{1/2}|q_i|\overline{q}_i`$. Hence the colour factor becomes
$$t_q^at_{\overline{q}}^a=\frac{1}{N}q_i|Q^a|q_j\overline{q}_i|Q^a|\overline{q}_j=\frac{1}{N}\mathrm{tr}(T^aT^a)=C_F,$$
(13)
where $`Q^a`$ is the colour charge operator and the anti-Hermitian generators $`T^a`$ are in the fundamental representation of $`SU(N)`$. The heavy quark and antiquark are again separated by $`𝒓`$. The sources only enter the Hamiltonian (10) in the last term, so the $`𝒓`$ dependent term here gives the potential between them.
Let us first explain how the lowest order result may be recovered in this approach. The relevant term in the Hamiltonian is
$$\frac{1}{2}\mathrm{d}^2x\varphi ^a^2\varphi ^a=g^2\mathrm{d}^2x\rho _{\overline{q}}(x)\frac{1}{^2}\rho _q(x),$$
(14)
where, as we are only interested in the potential, we have dropped separation independent terms. We may now evaluate the expectation value between the gluonic vacuum states. Expressing the delta functions as Fourier transforms and trivially performing the spatial integral, we obtain
$$V(q)=g^2C_F\frac{1}{𝒒^2},$$
(15)
from which we can read off the three dimensional generalisation of the Coulomb interaction, i.e., the lowest order term in (3).
We may now proceed to the $`g^4`$ contributions. From time independent perturbation theory we may write it as the sum of anti-screening and screening effects, $`V^{(4)}(r)=V_{\mathrm{anti}}^{(4)}(r)+V_{\mathrm{scr}}^{(4)}(r)`$, where
$`V_{\mathrm{anti}}^{(4)}(r)`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \mathrm{d}^2x0|\varphi ^a^2\varphi ^a|0},`$ (16)
$`V_{\mathrm{scr}}^{(4)}(r)`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{n0}{}}{\displaystyle \frac{1}{E_n}}{\displaystyle \mathrm{d}^2x0|\varphi ^a^2\varphi ^a|n\mathrm{d}^2xn|\varphi ^b^2\varphi ^b|0},`$ (17)
and $`E_n`$ is the energy of the state $`|n`$. In the second term it is sufficient, at this order, to sum over a complete set of intermediate states of two transverse gluons.
These two terms again represent the two distinct physical interactions that occur in QCD: the first is the non-abelian generalisation of the Coulombic interaction, while the second describes the exchange of physical, transverse gluons. This second term, since it comes from the exchange of physical quanta, represents the expected screening part of the potential which lowers the interaction energy.
Let us first verify that this anti-screening contribution agrees with our previous result. At this order we need to retain terms up to order $`g^3`$ in $`\varphi `$. From (12) we obtain three identical contributions
$$V_{\mathrm{anti}}^{(4)}(r)=\frac{3}{2}g^4f_{abc}f_{cde}\mathrm{d}^2x0|\rho ^a\frac{1}{^2}𝑨^b\mathbf{}\frac{1}{^2}𝑨^d\mathbf{}\frac{1}{^2}\rho ^e|0.$$
(18)
This result is precisely Eq. 5 as expected.
To find the screening contribution, we now need to insert physical two gluon states into (17). The sum over such states then becomes a sum over colour ($`e`$, $`f`$) and helicity ($`\lambda `$, $`\sigma `$) and an integral over momenta. Explicitly we have:
$$\underset{n=2\mathrm{gluon}}{}|nn|=\frac{1}{2}\underset{ef}{}\underset{\lambda \sigma }{}\mathrm{d}^2k\mathrm{d}^2la_e^{}(\lambda ,𝒌)a_f^{}(\sigma ,𝒍)|00|a_f(\sigma ,𝒍)a_e(\lambda ,𝒌).$$
(19)
The only terms from (12) that can contribute to these transverse states are given by $`\varphi ^a^2\varphi ^a2g^2f_{abc}\rho ^a^2𝑨^b𝑬_\mathrm{T}^c`$. We may thus write
$$V_{\mathrm{scr}}^{(4)}(r)=2g^4\underset{n=2\mathrm{gluon}}{}\frac{1}{E_n}\mathrm{d}^2xf_{abc}0|\rho _q^a\frac{1}{^2}𝑨^b𝑬_\mathrm{T}^c|n\mathrm{d}^2wf_{def}n|\rho _{\overline{q}}^d\frac{1}{^2}𝑨^e𝑬_\mathrm{T}^f|0,$$
(20)
and, in Coulomb gauge, we may identify the transverse electric field with $`\dot{A}_i`$. Using the standard commutator, $`[a_b(\lambda ,𝒌),a_c^{}(\sigma ,𝒍)]=\delta _{bc}\delta _{\lambda \sigma }\delta ^3(𝒌𝒍)`$ we then rapidly obtain
$`V_{\mathrm{scr}}^{(4)}(q)`$ $`=`$ $`{\displaystyle \frac{g^4}{4}}C_AC_F{\displaystyle \frac{1}{|𝒒|^4}}{\displaystyle \frac{\mathrm{d}^2l}{(2\pi )^2}\mathrm{d}^2k\frac{\delta ^3(𝒒𝒌𝒍)(|𝒍||𝒌|)^2}{|𝒍||𝒌|(|𝒍|+|𝒌|)}}`$ (21)
$`\times {\displaystyle \underset{\lambda }{}}ϵ^i(\lambda ,𝒌)ϵ^j(\lambda ,𝒌){\displaystyle \underset{\sigma }{}}ϵ^i(\sigma ,𝒍)ϵ^j(\sigma ,𝒍),`$
where we have already carried out the $`𝒙`$ integral and some trivial momentum integrals. We now exploit the Coulomb gauge relation
$$\underset{\lambda }{}ϵ^i(\lambda ,𝒌)ϵ^j(\lambda ,𝒌)=\delta ^{ij}\frac{k^ik^j}{𝒌^2},$$
(22)
to arrive at the final expression for the screening contribution
$$V_{\mathrm{scr}}^{(4)}(q)=\frac{g^4}{4}C_AC_F\frac{1}{|𝒒|^4}J(q),$$
(23)
where
$$J(q)=\frac{\mathrm{d}^2l}{(2\pi )^2}\frac{(|𝒍||𝒒𝒍|)^2}{|𝒍||𝒒𝒍|(|𝒍|+|𝒒𝒍|)}\left\{1\frac{𝒒^2}{(𝒒𝒍)^2}+\frac{(𝒒𝒍)^2}{𝒍^2(𝒒𝒍)^2}\right\},$$
(24)
The term $`|𝒍|+|𝒒𝒍|`$ in the denominator is unusual and makes this a more difficult integral<sup>4</sup><sup>4</sup>4We need to calculate the full, finite integral, while in $`3+1`$ dimensions, where we have the same integrand, we only need to extract the logarithmic divergence, so this denominator term effectively reduces to $`2|𝒍|`$ and the calculation is trivial..
To evaluate this integral it is convenient to go into polar co-ordinates, $`\rho ,\theta `$, and then make the change of variables: $`\rho =(\tau ^21)/[2(\tau \mathrm{cos}\theta )]`$. The angular integral may then be performed and afterwards it is not too difficult to evaluate the integral over $`\tau `$. The result is
$$J(q)=\left(\frac{7}{8}+\frac{3}{\pi }\right)|𝒒|.$$
(25)
We so obtain for the screening contribution to the interquark potential in $`2+1`$ dimensions
$$V_{\mathrm{scr}}^{(4)}(q)=g^4C_FC_A\frac{1}{4|𝒒|^3}\left(\frac{7}{8}\frac{3}{\pi }\right).$$
(26)
This, together with the anti-screening result (9), gives us the total order $`g^4`$ contribution to the interquark potential in $`2+1`$ dimensions:
$$V^{(4)}(q)=g^4C_FC_A\frac{1}{4|𝒒|^3}\left[\frac{3}{\pi }\left(\frac{3}{\pi }\frac{7}{8}\right)\right].$$
(27)
We see that, as expected, the sum of the dominant anti-screening contribution and this screening term gives exactly the correct result for the total potential (3). As had to be the case, the various factors of $`\pi `$ have combined to give one overall factor. This physical decomposition cannot be seen in the Wilson loop approach. There only the contributions from different classes of diagrams can be distinguished, however, they are gauge dependent (and all have the same $`\pi `$ factors).
The relative numerical weighting of the screening and anti-screening contributions to the potential is now $`8.37\%`$. This is remarkably within one part in a hundred of the split in $`3+1`$ dimensions!
There is a pressing need for a detailed understanding of the structure of the forces in QCD. In this paper we have calculated the screening and anti-screening contributions to the static inter-quark potential in $`2+1`$ dimensions:
$$V(r)=\frac{g^2C_F}{2\pi }\mathrm{ln}(g^2r)+\frac{g^4C_FC_A}{8\pi }\left[\frac{3}{\pi }\left(\frac{3}{\pi }\frac{7}{8}\right)\right]r.$$
(28)
This calculation is of interest in itself, given that the beta function now vanishes, in that it supports the idea that the $`2+1`$ theory models many of the important features of full QCD. We have seen that, to a first approximation, it is safe to neglect gluonic screening effects in $`2+1`$ dimensions. Additionally we note that it is important to understand the $`2+1`$ dimensional theory, since it is related by dimensional reduction to the high temperature limit of QCD (although this identification is not direct ). The physical decomposition that we have calculated exhibits a curious mathematical property (the differing transcendental factors) and the unexpected physical behaviour that the relative weights of screening and anti-screening are nearly identical in both $`2+1`$ and $`3+1`$ dimensions.
We note that the one loop anti-screening coefficient of the linearly rising term is only 1.35 times the lattice result in 2$`+`$1 dimension . It is not clear to us why there is such good agreement, nor how a linear potential emerges from the lattice when higher perturbative corrections will have the form of a power series in $`r`$.
We now want to extend this work in various directions. The separation into screening and anti-screening is not known at higher orders in the coupling. As noted by Drell , the method used in the latter part of this paper does not easily lend itself to such calculations. Our approach, based on a manifestly gauge invariant construction of quarks and gluons, can be readily extended to higher orders (see the appendix of ). Indeed we have previously shown in QED that such dressed fields have infra-red finite on-shell Green’s functions at all orders in perturbation theory. (The same decomposition of the dressing into a minimal and a separately gauge invariant part is also reflected in the infra-red structures of QED.) We are thus in the process of calculating the, hitherto unknown, decomposition of the potential into screening and anti-screening effects at order $`g^6`$ in both $`3+1`$ and $`2+1`$ dimensions. Another important extension of this work is to repeat the $`3+1`$ calculation at finite temperature. The results of this letter could be taken as indicating that the anti-screening/screening decomposition is insensitive to the temperature. If this is indeed the case, one needs to discover what aspect of strong interaction physics underlies this remarkable property.
Acknowledgements: This work was supported by the British Council/Spanish Education Ministry Acciones Integradas grant no. 1801/HB1997-0141. We thank S. Drell, Y. Schröder and S. Tanimura for correspondence, and Robin Horan and Ramon Muñoz-Tapia for discussions.
|
warning/0002/hep-th0002145.html
|
ar5iv
|
text
|
# Untitled Document
hep-th/0002145
DeSitter Entropy, Quantum Entanglement and AdS/CFT
Stephen Hawking, Juan Maldacena and Andrew Strominger
Department of Applied Mathematics and Theoretical Physics,
Centre for Mathematical Sciences
Wilberforce Road, Cambridge CB3 OWA, UK
Department of Physics
Harvard University
Cambridge, MA 02138, USA
Abstract
A deSitter brane-world bounding regions of anti-deSitter space has a macroscopic entropy given by one-quarter the area of the observer horizon. A proposed variant of the AdS/CFT correspondence gives a dual description of this cosmology as conformal field theory coupled to gravity in deSitter space. In the case of two-dimensional deSitter space this provides a microscopic derivation of the entropy, including the one-quarter, as quantum entanglement of the conformal field theory across the horizon.
1. Introduction
Despite advances in the understanding of black hole entropy, a satisfactory microscopic derivation of the entropy of deSitter space remains to be found. In this paper we address this issue in the context of a deSitter space arising as a brane-world of the type discussed by Randall and Sundrum <sup>1</sup> Unlike we include a nonzero cosmological constant on the brane. bounding two regions of anti-deSitter space. It is natural to suppose that such theories are dual, in the spirit of AdS/CFT , to a conformal theory on the brane-world coupled to gravity with a cutoff. The cutoff scales with the deSitter radius in such a way that the usual AdS/CFT correspondence is recovered when the cutoff is taken to infinity. This duality provides an alternate description of the deSitter cosmology which can be used, in the case of two dimensions, for a microscopic derivation of the deSitter entropy. We find that the entropy can be ascribed to the quantum entanglement of the CFT vacuum across the deSitter horizon. Quantum entanglement entropy can also be viewed as the entropy of the thermal Rindler particles near the horizon, thereby avoiding reference to the unobservable region behind the horizon.
Our derivation is closely related to the observation of reference that in two dimensions black hole entropy can be ascribed to quantum entanglement if Newton’s constant is wholly induced by quantum fluctuations of ordinary matter fields (see also \[5,,6,,7,,8\]). In the context of this seemed to be a rather artificial and unmotivated assumption. However the AdS brane-world scenarios do appear to have this feature. The basic reason is that, in a semiclassical expansion, the Einstein action on the brane arises mainly from bulk degrees of freedom<sup>2</sup> This follows when the AdS radius is large compared to the Planck length. which correspond, in the dual picture, to ordinary matter fields on the brane. The semiclassical expansion in the bulk corresponds to a large $`N`$ expansion in the brane, in which the leading term in Newton’s constant is induced by matter fields.
Our derivation of two-dimensional deSitter entropy is similar to the derivation of black hole entropy in in that it uses a brane field theory dual to the spacetime gravity theory to compute the entropy. However it differs in that in the black hole entropy was given by the logarithm of the number of unobserved microstates of the black hole, whereas here the deSitter entropy arises from entanglement with the unobserved states behind the horizon.<sup>3</sup> In general the entanglement entropy is less than the logarithm of the number of possible microstates of the unobserved sector of the Hilbert space. If the total system is in a pure state, the entanglement entropy is bounded from above by the logarithm of the number of possibly entangled states in the unobserved sector of the Hilbert space. Alternatively, it can be viewed as the number of microstates of the thermal gas of Rindler particles near the horizon. This latter viewpoint is closer to that of . This issue is explored in the final section by throwing a black hole in the bulk of AdS at the brane. When the bulk black hole reaches the brane, the brane state collapses to a brane black hole. At all stages the entropy is accounted for by a thermal gas on the brane.
Formally, the derivation can be generalized to higher-dimensional deSitter spaces which bound higher-dimensional anti-deSitter spaces. It was conjectured by Susskind and Uglum that there is a general a precise relationship between entanglement entropy and the one loop correction to Newton’s constant. Based on this, Jacobson argued that black hole entropy can be ascribed to quantum entanglement if Newton’s constant is wholly induced. However, while we are sympathetic to the conjecture of , and it fits well with the discussion herein, its status remains unclear \[10,,11\]. The basic problem is that in greater than two dimensions the corrections have power law divergences and hence are regulator dependent. This makes precise statements difficult above two dimensions.
A further significant fly in the ointment - even in two dimensions- is that there is no known example of the type of brane-world scenario considered in embedded in a fully consistent manner into string theory <sup>4</sup> See however for related scenarios in string theory.. The observations of the present paper are relevant only if such examples exist. For the time being however they provide intriguing connections along the circle of ideas pursued in \[1--9\].
2. Classical Geometry
Fig. 1: Euclidean instanton geometry. The brane is an $`S^2`$ which bounds two patches of euclidean $`AdS_3=H^3`$.
The euclidean action for a onebrane coupled to gravity with a negative cosmological constant ($`\mathrm{\Lambda }=\frac{1}{L^2}`$) is
$$S_{tot}=\frac{M_p}{16\pi }d^3x\sqrt{g}(R+\frac{2}{L^2})+Td^2\sigma \sqrt{h}.$$
$`M_p`$ here is the three dimensional Planck mass, $`T`$ is the brane tension and $`h`$ is the induced metric on the brane. We have assumed that there is no boundary. We wish to consider a spherically symmetric brane at radius $`r_B`$ which bounds two regions of $`AdS_3`$ with metrics
$$ds_3^2=L^2dr^2+L^2\mathrm{sinh}^2rd\mathrm{\Omega }_2^2,$$
where $`0rr_B`$, as shown in fig. 1. The two copies of $`AdS_3`$ are glued together along the $`S^2`$ at $`r=r_B`$ where the bulk curvature has a delta function. The topology of the spacetime is $`S^3`$. The induced metric on the brane is
$$ds_2^2=\mathrm{}^2d\mathrm{\Omega }_2^2,$$
with
$$\mathrm{}L\mathrm{sinh}r_B.$$
The action for such a configuration is
$$S_{tot}=\frac{M_p}{4\pi L^2}V_3\frac{M_p\mathrm{coth}r_B}{2\pi L}V_2+TV_2,$$
where $`V_3`$ is the bulk volume and $`V_2`$ is the brane volume. The second term arises from a delta function in the bulk curvature at $`r=r_B`$. One finds using (2.1) that
$$S_{tot}=\frac{LM_p}{2}(\mathrm{sinh}2r_B+2r_B)+4\pi TL^2\mathrm{sinh}^2r_B.$$
The action (2.1) has an extremum at
$$\mathrm{tanh}r_B=\frac{M_p}{4\pi TL}$$
for which
$$S_{tot}=LM_pr_B.$$
We are interested in the case that the right hand side of (2.1) is close to (but less than) one so that $`r_B`$ and $`\mathrm{}`$ are large. We may then approximate
$$S_{tot}=LM_p\mathrm{ln}\mathrm{}+\mathrm{}$$
The subleading corrections are suppressed for large $`\mathrm{}`$.
The induced brane metric (2.1) is the two-dimensional euclidean deSitter ($`i.e.`$ round $`S^2`$) metric with a large deSitter radius $`\mathrm{}`$. A lorentzian deSitter solution can be obtained by analytic continuation of the periodic angle $`\varphi `$ on $`S^2`$ to $`it`$. One finds
$$\begin{array}{cc}\hfill ds_3^2& =L^2dr^2+L^2\mathrm{sinh}^2r(d\theta ^2sin^2\theta dt^2),\hfill \\ \hfill ds_2^2& =\mathrm{}^2d\theta ^2\mathrm{}^2sin^2\theta dt^2.\hfill \end{array}$$
Fig. 2: Penrose diagram of Lorentzian two dimensional de-Sitter space. The dotted line indicates the trajectory of a geodesic observer. We have also indicated the past and future horizons for that observer, and the shaded region indicates the patch covered by the coordinates (2.1).
These coordinates cover the diamond-shaped region of $`DS_2`$ illustrated in fig. 2. It is the region outside both the future and past horizons of any timelike observer at constant $`\theta 0,\pi `$.
3. Dual Representations
Let us assume there is a unitary quantum theory whose semiclassical gravitational dynamics is described by (2.1). Such a theory should have two dual descriptions.<sup>5</sup> Related discussions appear in \[13,,14\]. The first is, as described above, a three-dimensional bulk theory containing gravity and a brane.
The second description is as a two-dimensional effective theory of the light fields on the brane worldvolume. These light fields include holographic matter living on the brane. To see this we first consider a single copy of $`AdS_3`$ in coordinates (2.1) with a fixed boundary at $`r=r_B`$, for large $`r_B`$. If we keep the metric on the boundary fixed, but integrate over the bulk metric, the resulting theory has a holographic description as a $`1+1`$ conformal field theory on a sphere of radius $`\mathrm{}`$ with central charge $`c=\frac{3LM_p}{2}`$ and a cutoff at $`L`$\[3,,16,,17\].<sup>6</sup> Alternately one may take a sphere of unit radius and a cutoff at $`L/\mathrm{}`$. To recover the geometry under consideration, we must take two copies of such bounded $`AdS_3`$ spacetimes, identify them along their boundaries, and then integrate over boundary metrics. Because there are two copies, one has two copies of the matter action on the boundary, with a total central charge
$$c=3LM_p.$$
The second brane tension term in the action (2.1) corresponds to a counterterm which renormalizes the cosmological constant.
We are not fixing the boundary by hand so we have two-dimensional gravity because, as shown in , there is a graviton zero mode trapped on the brane, as well as the radion field representing the radial position of the brane. The radion has a small mass in the case of a large deSitter boundary. For a $`D`$-dimensional boundary brane, gravity plus the radion comprise $`\frac{1}{2}(D^23D+2)`$ local degrees of freedom (after implementing the constraints). Hence for the present case of $`D=2`$ the radion-gravity system has no local degrees of freedom.
For our purposes we need only the gravity part of the effective action, with the radion field set at the minimum of its potential. The gravity effective action is most easily represented in conformal gauge
$$ds_2^2=e^{2\rho }d\widehat{s}_2^2,$$
where $`d\widehat{s}^2`$ is the unit metric on $`S^2`$ obeying
$$\widehat{R}_{z\overline{z}}=\widehat{g}_{z\overline{z}},d^2z\widehat{g}_{z\overline{z}}=4\pi $$
in complex coordinates. One then finds<sup>7</sup> This is equivalent to the computation in \[18,,19\].
$$S_g=\frac{LM_p}{4\pi }d^2z\left(_z\rho _{\overline{z}}\rho +\widehat{R}_{z\overline{z}}\rho \frac{1}{2\mathrm{}^2}\widehat{g}_{z\overline{z}}e^{2\rho }\right).$$
The equations of motion for constant fields give
$$\rho =\mathrm{ln}\mathrm{}.$$
The action (3.1) evaluated at this solution agrees with (2.1) We note also that the total gravity plus matter central charge vanishes, as required for general covariance. These considerations determine (3.1).
4. DeSitter Entropy
In this section we give macroscopic and microscopic derivations of the entropy.
4.1. Semiclassical Macroscopic Entropy
The macroscopic entropy can be computed directly in three dimensions from the area entropy-law
$$S_{dS}=\frac{Area}{4G_3}.$$
In this expression $`G_3=1/M_p`$ and the horizon area is the area of the fixed point of a $`U(1)`$ isometry of the instanton geometry (2.1). This consists of a geodesic circle intersecting the $`S^2`$ brane at the north and south poles. The area (length) of this circle is $`4Lr_B4L\mathrm{ln}\mathrm{}`$. Hence we obtain
$$S_{dS}=LM_p\mathrm{ln}\mathrm{}.$$
An alternate derivation can be given from the two dimensional deSitter space using
$$S_{dS}=\frac{Area}{4G_2}.$$
The area in this formula is just the area of the observer horizon ($`\theta =0,\pi `$ in (2.1) ) which consists of two points and is therefore equal to 2. $`G_2`$ is determined as the (field-dependent) coefficient of the the scalar curvature $`R=\frac{1}{2}g^{z\overline{z}}R_{z\overline{z}}`$. From (3.1) this is
$$\frac{1}{G_2}=2LM_p\rho =2LM_p\mathrm{ln}\mathrm{}.$$
Inserting (4.1) into (4.1) reproduces (4.1).
4.2. Microscopic Entropy
Let us now consider the entropy from the point of view of the brane CFT with $`c=3M_pL`$. An $`SO(2,1)`$ invariant vacuum for quantum field theory in lorentzian deSitter space $`|0`$ can be defined as the state annihilated by positive frequency modes in the metric
$$ds_2^2=\mathrm{}^2\frac{dt^2+dx^2}{cos^2t}.$$
The proper time $`\tau `$ of an observer moving along a geodesic at $`x=0`$ is related to the time $`t`$ in (4.1) by time $`e^{\tau /\mathrm{}}=\mathrm{tan}(\frac{t}{2}+\frac{\pi }{4})`$. Green functions in this vacuum are single valued functions of $`t`$. Therefore they are periodic in imaginary $`\tau `$ with period $`2\pi i\mathrm{}`$, and the observer accordingly detects a thermal bath of particles with temperature $`\frac{1}{2\pi \mathrm{}}`$.
The vacuum $`|0`$ is a pure state of this CFT. However a single observer can probe features of this state only within the observer horizons, i.e. in the diamond region covered by the coordinates (2.1). The results of all such measurements are described by an observable density matrix $`\rho _{\mathrm{obs}}`$. $`\rho _{\mathrm{obs}}`$ is constructed from the pure density matrix $`|00|`$ by tracing over the unobservable sector of the Hilbert space supported behind the horizon. The entropy
$$S_{ent}=tr\rho _{\mathrm{obs}}\mathrm{ln}\rho _{\mathrm{obs}}$$
is nonzero because of correlations between the quantum states inside and outside of the horizon. $`S_{ent}`$ is called the entanglement entropy because it measures the extent to which the observable and unobservable Hilbert spaces are entangled. Note that the entanglement entropy, defined this way, agrees with the entropy of the gas of particles at the local Rindler temperature. A general formula for $`S_{ent}`$ was derived in \[21,,4\]:
$$S_{ent}=\frac{c}{6}\rho |_{boundary}\frac{c\mathrm{\Delta }}{6},$$
where $`\mathrm{\Delta }`$ is the short distance cutoff and $`\rho `$ is the conformal factor of the metric in the coordinates (in our case (4.1) ) used to define the vacuum evaluated at the boundary (consisting of two points) of the unobserved region. From (4.1) we see that the boundary is at $`t=0`$, so $`\rho =\mathrm{ln}\mathrm{}`$. Putting this all together and using $`c=3M_pL`$ we get
$$S_{ent}=LM_p\mathrm{ln}\mathrm{},$$
in agreement with (2.1).
This result is a generalization to the two-dimensional deSitter case of the observation of that, in a two-dimensional theory in which the entirety of Newton’s constant is induced from matter, the Bekenstein-Hawking black hole entropy can be microscopically derived as entanglement entropy. The missing ingredient in both of these previous discussions was a motivation for the assumption that Newton’s constant is induced. Here we see it is natural - or at least equivalent to other assumptions - in the brane-world context.
5. Four Dimensions
We can also consider a four dimensional brane world model. We have a four dimensional brane bounding two $`AdS_5`$ regions. If we consider perturbations of the four dimensional metric we can analyze the system by first finding a five dimensional solution which has a given four dimensional metric at the brane. The solution will look like
$$ds^2=L^2\left(\frac{g_{\mu \nu }(z,x)dx^\mu dx^\nu +dz^2}{z^2}\right)+\mathrm{}zϵ$$
where $`g_{\mu \nu }(z=ϵ,x)=g_{4\mu \nu }`$ is the four dimensional metric. We can then insert the solution with a given four dimensional metric back into the action and get an effective action for the four dimensional metric. This effective action will contain a leading term going like $`1/ϵ^4`$ which will be canceled by the brane tension so that the next nontrivial term will be
$$S(g_4)=\frac{2}{16\pi G_5}_{zϵ}d^5x\sqrt{g_5}R_5=\frac{L^3}{16\pi G_5ϵ^2}d^4x\sqrt{g_4}R_4,$$
where to this order of approximation we can assume that the five dimensional metric is independent of $`z`$ (the $`z`$ dependent parts give terms going like lower powers of $`1/ϵ`$) and we took into account the two copies of $`AdS_5`$. This of course the way that the four dimensional Newton’s constant is computed in .<sup>8</sup> In $`ϵ`$ does not appear since the four dimensional metric is rescaled by $`\frac{L^2}{ϵ^2}`$. We have phrased the calculation in this way to make connection with AdS/CFT \[3,,16,,17\], so that the integral over five dimensional metrics with the boundary metric at $`z=ϵ`$ held fixed can be interpreted as a field theory with a cutoff $`ϵ`$ on that particular four dimensional space. So the physics of is the same as the physics of a conformal field theory with a cutoff coupled to four dimensional gravity (as considered in ), where the four dimensional conformal field theory as an $`AdS`$ dual, (see also \[13,,23\]). Notice that the four dimensional Newton constant can be written as
$$\frac{1}{G_4}=\frac{8N_{dof}}{\pi ϵ^2},N_{dof}\frac{\pi L^3}{8G_5}$$
where $`N_{dof}`$ is the quantity that appears in all AdS/CFT calculations involving the stress tensor, calculations such as the two point function of the stress tensor or the free energy at finite temperature, etc. It can be viewed as the effective number of degrees of freedom of the CFT, ($`N_{dof}=N^2/4`$ for $`𝒩=4`$ SYM). This form for the four dimensional Newton constant is very suggestive. It is of the general form expected for induced gravity in four dimensions. If we start in four dimensions with a theory with infinite or very large Newton constant and we integrate out the matter fields we expect to get a four dimensional value for the Newton constant which is rougly as in (5.1)\[24,,25,,26,,6,,27,,10,,28,,29\]. The precise value that we would get seems to depend on the cutoff procedure. Indeed, if we use heat kernel regularization we would get that for $`N=4`$ Yang Mills this cuadratic divergence cancels. The gravity procedure of fixing the boundary at some finite distance must correspond to a suitable cutoff for the field theory and it is not obvious that we should get the same results for divergent terms. Indeed, the supergravity regularization procedure would also give a divergent value for the vacuum energy (which is being cancelled by the brane). Again in theories where the four dimensional Newton constant is induced one can interpret black hole entropy as entanglement entropy . If we consider a four dimensional metric with a horizon, like a black hole or de-Sitter space we indeed find that the entropy is given by
$$S=\frac{A_4}{4G_4}=\frac{2N_{dof}A_4}{ϵ^2\pi }$$
where we just used the relation of the 4d Newton constant and the four dimensional parameters. The right hand side can be interpreted as entanglement entropy. In other words, we can compute the entanglement entropy in the field theory as entropy of the gas of particles in thermal Rindler space and we would obtain precisely the right hand side of (5.1). We can do the entropy calculation at weak coupling in weakly coupled $`𝒩=4`$ SYM and we would obtain agreement up to a numerical factor, which could be be related to the ignorance of the cutoff procedure, but more fundamentally can also be related to strong coupling effects like the $`3/4`$ appearing in the relation between the weakly coupled and the strongly coupled expressions for the free energy.
6. Black Hole Formation on the Brane
The entropy of a bulk black hole in the interior of AdS can be accounted for by representing it as a thermal state in the brane theory on its boundary. At first this may seem to be at odds with the accounting given here of the entropy of a black hole on the brane in terms of quantum entanglement. In this section we will attempt to reconcile the accounts by throwing a bulk black hole at the brane and watching it turn into a brane black hole.<sup>9</sup> Similar ideas are being pursued by H. Verlinde.
Consider a bulk black hole at the origin of AdS at temperature $`T_H`$ whose size is large compared to the AdS radius $`L`$ but small compared to the brane radius $`\mathrm{}`$. This has a stable ground state in which there is a cloud of thermal radiation surrounding the black hole. In the brane theory, this is represented as homogeneous thermal state at temperature $`T_H`$. The statistical entropy of this state agrees with Bekenstein-Hawking entropy of the black hole.<sup>10</sup> The equality is precise for AdS<sub>3</sub>. In higher dimensional cases such as AdS<sub>5</sub> it follows if one accepts the factor of $`4/3`$ as a feature of strongly coupled gauge theory, which we shall for the purposes of this discussion.
The center of mass of the black hole can be given momentum by the action of an AdS<sub>D+1</sub> $`SO(D,2)`$ isometry. These isometries are broken by the presence of the brane, but if black hole is not too near the brane this should not matter. The $`SO(D,2)`$ action will impart momenta to the black hole and make it oscillate about the origin. The brane version of such a state can be found by applying an $`SO(D,2)`$ conformal transformation to the thermal brane state. The resulting state will carry conformal charges and have energy densities with bipolar oscillations. The statistical entropy of this oscillating state will of course still agree with Bekenstein-Hawking entropy of the oscillating black hole.
In the above discussion we implicitly assumed that the field theory was defined on the cylinder ($`S^3\times R`$). When we think of the field theory defined in flat space or de-Sitter space we are looking only at some coordinate patch of the AdS cylinder. In this case we only see half a period of oscillation which can be interpreted as a gas of particles in the field theory that contracts and expands again.
If the oscillation is made large enough, the black hole actually reaches the brane where it will stick, at least temporarily. The brane picture of this process is that the oscillations in the energy density have become so large that the thermal radiation collapses to from a black hole.<sup>11</sup> Note that the coupling to gravity breaks conformal invariance so the conformal charges are not conserved when the collapse occurs.
Before collapse, the entropy is accounted for on the brane as the entropy of thermal radiation. After collapse, it is accounted for by the thermal gas of Rindler particles near the horizon. (In general the black hole formation is not adiabatic.) This latter entropy is localized within a distance of the order of the cutoff from the horizon. This could be described by saying that all stages the entropy is stored in thermal radiation, and this radiation hovers outside the horizon when the black hole is formed. In this description the statistical origin of the entropy of bulk and brane black holes appears to be similar. Eventually the black hole will evaporate and the final state will be just outgoing thermal radiation on the brane theory.
It is interesting that there is a “correspondence principle” in the sense that when the AdS black hole has a radius of the order of the five dimensional anti-de-Sitter space and it makes a grazing collision with the brane, then the entropies calculated as a thermal gas and as entanglement entropy are the same up to a numerical constant. Such a black hole would have a Schwarschild radius in the boundary theory equal to the field theory cutoff $`ϵ`$.
In closing, it remains to find a fully consistent quantum realization of such a brane-world scenario to which our observations can be applied. Alternately, perhaps it is applicable in a more general setting. One of the important lessons of string duality is that something which is classical from one point of view can be quantum from another. What is needed here is a point of view from which Newton’s constant - usually regarded as a largely classical quantity - is a purely quantum effect. It is notable in this regard that closed string poles arise as a loop effect in open string theory, indicating that there might be a way in which the full closed string dynamics is contained in open string field theory. In that case Newton’s constant could be induced.
Acknowledgements
This work was supported in part by DOE grant DE-FG02-91ER40654. J.M. was also supported in part by the Sloan and Packard fellowships. We would like to thank T. Jacobson, A. Tseytlin and H. Verlinde for discussions.
References
relax G. W. Gibbons and S. W. Hawking, Action integrals and partition functions in quantum gravity, Phys. Rev. D15 (1977) 2752. relax L. Randall and R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064. relax J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200. relax T. Fiola, J. Preskill, A. Strominger and S. P. Trivedi, Black hole thermodynamics and information loss in two-dimensions, Phys. Rev. D50 (1994) 3987, hep-th/9403137. relax T. Jacobson, Black hole entropy and induced gravity, gr-qc/9404039. relax L. Susskind and J. Uglum, Black hole entropy in canonical quantum gravity and superstring theory, Phys. Rev. D50 (1994) 2700, hep-th/9401070. relax V. Frolov, D. Fursaev and A. I. Zelnikov, Statistical origin of black hole entropy in induced gravity, Nucl. Phys. B486 (1997) 339, hep-th/9607104. relax V. Frolov and D. Fursaev, Mechanism of generation of black hole entropy in Sakharov’s induced gravity, Phys. Rev. D56 (1997) 2212, hep-th/9703178. relax A. Strominger and C. Vafa, Microscopic origin of the Beckenstein-Hawking entropy, Phys. Lett. B379 (1996) 99, hep-th/9601029. relax D. Kabat, Black hole entropy and entropy of entanglement, Nucl. Phys. B453 (1995), hep-th/9503016. relax T. Jacobson, On the nature of black hole entropy, gr-qc/9908031. relax H. Verlinde, Holography and compactification, hep-th/9906182. relax S. S. Gubser, AdS/CFT and gravity, hep-th/9912001. relax J. DeBoer, E. Verlinde. and H. Verlinde, On the holographic renormalization group, hep-th/9912012. relax J. D. Brown and M. Henneaux, Central charges in the canonical realization of asymptotic symmetries: An example from three dimensional gravity, Commun. Math. Phys. (1986), 104. relax L. Susskind and E. Witten, The holographic bound in Anti-de-Sitter space, hep-th/9805114. relax A. Peet and J. Polchinski, UV/IR relations in AdS dynamics, Phys. Rev. D59 (1999) 065011, hep-th/9809022. relax M. Henningson and K. Skenderis, JHEP 9807 (1998) 023, hep-th/9806087. relax V. Balasubramanian and P. Kraus, A stress tensor for anti de Sitter gravity, Commun. Math. Phys. (1999) 208, hep-th/9902121. relax S. Hawking and C. Hunter, Gravitational entropy and global structure, Phys. Rev. D59 (1999)044025, hep-th/9808085. relax C. Holzhey, F. Larsen and F. Wilczek, Geometric and renormalized entropy in conformal field theory, Nucl. Phys. B424 (1994) 443, hep-th/9403108. relax P. Frampton and C. Vafa, Conformal approach to particle phenomenology, hep-th/9903226. relax E. Witten, New dimensions in field theory and string theory, http://www.itp.ucsb.edu/online/susyc99/discussion. relax L. Bombelli, R. Koul, J. Lee and R. Sorkin, A quantum source of entropy for black holes, Phys. Rev. D34 (1986) 373. relax M. Srednicki, Phys. Rev. Lett. 71 (1993) 666, Entropy and area, hep-th/9303048. relax S. Carlip and C. Teitelboim, The off shell black hole gr-qc/9312002. relax C. Callan and F. Wilczek, On geometric entropy, Phys. Lett. B333 (1994) 55; hep-th/9401072. relax J-G. Demers, R. LaFrance and R. Myers, Black hole entropy and renormalization, gr-qc/9507042. relax F. Larsen and F. Wilczek, Renormalization of black hole entropy and of the gravitational coupling constant, Nucl. Phys. B58 (1996) 249, hep-th/9506066.
|
warning/0002/hep-ph0002062.html
|
ar5iv
|
text
|
# Theorems on the Renormalization Group Evolution of Quark Yukawa Couplings and CKM Matrix
## I Introduction
The problem of the Renormalization Group EVolution (RGEV) of the Quark Yukawa Couplings (QYC) and the Cabibbo-Kobayashi-Maskawa matrix (CKM) has been studied in many earlier papers, Refs. . Despite the fact that many, both analytical end numerical results have been obtained on the evolution of the QYC and CKM matrix there is no single reference where some general properties of the evolution have been discussed. This paper is aimed to fill this gap and it discusses the general properties of the evolution of the CKM matrix and quark masses that follow from the hierarchical structure of the QYC.
The information on the properties of the QYC come from the experimental values of the quark masses and the CKM matrix. Both the quark masses and the CKM matrix show the hierarchical structure with the parameter $`\lambda `$$`\mathrm{sin}\theta _C`$$`0.22.`$ This hierarchical structure must also be present in the QYC and is used to derive some conditions for the unknown QYC, Refs. which reduce the number of parameters of the standard model.
In this paper we systematically investigate the influence of the hierarchical structure on the evolution of the CKM matrix and quark masses by constructing the exact solution of the one loop RGEV equations compatible with the observed hierarchy and then considering the two loop RGEV equations with the next order corrections in $`\lambda .`$
The most important result that we derive is that the CKM evolution depends only on one parameter which is a suitable integral that depends on the model (Standard Model (SM), Minimal Supersymmetric Standard Model (MSSM) and Double Higgs Model (DHM)). The corrections to the one loop evolution of the CKM matrix are of the relative order $`\lambda ^5.`$ We next show that the evolution of the ratios of the down quark masses $`m_d/m_s`$ and $`m_s/m_b`$ depends on the same parameter as the CKM matrix. The evolution of the up quark masses is different. Here the quark masses $`m_u`$, $`m_c`$ depend linearly on the corresponding initial values and their ratio $`m_u/m_c`$ is constant while the dependence for $`m_t`$ is non linear.
The organization of the paper is the following. In Section II we introduce the notation and the essential experimental facts. Afterwards we discuss the RGEV equations up to two loops and their approximate form compatible with the observed hierarchy. This gives the basis for the derivation in Section III of the exact solutions of the one loop RGEV equations. We also present there the perturbative scheme for the study of the corrections coming from the higher order terms in $`\lambda `$ and the two loop contributions. In Section IV we discuss the properties of the solutions derived earlier and their physical importance. In Section V we derive the approximate evolution equations for the squares of the absolute values of the CKM matrix elements and give their explicit solution. In such a way we give the explicit form of the one loop renormalization group evolution of the full CKM matrix. Section VI is devoted to the conclusions.
## II Quark Yukawa Couplings and CKM Matrix–Hierarchy and Evolution
The quark and lepton masses in the standard model arise through the Higgs mechanism from the Yukawa couplings which have the following structure in the SM
$$\underset{i,j=1}{\overset{3}{}}(f_{ij}^{(e)}\overline{e}_L^i\varphi e_R^j+(y_u)_{ij}\overline{u}_L^i\stackrel{~}{\varphi }u_R^j+(y_d)_{ij}\overline{d}_L^i\varphi d_R^j+\text{h.c.}).$$
(1)
Here $`f^{(e)}`$, $`y_u`$, and $`y_d`$ are the matrices of the Yukawa couplings of leptons and up and down quarks, respectively. $`\varphi `$ is the scalar Higgs field. The theory itself does not put any restrictions or conditions on the $`f^{(e)}`$, $`y_u`$ and $`y_d`$ matrices and from the phenomenological point of view they are constrained by the values of the physical lepton and quark masses
$$\text{Diag}(m_u,m_c,m_t)=(U_u)_L^{}y_u(U_u)_R^{},\text{Diag}(m_d,m_s,m_b)=(U_d)_L^{}y_d(U_d)_R^{}$$
(2)
with diagonal elements being the up and down quark masses after the spontaneous symmetry breaking and the CKM matrix. The diagonalizing matrices $`(U_{u,d})_{L,R}^{}`$ of the biunitary transformations transform the quark fields in Eq. (1) into the physical quark fields. As a consequence the unitary matrix (CKM matrix)
$$V_{CKM}=(U_u)_L^{}(U_d)_L^{}$$
(3)
appears in the charged current.
The observed hierarchy of the quark and lepton masses and the CKM matrix is the following, Refs.
$$\frac{m_u}{m_c}\lambda ^4,\frac{m_c}{m_t}\lambda ^4,\frac{m_d}{m_s}\lambda ^2,\frac{m_s}{m_b}\lambda ^2,\frac{m_b}{m_t}\lambda ^2,\frac{m_e}{m_\mu }\lambda ^4,\frac{m_\mu }{m_\tau }\lambda ^2,\frac{m_\tau }{m_t}\lambda ^3,$$
(4)
$$\left(\begin{array}{ccc}1\frac{1}{2}\lambda ^2& \lambda & \lambda ^3A(\rho i\eta )\\ \lambda & 1\frac{1}{2}\lambda ^2& \lambda ^2A\\ \lambda ^3A(1\rho i\eta )& \lambda ^2A& 1\end{array}\right)$$
(5)
with $`\lambda 0.22.`$
The hypothesis of grand unification assumes that the 3 coupling constants of the standard model converge at the scale $`10^{1516}`$ GeV and the symmetry group of the model becomes larger (e.g. SU(5)). This may also be an origin of the additional symmetries or textures of the QYC at this energy scale. To relate the parameters of the QYC at the GU scale with the observables at low energy one conventionally uses the Renormalization Group Equations (RGE) for the coupling constants and the quark and lepton Yukawa couplings. The structure of the two loop RGE is the following Refs.
$$\frac{dg_l}{dt}=\frac{1}{(4\pi )^2}b_lg_l^3\frac{1}{(4\pi )^4}G_lg_l^3,$$
(7)
$$\frac{dy_{u,d,e,\nu }}{dt}=\left[\frac{1}{(4\pi )^2}\beta _{u,d,e,\nu }^{(1)}+\frac{1}{(4\pi )^4}\beta _{u,d,e,\nu }^{(2)}\right]y_{u,d,e,\nu }.$$
(8)
The variable $`t`$ is defined as $`t=\mathrm{ln}(E/\mu )`$ and the constants $`b_l`$ and functions $`G_l`$, $`\beta _{u,d,e,\nu }^{(1)}`$ and $`\beta _{u,d,e,\nu }^{(2)}`$ are defined for various models in Appendix I. The coefficients $`G_l,`$ $`\beta _{u,d,e,\nu }^{(1)},`$ $`\beta _{u,d,e,\nu }^{(2)}`$ are functions of $`g_l`$ and $`y_{u,d,e,\nu }`$ so Eqs. (8) form a system of coupled non linear equations and their explicit solution is not known. The Yukawa couplings $`y_{u,d,e,\nu }`$ are normalized in the following way
$$[y_u]_{33}1,[y_d]_{33}\frac{m_b}{m_t}\lambda ^2,[y_e]_{33}\frac{m_\tau }{m_t}\lambda ^3.$$
(9)
The normalization in Eq. (9) is the origin of the hierarchy in Eqs. (8) because the functions $`G_l,`$ $`\beta _{u,d,e,\nu }^{(1)}`$ and $`\beta _{u,d,e,\nu }^{(2)}`$ contain the squares and higher powers of $`y_{u,d,e,\nu }.`$
From Eq. (9) it follows that positive powers of $`y_{d,e,\nu }`$ are much smaller than the corresponding powers of $`y_u`$. This information enables to find the solution of Eqs. (8) and to study the properties of the solutions. The other origin of the hierarchy in Eqs. (8) is related to the number of loops. Here the inclusion of each additional loop is suppressed by the factor $`1/(4\pi )^2\lambda ^4`$ which is of the same order as $`([y_d]_{33})^2`$ and the correct approximation procedure must take this into account.
Eqs. (8) will be solved and analyzed in the following steps:
1. Definition of the hierarchy of the equations;
2. *Exact* solution of the basic equations;
3. Perturbative corrections to the exact solution;
4. Implications of the former results for the quark masses and the CKM matrix
In the rest of the paper we will discuss each step in detail.
Eqs. (8) are the two loop RGE for coupling constants $`g_l`$ and Yukawa couplings $`y_{u,d,e,\nu }.`$ In these equations the terms of different order in $`\lambda `$ are present. As the first step we will find the approximate form of these equations neglecting all the terms of $`\lambda ^4`$ and higher. These approximate equations have the following form
$$\frac{dg_l}{dt}=\frac{1}{(4\pi )^2}b_lg_l^3,$$
(11)
$$\frac{dy_u}{dt}=\frac{1}{(4\pi )^2}[\alpha _1^u(t)+\alpha _2^uy_u^{}y_u^{}+\alpha _3^u\text{Tr}(y_u^{}y_u^{})]y_u=\frac{1}{(4\pi )^2}E_1^uy_u,$$
(12)
$$\frac{dy_d}{dt}=\frac{1}{(4\pi )^2}[\alpha _1^d(t)+\alpha _2^dy_u^{}y_u^{}+\alpha _3^d\text{Tr}(y_u^{}y_u^{})]y_d=\frac{1}{(4\pi )^2}E_1^dy_d,$$
(13)
$$\frac{dy_e}{dt}=\frac{1}{(4\pi )^2}[\alpha _1^e(t)+\alpha _2^ey_u^{}y_u^{}+\alpha _3^e\text{Tr}(y_u^{}y_u^{})]y_e=\frac{1}{(4\pi )^2}E_1^ey_e,$$
(14)
$$\frac{dy_\nu }{dt}=\frac{1}{(4\pi )^2}[\alpha _1^\nu (t)+\alpha _2^\nu y_u^{}y_u^{}+\alpha _3^\nu \text{Tr}(y_u^{}y_u^{})]y_\nu =\frac{1}{(4\pi )^2}E_1^\nu y_\nu $$
(15)
where the values of $`\alpha _i^{u,d,e,\nu }`$ are given in Appendix. Eqs. (15) will be solved exactly and as a next step we will consider Eqs. (8) where the terms of the order $`\lambda ^4`$ are kept:
$$\frac{dg_l}{dt}=\frac{1}{(4\pi )^2}b_lg_l^3\frac{g_l^3}{(4\pi )^4}\left(C_{lu}\text{Tr}(y_uy_u^{})+\underset{k}{}b_{kl}g_k^2\right),$$
(17)
$`{\displaystyle \frac{dy_u}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}E_1^uy_u+{\displaystyle \frac{1}{(4\pi )^2}}[\alpha _4^uy_dy_d^{}+\alpha _5^u\text{Tr}(y_dy_d^{})]y_u`$ (18)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^4}}F^u(y_uy_u^{},g_l)y_u={\displaystyle \frac{1}{(4\pi )^2}}E_1^uy_u+{\displaystyle \frac{1}{(4\pi )^4}}E_2^uy_u,`$ (19)
$`{\displaystyle \frac{dy_d}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}E_1^dy_d+{\displaystyle \frac{1}{(4\pi )^2}}[\alpha _4^dy_dy_d^{}+\alpha _5^d\text{Tr}(y_dy_d^{})]y_d`$ (20)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^4}}F^d(y_uy_u^{},g_l)y_d={\displaystyle \frac{1}{(4\pi )^2}}E_1^dy_d+{\displaystyle \frac{1}{(4\pi )^4}}E_2^dy_d,`$ (21)
$`{\displaystyle \frac{dy_e}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}E_1^ey_e+{\displaystyle \frac{1}{(4\pi )^2}}[\alpha _4^ey_dy_d^{}+\alpha _5^e\text{Tr}(y_dy_d^{})]y_e`$ (22)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^4}}F^e(y_uy_u^{},g_l)y_e={\displaystyle \frac{1}{(4\pi )^2}}E_1^ey_e+{\displaystyle \frac{1}{(4\pi )^4}}E_2^ey_e,`$ (23)
$`{\displaystyle \frac{dy_\nu }{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}E_1^\nu y_\nu +{\displaystyle \frac{1}{(4\pi )^2}}[\alpha _4^\nu y_dy_d^{}+\alpha _5^\nu \text{Tr}(y_dy_d^{})]y_\nu `$ (24)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^4}}F^\nu (y_uy_u^{},g_l)y_\nu ={\displaystyle \frac{1}{(4\pi )^2}}E_1^\nu y_\nu +{\displaystyle \frac{1}{(4\pi )^4}}E_2^\nu y_\nu .`$ (25)
The functions $`F^{u,d,e,\nu }(y_uy_u^{},g_l)`$ are obtained from the functions $`\beta _{u,d,e,\nu }^{(2)}`$ given in Appendix by putting $`y_d=y_e=0.`$ Eqs. (II) cannot be explicitly solved but they allow the perturbative solution by transforming them into integral equations. This allows the study of the corrections to the exact solutions of Eqs. (15).
## III Solution of the Renormalization Group Equations for $`y_u`$ and $`y_d`$
The renormalization group equations (8) will be solved in two steps. We will start by solving the one loop equations (15). Eqs. (11) are very easy to solve and their solution reads
$$g_l(t)=\frac{g_l(t_0)}{\sqrt{1\frac{2b_lg_l^2(t_0)(tt_0)}{(4\pi )^2}}}$$
(26)
Eq. (12) is decoupled from Eqs. (13) and (14) and is non linear. Eqs. (13) and (14) become linear once $`y_u`$– the solution of Eq. (12) is known. We therefore solve Eq. (12) first.
### A The evolution of $`y_u`$ from one loop RGE
The crucial fact enabling the exact solution of Eq. (12) is the observation that the diagonalizing matrices of the biunitary transformation do not depend on the energy. As mentioned before $`y_u`$ can be diagonalized by the biunitary transformation, Eq. (2). The unitary matrices $`(U_u)_{L,R}`$ are the diagonalizing matrices of the hermitian matrices $`H_u^1=y_u^{}(t)y_u^{}(t)`$ and $`H_u^2=y_u^{}(t)y_u^{}(t)`$ and they fulfill the equations which follow from Eq. (12)
$$\frac{d}{dt}H_u^i(t)=\frac{2}{(4\pi )^2}\{\alpha _1^u(t)+\alpha _2^uH_u^i+\alpha _3^u\text{Tr}(H_u^i)\}H_u^i,i=1,2.$$
(27)
The solution of Eq. (27) can be written in the following form
$$H_u^i(t)=H_u^i(t_0)+\underset{k=1}{\overset{\mathrm{}}{}}\frac{(tt_0)^k}{k!}\frac{d^kH_u^i(t)}{dt^k}|_{t=t_0}.$$
(28)
From Eq. (27) one can see that the derivative $`\frac{d^kH_u^i}{dt^k}|_{t=t_0}`$ is the sum of the powers of $`H_u^i(t_0)`$ with scalar coefficients. Thus from the hermiticity of $`H_u^i(t_0)`$ it follows that $`H_u^i(t)`$ are diagonalized by the same matrices as $`H_u^i(t_0)`$ so the diagonalizing matrices $`(U_u)_L`$ and $`(U_u)_R`$ of $`H_u^1(t)`$ and $`H_u^2(t)`$ are energy independent. Thus the matrix $`y_u(t)`$ has the following representation
$$y_u(t)=(U_u)_L^{}\mathrm{\Delta }_u(t)(U_u)_R^{}$$
(29)
where on the right hand side of Eq. (29) only the diagonal matrix $`\mathrm{\Delta }_u(t)=\text{Diag}(m_u(t),m_c(t),m_t(t))`$ does depend on $`t`$ and $`(U_u)_L`$ and $`(U_u)_R`$ are constant matrices that diagonalize $`y_u(t_0)`$. The matrix $`\mathrm{\Delta }_u(t)`$ fulfills the following equation
$$\frac{d}{dt}\mathrm{\Delta }_u=\frac{1}{(4\pi )^2}\{\alpha _1^u(t)+\alpha _2^u\mathrm{\Delta }_u^2+\alpha _3^u\text{Tr}(\mathrm{\Delta }_u^2)\}\mathrm{\Delta }_u.$$
(30)
Eq. (30) splits into 3 equations for the up quark masses
$`{\displaystyle \frac{dm_u}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\{\alpha _1^u(t)+\alpha _3^um_t^2\}m_u,`$ (32)
$`{\displaystyle \frac{dm_c}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\{\alpha _1^u(t)+\alpha _3^um_t^2\}m_c,`$ (33)
$`{\displaystyle \frac{dm_t}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\{\alpha _1^u(t)+(\alpha _2^u+\alpha _3^u)m_t^2\}m_t.`$ (34)
We first solve Eq. (34)
$$m_t(t)=\frac{m_t(t_0)r_g^{1/2}(t)}{\sqrt{1\frac{2}{(4\pi )^2}(\alpha _2^u+\alpha _3^u)m_t^2(t_0)_{t_0}^tr_g(\tau )𝑑\tau }}$$
(35)
where
$$r_g(t)=\mathrm{exp}(\frac{2}{(4\pi )^2}_{t_0}^t\alpha _1^u(\tau )𝑑\tau ).$$
(36)
Eqs. (32) and (33) are identical and their solution reads
$$m_{u,c}(t)=m_{u,c}(t_0)r_g^{1/2}(t)\mathrm{exp}(\frac{1}{(4\pi )^2}\alpha _2^u_{t_0}^tm_t^2(\tau )𝑑\tau ).$$
(37)
The full energy evolution of $`y_u(t)`$ is determined from Eq. (29) using Eqs. (35) and (37).
### B The evolution of $`y_d`$ from one loop RGE
The one loop evolution of $`y_d`$ is determined from Eq. (13). The fact that the diagonalizing matrices of $`y_u(t)`$ do not depend on $`t`$ simplifies Eq. (13) and justifies the following substitution
$$y_d(t)=(U_u)_L^{}W(t)$$
(38)
so $`W(t)`$ fulfills the equation
$$\frac{dW}{dt}=\frac{1}{(4\pi )^2}\{\alpha _1^d(t)+\alpha _2^d\mathrm{\Delta }_u^2+\alpha _3^d\text{Tr}(\mathrm{\Delta }_u^2)\}W.$$
(39)
The matrix on the right hand side of Eq. (39) has the following form
$`\{\alpha _1^d(t)+\alpha _2^d\mathrm{\Delta }_u^2+\alpha _3^d\text{Tr}(\mathrm{\Delta }_u^2)\}`$ (40)
$`=(\alpha _1^d(t)+\alpha _3^dm_t^2)\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& 1\end{array}\right)+\alpha _2^dm_t^2\left(\begin{array}{ccc}0& 0& 0\\ 0& 0& 0\\ 0& 0& 1\end{array}\right)`$ (47)
and the solution $`W(t)`$ of Eq. (39) is the following
$$W(t)=(r_g^{}(t))^{1/2}\mathrm{exp}(\frac{1}{(4\pi )^2}\alpha _3^d_{t_0}^tm_t^2(\tau )𝑑\tau )Z(t)W(t_0)$$
(48)
where
$$r_g^{}(t)=\mathrm{exp}(\frac{2}{(4\pi )^2}_{t_0}^t\alpha _1^d(\tau )𝑑\tau )$$
(49)
and
$$Z(t)=\left(\begin{array}{ccc}1& 0& 0\\ 0& 1& 0\\ 0& 0& h(t)\end{array}\right)$$
(50)
with
$$h(t)=\mathrm{exp}(\frac{1}{(4\pi )^2}\alpha _2^d_{t_0}^tm_t^2(\tau )𝑑\tau ).$$
(51)
Putting Eqs. (38)–(51) together we obtain the following result for the one loop evolution of $`y_d(t)`$
$$y_d(t)=(r_g^{}(t))^{1/2}(h(t))^{(\alpha _3^d/\alpha _2^d)}(U_u)_L^{}Z(t)(U_u)_L^{}y_d(t_0).$$
(52)
### C Higher order corrections to the $`y_u`$ and $`y_d`$ evolution
The two loop RGE for $`y_u`$ and $`y_d`$, Eqs. (19) and (21) have the following structure
$$\frac{dy_{u,d}}{dt}=\frac{1}{(4\pi )^2}E_1^{u,d}y_{u,d}+\frac{1}{(4\pi )^4}E_2^{u,d}y_{u,d}.$$
(53)
In Eqs. (53) the terms containing the functions $`E_2^{u,d}`$ are suppressed by the factor $`1/(4\pi )^2\lambda ^4`$ in comparison to the terms with $`E_1^{u,d}.`$ Eqs. (53) can be transformed into the integral equations
$$y_{u,d}(t)=y_{u,d}(t_0)+\frac{1}{(4\pi )^2}_{t_0}^tE_1^{u,d}y_{u,d}(\tau )𝑑\tau +\frac{1}{(4\pi )^4}_{t_0}^tE_2^{u,d}y_{u,d}(\tau )𝑑\tau $$
(54)
and Eqs. (54) can be transformed into the following recurrence relation
$$y_{u,d}^{(n)}(t)=y_{u,d}(t_0)+\frac{1}{(4\pi )^2}_{t_0}^tE_1^{u,d}y_{u,d}^{(n1)}(\tau )𝑑\tau +\frac{1}{(4\pi )^4}_{t_0}^tE_2^{u,d}y_{u,d}^{(n1)}(\tau )𝑑\tau .$$
(55)
The functions $`y_{u,d}^{(n)}(t)`$ converge for $`n\mathrm{}`$ to the solution of Eq. (53) for an arbitrary initial function $`y_{u,d}^{(0)}(t)`$. If we choose $`y_{u,d}^{(0)}(t)`$ to be the solution of the one loop RGE, Eqs. (29) and (52) then we obtain
$$y_{u,d}^{(1)}(t)=y_{u,d}(t_0)+\frac{1}{(4\pi )^2}_{t_0}^tE_1^{u,d}y_{u,d}^{(0)}𝑑\tau +\frac{1}{(4\pi )^4}_{t_0}^tE_2^{u,d}y_{u,d}^{(0)}𝑑\tau .$$
(56)
Now
$$y_{u,d}(t_0)+\frac{1}{(4\pi )^2}_{t_0}^tE_1^{u,d}y_{u,d}^{(0)}𝑑\tau =y_{u,d}^{(0)}(t)$$
(57)
because $`y_u^{(0)}`$ and $`y_d^{(0)}`$ fulfill Eqs. (12) and (13), respectively. It thus follows that
$$y_{u,d}^{(1)}(t)=y_{u,d}^{(0)}(t)+\frac{1}{(4\pi )^4}_{t_0}^tE_2^{u,d}y_{u,d}^{(0)}𝑑\tau $$
(58)
which means that the lowest order corrections to the solution of the one loop RGE for $`y_{u,d}^{(0)}(t)`$ are of the relative order $`\lambda ^4`$ when compared to the $`y_{u,d}^{(0)}(t)`$ and Eq. (58) gives the explicit form of this correction.
## IV Properties of the Renormalization Group Evolution of the Quark Masses and the CKM Matrix
In this section we will study the physical implications of the results obtained in the previous section. We will present them in the series of the theorems
###### Theorem 1
The one loop RGEV of the CKM matrix depends on only one function of the energy $`h(t)`$, given in Eq. (51).
*Proof:* The CKM matrix, given in Eq. (3) is constructed from the matrices $`(U_u)_L^{}`$ and $`(U_d)_L^{}`$. In section III.A we have shown that $`(U_u)_L^{}`$ does not depend on the energy. The dependence on $`t`$ can only be contained in $`(U_d)_L^{}`$ which is determined from the matrix $`y_d(t)`$ of the down QYC. The matrices $`(U_d)_L^{}`$ and $`(U_d)_R^{}`$ do not depend on the normalization of $`y_d(t)`$. From Eq. (52) one can see that the only dependence of $`y_d(t)`$ on $`t`$ apart from the normalization is contained in the matrix $`Z(t)`$, which is the diagonal matrix and as can be seen from Eq. (50) it depends only on the function $`h(t).`$ This completes the proof of Theorem 1.
###### Theorem 2
The ratios of the down quark masses $`m_d/m_s`$ and $`m_s/m_b`$ are the functions of only $`h(t)`$, given in Eq. (51)
*Proof:* The down quark masses are obtained after the diagonalization of the down QYC $`y_d(t)`$ and the ratios of $`m_d/m_s`$ and $`m_s/m_b`$ do not depend on the normalization of $`y_d(t).`$ It thus follows from the proof of Theorem 1 that these ratios are the functions of $`h(t)`$.
###### Theorem 3
The ratio of the up quark masses $`m_u/m_c`$ is energy independent.
*Proof:* This theorem follows directly from Eq. (37).
###### Theorem 4
The next order corrections<sup>*</sup><sup>*</sup>*we include in RGEV equations the terms of the order $`\lambda ^4`$ that come from the one loop RGE that contain the terms $`y_d^{}y_d^{}`$ and the two loop contributions of the order $`1`$ multiplied by $`1/(4\pi )^4.`$ of the RGEV of the CKM matrix are of the order $`\lambda ^5.`$
*Proof:* If we consider RGE in the next order then the terms of the order $`\lambda ^4`$ are preserved in the equations and the evolution is governed by Eqs. (II) and the explicit solution in the next leading order is given by Eq. (58). From Eq. (58) one obtains by the direct calculation the following result for the commutators
$`[y_{u,d}^{(1)}y_{u,d}^{(1)^{}},y_{u,d}^{(0)}y_{u,d}^{(0)^{}}]\lambda ^5,`$ (60)
$`[y_{u,d}^{(1)^{}}y_{u,d}^{(1)},y_{u,d}^{(0)^{}}y_{u,d}^{(0)}]\lambda ^5.`$ (61)
From Eqs. (IV) using the time independent perturbation theory one obtains that the diagonalizing matrices of the biunitary transformations in the next order are corrected by the terms of the order $`\lambda ^5`$ and this completes the proof.
## V Evolution of the CKM matrix
In this section we will find the equations for the one loop evolution of the squares of the absolute values of the off-diagonal elements of the CKM matrix.
The CKM matrix is defined in Eq. (3). Using the fact that $`(U_u)_L`$ is energy independent we obtain from Eq. (52)
$$(U_u)_Ly_d(t)y_d(t)^{}(U_u)_L^{}=r_g^{^{}}(t)(h(t))^{(2\alpha _3^d/\alpha _2^d)}Z(t)(U_u)_Ly_d(t_0)y_d(t_0)^{}(U_u)_L^{}Z(t)$$
(62)
which can be written in the following way
$$V_{CKM}(t)M_d^2(t)V_{CKM}^{}(t)=r_g^{^{}}(t)(h(t))^{(2\alpha _3^d/\alpha _2^d)}Z(t)(U_u)_Ly_d(t_0)y_d(t_0)^{}(U_u)_L^{}Z(t)$$
(63)
where $`M_d^2(t)`$ is the diagonal matrix of the squares of the physical down quarks Yukawa couplings which become the squares of the down quark masses after the spontaneous symmetry breaking. Now differentiating Eq. (63) with respect to $`t`$ we obtain the following result
$`V_{CKM}^{}(t){\displaystyle \frac{dV_{CKM}}{dt}}`$ (64)
$`=(M_d^2)^1V_{CKM}^{}(t){\displaystyle \frac{dV_{CKM}}{dt}}M_d^2(M_d^2)^1V_{CKM}^{}(t){\displaystyle \frac{d}{dt}}((U_u)_Ly_d(t)y_d(t)^{}(U_u)_L^{})V_{CKM}(t)+(M_d^2)^1{\displaystyle \frac{dM_d^2}{dt}}.`$ (65)
The second term on the right hand side of Eq. (65) is equal
$$(M_d^2)^1V_{CKM}^{}(t)\frac{d}{dt}((U_u)_Ly_d(t)y_d(t)^{}(U_u)_L^{})V_{CKM}(t)=\frac{d\mathrm{ln}(h)}{dt}(𝐑^{}𝐑+(M_d^2)^1𝐑^{}𝐑M_d^2)+\frac{d\mathrm{ln}(r_g^{^{}}(t)(h(t))^{(2\alpha _3^d/\alpha _2^d)})}{dt}I$$
(66)
where the vector $`𝐑=(V_{td},V_{ts},V_{tb})`$ and $`h(t)`$ is given in Eq. (51).
The off diagonal matrix elements of $`V_{CKM}^{}(t)\frac{dV_{CKM}}{dt}`$ can now be evaluated from Eq. (65). Using these matrix elements we obtain the following evolution equations
$$\frac{d|V_{ub}|^2}{dt}=\frac{h^{^{}}}{h}\frac{m_d^2+m_b^2}{m_d^2m_b^2}(V_{ub}^{}V_{ud}V_{td}^{}V_{tb})\frac{h^{^{}}}{h}\frac{m_s^2+m_b^2}{m_s^2m_b^2}(V_{ub}^{}V_{us}V_{ts}^{}V_{tb})+\text{c.c}\frac{2h^{^{}}}{h}|V_{ub}|^2|V_{tb}|^2,$$
(68)
$$\frac{d|V_{cb}|^2}{dt}=\frac{h^{^{}}}{h}\frac{m_d^2+m_b^2}{m_d^2m_b^2}(V_{cb}^{}V_{cd}V_{td}^{}V_{tb})\frac{h^{^{}}}{h}\frac{m_s^2+m_b^2}{m_s^2m_b^2}(V_{cb}^{}V_{cs}V_{ts}^{}V_{tb})+\text{c.c}\frac{2h^{^{}}}{h}|V_{cb}|^2|V_{tb}|^2,$$
(69)
$$\frac{d|V_{td}|^2}{dt}=\frac{h^{^{}}}{h}\frac{m_s^2+m_d^2}{m_s^2m_d^2}(V_{td}^{}V_{ts}V_{ts}^{}V_{td})\frac{h^{^{}}}{h}\frac{m_b^2+m_d^2}{m_b^2m_d^2}(V_{td}^{}V_{tb}V_{tb}^{}V_{td})+\text{c.c}\frac{2h^{^{}}}{h}|V_{td}|^2|(1|V_{td}|^2),$$
(70)
$$\frac{d|V_{cd}|^2}{dt}=\frac{h^{^{}}}{h}\frac{m_s^2+m_d^2}{m_s^2m_d^2}(V_{cd}^{}V_{cs}V_{ts}^{}V_{td})\frac{h^{^{}}}{h}\frac{m_b^2+m_d^2}{m_b^2m_d^2}(V_{cd}^{}V_{cb}V_{tb}^{}V_{td})+\text{c.c}\frac{2h^{^{}}}{h}|V_{cd}|^2|V_{td}|^2.$$
(71)
Eqs. (V) can be solved explicitly and the solution reads
$$|V_{ub}|^2=\frac{|V_{ub}^0|^2}{|V_{tb}^0|^2(h^21)+1},$$
(73)
$$|V_{cb}|^2=\frac{|V_{cb}^0|^2}{|V_{tb}^0|^2(h^21)+1},$$
(74)
$$|V_{td}|^2=\frac{|V_{td}^0|^2}{|V_{td}^0|^2(1h^2)+h^2},$$
(75)
$$|V_{cd}|^2=\frac{|V_{cd}^0|^2}{|V_{td}^0|^2(1h^2)+h^2}.$$
(76)
Here $`|V_{ij}^0|^2`$ are the initial values of the squares of the absolute values of the corresponding CKM matrix elements and $`h`$ is given in Eq. (51). From the unitarity of the CKM matrix and using Eqs. (71) one can calculate the absolute values of all the remaining matrix elements of the CKM matrix and thus we determine the renormalization group evolution of the full CKM matrix.
## VI Conclusions
In this paper we have analyzed the renormalization group evolution of the quark Yukawa couplings and the CKM matrix based on the observed hierarchy of the quark masses and the CKM matrix. The inclusion of the hierarchy greatly simplifies the analysis and leads to simple, explicit results for the evolution of the Yukawa couplings (Eqs. (29) and Eqs. (52)) and the CKM matrix (Eqs. (71)). The other remarkable result is that the diagonalizing matrices of the up quark Yukawa couplings are energy independent in the leading order. This means that the transformation $`(\psi _u)_{L,R}(U_u)_{L,R}(\psi _u)_{L,R}`$, $`(\psi _d)_{L,R}(U_u)_{L,R}(\psi _d)_{L,R}`$ will diagonalize the matrix of the up quark Yukawa couplings and it will stay diagonal upon the renormalization group evolution and the CKM matrix will be determined only from the down quarks Yukawa couplings. This fact may simplify the model building based on the symmetries of the quark Yukawa couplings.
###### Acknowledgements.
This paper is partially supported by the CoNaCyT (Mexico) project 3512P-E9608. S.R.J.W. gratefully acknowledges partial support by Comisión de Operación y Fomento de Actividades Académicas - COFAA (Instituto Politécnico Nacional). J.G. Mora H. thanks dr. A. Odzijewicz for the hospitality in the Institute of Theoretical Physics of the Białystok University (Poland) during the preparation of the paper.
## Constants in the renormalization group equations
The two loop renormalization group equations for various models have the following structure
$$\frac{dg_l}{dt}=\frac{1}{(4\pi )^2}b_lg_l^3\frac{1}{(4\pi )^4}G_lg_l^3,$$
(77)
$$\frac{dy_{u,d,e,\nu }}{dt}=\left[\frac{1}{(4\pi )^2}\beta _{u,d,e,\nu }^{(1)}+\frac{1}{(4\pi )^4}\beta _{u,d,e,\nu }^{(2)}\right]y_{u,d,e,\nu }$$
(78)
where $`b_l`$ are constants dependent on the model and $`G_l`$, $`\beta _{u,d,e,\nu }^{(1)}`$ and $`\beta _{u,d,e,\nu }^{(2)}`$ are the following functions of the coupling constants and the squares of the Yukawa couplings $`H_{u,d,e,\nu }^{(1)}=y_{u,d,e,\nu }y_{u,d,e,\nu }^{}`$
$$G_l=C_{lu}\text{Tr}(H_u^{(1)})+\underset{k}{}b_{kl}g_k^2,$$
(79)
$`\beta _l^{(1)}`$ $`=`$ $`\alpha _1^l(t)+\alpha _2^lH_u^{(1)}+\alpha _3^l\text{Tr}(H_u^{(1)})+\alpha _4^lH_d^{(1)}+\alpha _5^l\text{Tr}(H_d^{(1)})`$ (80)
$`+`$ $`\alpha _6^lH_e^{(1)}+\alpha _7^l\text{Tr}(H_e^{(1)})+\alpha _8^lH_\nu ^{(1)}+\alpha _9^l\text{Tr}(H_\nu ^{(1)}),`$ (81)
$`\beta _l^{(2)}`$ $`=`$ $`{\displaystyle \underset{lkn}{}}B_{lkn}^1g_k^2g_n^2+{\displaystyle \underset{lkn}{}}B_{lkn}^2g_k^2H_n^{(1)}+{\displaystyle \underset{lkn}{}}B_{lkn}^3g_k^2\text{Tr}(H_n^{(1)})`$ (82)
$`+`$ $`{\displaystyle \underset{lkn}{}}B_{lkn}^4H_k^{(1)}H_n^{(1)}+{\displaystyle \underset{lkn}{}}B_{lkn}^5H_k^{(1)}\text{Tr}(H_n^{(1)}).`$ (83)
The functions $`\alpha _1^l(t)`$ are equal
1. for the SM and DHM
$`\alpha _1^u(t)=({\displaystyle \frac{17}{20}}g_1^2+{\displaystyle \frac{9}{4}}g_2^2+8g_3^2),`$ (84)
$`\alpha _1^d(t)=({\displaystyle \frac{1}{4}}g_1^2+{\displaystyle \frac{9}{4}}g_2^2+8g_3^2),`$ (85)
$`\alpha _1^e(t)=({\displaystyle \frac{9}{20}}g_1^2+{\displaystyle \frac{9}{4}}g_2^2),`$ (86)
$`\alpha _1^\nu (t)=({\displaystyle \frac{9}{20}}g_1^2+{\displaystyle \frac{9}{4}}g_2^2).`$ (87)
2. for the MSSM
$`\alpha _1^u(t)=({\displaystyle \frac{13}{15}}g_1^2+3g_2^2+{\displaystyle \frac{16}{3}}g_3^2),`$ (88)
$`\alpha _1^d(t)=({\displaystyle \frac{7}{15}}g_1^2+3g_2^2+{\displaystyle \frac{16}{3}}g_3^2),`$ (89)
$`\alpha _1^e(t)=({\displaystyle \frac{9}{5}}g_1^2+3g_2^2),`$ (90)
$`\alpha _1^\nu (t)=({\displaystyle \frac{3}{5}}g_1^2+3g_2^2).`$ (91)
The values of the coefficients $`b_l`$ and $`\alpha _k^l`$ for the one loop renormalization group equations are given in Tables I, II and III (see e.g. ).
The coefficients for the two loop renormalization group equations that appear in the functions $`G_l`$ and $`\beta _l^{(2)}`$ can be found in Ref. .
|
warning/0002/cond-mat0002002.html
|
ar5iv
|
text
|
# Localization-delocalization transition of disordered 𝑑-wave superconductors in class 𝐶I
## Abstract
A lattice model for disordered $`d`$-wave superconductors in class $`C`$I is reconsidered. Near the band-center, the lattice model can be described by Dirac fermions with several species, each of which yields WZW term for an effective action of the Goldstone mode. The WZW terms cancel out each other because of the four-fold symmetry of the model, which suggests that the quasiparticle states are localized. If the lattice model has, however, symmetry breaking terms which generate mass for any species of the Dirac fermions, remaining WZW term which avoids the cancellation can derive the system to a delocalized strong-coupling fixed point.
Dirty superconductors have attracted much interest, since they provide wider universality classes of disordered systems . In particular, it is quite interesting to ask what the universality class of the disordered $`d`$-wave superconductors is . A remarkable property of this unconventional superconductors is that near the band center, quasiparticle states can be described by Dirac fermions . Such a description enables us, for example, to relate $`d`$-wave superconductors with the quantum Hall effect and to predict a new spin phase called spin quantum Hall fluid . They should also produce the well-known effect of chiral anomaly to $`d`$-wave superconductors. Recently, Senthil et al have studied disordered $`d`$-wave superconductors with spin rotational symmetry and reached the conclusion that all states are localized. One knows, however, that the WZW term due to chiral anomaly plays a crucial role in two dimensional critical phenomena . Therefore, it is quite important to take the WZW term missing in into account, or to answer the question why it vanishes if it does not exist.
In this paper, we reconsider disorder effects on the quasiparticle properties of the $`d`$-wave superconductors in the class $`C`$I (those with spin rotational and time-reversal symmetries) using a replica technique. It is shown that each Dirac fermion associated with four nodes creates the WZW term. It turns out that they cancel each other and the resultant nonlinear sigma model suggests localization of the quasiparticles in dirty $`d`$-wave superconductors, as was shown by Senthil et al . It should be stressed, however, that the WZW term is potentially realizable and the cancellation is accidental: It is due to the four-fold symmetry of the model. Therefore, if such symmetry is broken, the system flows to the strong-coupling fixed point described by the WZW model.
Let us begin with a lattice Hamiltonian for singlet superconductors ,
$$H=\underset{i,j}{}\left(t_{ij}\underset{\sigma }{}c_{i\sigma }^{}c_{j\sigma }+\mathrm{\Delta }_{ij}c_i^{}c_j^{}+\mathrm{\Delta }_{ij}^{}c_jc_i\right).$$
(1)
We can choose real and symmetric matrices $`t_{ij}=t_{ji}`$ and $`\mathrm{\Delta }_{ij}=\mathrm{\Delta }_{ji}`$ taking account of the hermiticity as well as the spin-rotational and the time-reversal symmetries. In the absence of randomness, we choose the following parameters for a pure Hamiltonian $`H_0`$ with $`d`$-wave symmetry
$$t_{j,j\pm \widehat{x}}=t_{j,j\pm \widehat{y}}=t_0,\mathrm{\Delta }_{j,j\pm \widehat{x}}=\mathrm{\Delta }_{j,j\pm \widehat{y}}=\mathrm{\Delta }_0,$$
(2)
where $`\widehat{x}=(1,0)`$ and $`\widehat{y}=(0,1)`$. Moreover, consider introducing small terms $`H_1`$ to break the $`C`$I symmetry,
$$\mathrm{\Delta }_{j,j+\widehat{x}\pm \widehat{y}}=\mathrm{\Delta }_{j+\widehat{x}\pm \widehat{y},j}=\pm \frac{i\mathrm{\Delta }_1}{4},\mathrm{\Delta }_{j,j}=i\mathrm{\Delta }_1.$$
(3)
It will be shown momentarily that this parameter controls the localization-delocalization transition of the present model.
The pure Hamiltonian $`H_0`$ has four nodes, where gapless quasi-particle excitations exist . Therefore, we can firstly take the continuum limit around the nodes of $`H_0`$ and next incorporate the continuum expression of $`H_1`$, provided that $`H_1`$ is small. The lattice operators are then described by the continuum slowly-varying fields near the band center as,
$`c_j/a`$ $`i^{j_x+j_y}\psi _1^1(x)i^{j_xj_y}\psi _2^1(x)`$ (5)
$`+i^{j_x+j_y}\psi _1^2(x)i^{j_xj_y}\psi _2^2(x),`$
$`c_j/a`$ $`i^{j_x+j_y}\psi _1^1(x)+i^{j_xj_y}\psi _2^1(x)`$ (7)
$`+i^{j_x+j_y}\psi _1^2(x)+i^{j_xj_y}\psi _2^2(x),`$
where $`a`$ is a lattice constant, $`x=aj`$, and two kinds of lower indices of the field $`\psi `$ are referred to as spin, left-right (LR) movers, respectively, and upper index as node. Namely, the field variable $`\psi (x)`$ lives in the space $`V=𝑪^2𝑪^2𝑪^2`$. The pure Hamiltonian in the continuum limit is then $`H_0=d^2x\psi ^{}(_0+_1)\psi `$ with
$$_0=\left(\begin{array}{cc}\gamma _\mu i_\mu \hfill & \\ & (xy)\hfill \end{array}\right),$$
(8)
where the coordinates have been transformed as $`x,y\frac{\pm x+y}{\sqrt{2}}`$. The explicit matrix in Eq. (8) denotes the node space, and matrices $`\gamma _\mu `$ belong to the other space of $`V`$, calculated initially as $`\gamma _1=v_F1_2\sigma _3`$ and $`\gamma _2=v_\mathrm{\Delta }1_2\sigma _1`$, where $`v_F=2\sqrt{2}t_0a`$ and $`v_\mathrm{\Delta }=2\sqrt{2}\mathrm{\Delta }_0a`$. It may be more convenient to choose
$$\gamma _1=v_F1_2\sigma _2,\gamma _2=v_\mathrm{\Delta }1_2\sigma _1,$$
(9)
via suitable rotation in LR-space of $`V`$. In this basis, $`_1`$ is given by
$$_1=\left(\begin{array}{cc}0\hfill & \\ & m1_2\sigma _3\hfill \end{array}\right),$$
(10)
where $`m=\mathrm{\Delta }_1`$. Namely, $`H_1`$ yields asymmetric mass term in the continuum Hamiltonian. For the time being we neglect it, but it will be shown that vanishing $`m`$ leads to localization whereas finite $`m`$ drives the system to delocalization.
The spin-rotational and time-reversal symmetries of the lattice model translate, respectively, into the continuum model as
$$\begin{array}{cc}=𝒞^\mathrm{t}𝒞^1,\hfill & 𝒞=i\sigma _2\sigma _11_2,\hfill \\ =𝒯𝒯^1,\hfill & 𝒯=1_2\sigma _31_2,\hfill \end{array}$$
(11)
where t means the transpose. The total Hamiltonian density is given by $`=_0+_\mathrm{d}`$, where $`_\mathrm{d}`$ is disorder potential satisfying Eq. (11). It is stressed that we take account of all kinds of disorder potentials satisfying Eq. (11). This implies that the summation with respect to $`i,j`$ in Eq. (1) should be over on-site, nearest-neighbor, and diagonal-second-neighbor pairs. Namely, we can achieve “maximum information entropy” for the Dirac Hamiltonian with the symmetries (11) when we introduce not only on-site disorder potentials but also disordered hopping and off-diagonal pairing for the lattice model in Eq. (1). Actually Ludwig et al have derived generic Dirac Hamiltonian for the integer quantum Hall transition considering lattice model in a similar situation . It may be straightforward to explicitly calculate disorder potentials but tedious to average over them, because there are no less than twenty independent potentials satisfying Eq. (11). The key point for the ensemble average is that if the model has “maximum entropy”, we can use the technique developed by Zirnbauer .
To study one-quasiparticle properties of the model, we introduce the Green function $`G_{aa^{}}(x,x^{};iϵ)=x,a|(iϵ)^1|x^{},a^{}`$, where index $`a`$ denotes the set of spin, LR, and node species in the space $`V`$ and $`|x,a=\psi _a^{}(x)|0`$. Especially, we need
$`G(x)={\displaystyle \underset{a}{}}G_{aa}(x,x;iϵ),`$ (12)
$`K(x,x^{})={\displaystyle \underset{a,a^{}}{}}G_{aa^{}}(x,x^{};iϵ)G_{a^{}a}(x^{},x;iϵ)`$ (13)
to compute the DOS and the conductance of the quasiparticle transport . Some notations are convenient for this purpose. The introduction of replica for the field $`\psi `$ enables us to express the generating functional of these Green functions by path integrals: $`\psi _a\psi _{a\alpha }`$ and $`\psi _a^{}\psi _{\alpha a}^{}`$, where $`a`$ and $`\alpha `$ are indices denoting $`V`$ and the replica space $`W_\mathrm{r}=𝑪^n`$, respectively. The fields $`\psi `$ and $`\psi ^{}`$ have been converted into matrix fields, which makes it simpler to define an order parameter field. It should be stressed that the fields $`\psi `$ and $`\psi ^{}`$ are completely independent variables. Lagrangian density is then described symbolically as $`=\mathrm{tr}_{W_\mathrm{r}}\psi ^{}\left(iϵ\right)\psi `$, where $`\mathrm{tr}_{W_\mathrm{r}}`$ is the trace in the replica space and the summation over the indices $`a`$ of the $`V`$ space is implied according to the rule of the matrix product. It should be noted again that $`\psi ^{}`$ is independent of $`\psi `$. Moreover, we introduce an auxiliary space to reflect the symmetries in the $`V`$ space (11) to an auxiliary field introduced later \[See Eq. (23)\], $`W_\mathrm{r}W=W_\mathrm{r}W_\mathrm{a}`$ with $`W_\mathrm{a}=𝑪^2𝑪^2`$, which are associated with the spin-rotational and the time-reversal symmetries, respectively. Fermi fields are now denoted by $`\stackrel{~}{\mathrm{\Psi }}_{\alpha i}`$ and $`\mathrm{\Psi }_{i\alpha }`$. One of simpler choices is
$`\mathrm{\Psi }=(\psi _+,\psi _{}),\psi _\pm =𝒯_\pm \stackrel{~}{\psi },`$ (14)
$`\stackrel{~}{\mathrm{\Psi }}=\left(\begin{array}{c}\overline{\psi }_+\\ \overline{\psi }_{}\end{array}\right),\overline{\psi }_\pm =\stackrel{~}{\psi }^{}𝒯_{},`$ (17)
where $`𝒯_\pm =(1\pm 𝒯)/2`$ serves as a projection operator ( $`𝒯_++𝒯_{}=1_21_21_2`$, $`𝒯_\pm ^2=𝒯_\pm `$, and $`𝒯_+𝒯_{}=0`$ ) into each chiral component of Sp($`n`$)$`\times `$Sp($`n`$) symmetry, as we shall see later, and $`\stackrel{~}{\psi }=\frac{1}{2}(\psi _1,i\psi _2)`$ with $`\psi _{1,2}=\psi \pm i𝒞^1\psi ^{}`$. The newly introduced fields $`\mathrm{\Psi }`$ and $`\stackrel{~}{\mathrm{\Psi }}`$ are subject to ,
$$\begin{array}{cc}\stackrel{~}{\mathrm{\Psi }}=\gamma \mathrm{\Psi }^\mathrm{t}𝒞^1,\hfill & \mathrm{\Psi }=𝒞\stackrel{~}{\mathrm{\Psi }}^\mathrm{t}\gamma ^1,\hfill \\ \stackrel{~}{\mathrm{\Psi }}=\pi \stackrel{~}{\mathrm{\Psi }}𝒯^1,\hfill & \mathrm{\Psi }=𝒯\mathrm{\Psi }\pi ^1.\hfill \end{array}$$
(18)
Matrices $`\gamma `$ and $`\pi `$ are defined in the $`W`$ space by
$`\gamma =1_ni\sigma _21_2\gamma _01_2,`$ (19)
$`\pi =1_n1_2\sigma _3.`$ (20)
The identity $`\mathrm{tr}_W(iϵ\omega \stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi })=\mathrm{tr}_{W_\mathrm{r}}\psi ^{}(iϵ)\psi `$, where $`\omega =1_n\sigma _2\sigma _1`$, leads to the generating functional $`𝒵=𝒟\mathrm{\Psi }𝒟\stackrel{~}{\mathrm{\Psi }}e^S`$ with
$$S=d^2x\mathrm{tr}_W\left(iϵ\omega \stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }+J\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }\right).$$
(21)
Assume that disorder potential $`_\mathrm{d}`$ obeys the Gaussian distribution $`P[_\mathrm{d}]=\mathrm{exp}\left(\frac{1}{2g}\mathrm{tr}_V_\mathrm{d}^2\right)`$. Then ensemble average over disorder is quite simple. The procedure is as follows: The disorder potentials are integrated out by using the identity $`\frac{1}{2g}\mathrm{tr}_V_\mathrm{d}^2+\mathrm{tr}_V_\mathrm{d}\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }}=\frac{1}{2g}(\mathrm{tr}_V_\mathrm{d}g\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }})^2+\frac{g}{2}\mathrm{tr}_V(\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }})^2`$. It turns out that the integration over $`_\mathrm{d}`$ is automatic because $`\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }}`$ satisfy the same symmetries as those of $`_\mathrm{d}`$ due to Eq. (18). This is actually a merit to consider the disorder potentials with maximum entropy. If disordered hopping and off-diagonal pairing of the lattice model are neglected and on-site disorder potentials are merely taken into account, some other conditions should be imposed on $`_\mathrm{d}`$ and therefore on the fields $`\mathrm{\Psi }`$ and $`\stackrel{~}{\mathrm{\Psi }}`$. Now we have interaction terms of fermions due to ensemble average. Note the identity $`\mathrm{tr}_V(\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }})^2=\mathrm{tr}_W(\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi })^2`$. Then, the four fermi interactions are decoupled via auxiliary matrix (order parameter) field defined in the $`W`$ space. To be concrete, add the following term into the action, $`\frac{1}{2g}\mathrm{tr}_W(Q+g\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }\omega )^2`$, which is actually a constant after integration over $`Q`$. Then we reach an effective Lagrangian density,
$`=`$ $`\mathrm{tr}_W\left[{\displaystyle \frac{1}{2g}}\left(Q^22iϵ\omega Q\right)+Q\stackrel{~}{\mathrm{\Psi }}\mathrm{\Psi }\stackrel{~}{\mathrm{\Psi }}_0\mathrm{\Psi }\right].`$ (22)
Here we have set $`J=0`$ for simplicity. Notice that the anti-Hermitian auxiliary field $`Q=Q^{}`$ is subject to
$$Q=\gamma Q^\mathrm{t}\gamma ^1,Q=\pi Q\pi ^1.$$
(23)
The solution of these equations is
$$Q=\left(\begin{array}{cc}& M^{}\hfill \\ M\hfill & \end{array}\right)$$
(24)
with a condition $`M=\gamma _0M^{}\gamma _0^1`$, where the explicit matrix in the above equation denotes the time-reversal space of $`W`$.
The Lagrangian (22) has G=Sp($`n`$)$`\times `$Sp($`n`$) symmetry. To see this, let us consider the transformation $`QgQg^1`$ which keeps the symmetry relations (23). It turns out that $`g`$ should satisfy $`\gamma =g\gamma g^\mathrm{t}`$ and $`\tau g\tau ^1=g`$ as well as $`gg^{}=1`$, and therefore $`g`$ is explicitly given by
$$g=\left(\begin{array}{cc}g_+\hfill & \\ & g_{}\hfill \end{array}\right),$$
(25)
where $`g_\pm `$ Sp($`n`$) is a $`2n\times 2n`$ matrix in the replica and the spin space of $`W`$. So far we have derived the Lagrangian as well as its symmetry group. To integrate out the fermi fields, it may be convenient to use the notations where the fermi fields $`\mathrm{\Psi }`$ and $`\stackrel{~}{\mathrm{\Psi }}`$ are column and row vector, respectively, as usual. Using Eq. (23), fermion part of the Lagrangian (22) is rewritten as $`_\mathrm{F}=\stackrel{~}{\mathrm{\Psi }}(_011Q^\mathrm{t})\mathrm{\Psi }=\stackrel{~}{\mathrm{\Psi }}1\gamma (_01+1Q)1\gamma ^{}\mathrm{\Psi }`$. Transform the fields as $`\stackrel{~}{\mathrm{\Psi }}\stackrel{~}{\mathrm{\Psi }}1\gamma ^{}`$ and $`\mathrm{\Psi }1\gamma \mathrm{\Psi }`$. Then the Lagrangian is given by, in terms of the fields $`M`$ and $`\psi _\pm `$,
$``$ $`={\displaystyle \frac{1}{g}}\mathrm{tr}_{W_{\mathrm{rs}}}\left[M^{}Mϵ\gamma _0(MM^{})\right]+_{\mathrm{F1}}+_{\mathrm{F2}},`$ (26)
where $`W_{\mathrm{rs}}`$ is the replica and the spin space of $`W`$ and $`_{\mathrm{F}j}`$ describes the Lagrangian of the $`j`$th-node fermion defined by
$$_{\mathrm{F1}}=\overline{\psi }_+^1i\overline{)}\psi _+^1+\overline{\psi }_{}^1i\overline{)}\psi _{}^1\overline{\psi }_+^1M^{}\psi _{}^1+\overline{\psi }_{}^1M\psi _+^1,$$
(27)
and $`_{\mathrm{F2}}=_{\mathrm{F1}}(12,xy)`$. Here and hereafter, the identity matrices such as those in $`\overline{)}1`$ and $`1M`$ are suppressed. The transformation laws of $`M`$ and $`\psi _\pm `$ fields are
$`Mg_{}Mg_+^{},\overline{\psi }_\pm \overline{\psi }_\pm g_\pm ^{},\psi _\pm g_\pm \psi _\pm .`$ (28)
We have used a bit complicated basis for $`W`$ in the definition of $`\mathrm{\Psi }`$ and $`\stackrel{~}{\mathrm{\Psi }}`$ in Eq. (17), since $`Q`$ and $`g`$ become simpler in this basis. On the other hand, it may cause a difficulty in computing the saddle points. We have known, however, from the $`ϵ`$-term in Eqs. (21) and (22) that $`\omega `$ serves as a “metric” in the extended auxiliary space. Usually we may choose a basis of the space $`W`$ with a diagonal metric, owing to which we can assume that $`Q`$ is also diagonal on the saddle points. In the present case, therefore, it is natural to assume that $`Q`$ should have the same structure as $`\omega `$, and hence $`M_0=v\gamma _0`$ with real diagonal matrix $`v=\text{diag}(v_1,\mathrm{},v_n)`$. Then the variation with respect to $`v`$ after the integration over fermi fields tells that the saddle points are given by $`v_\alpha =v_0\mathrm{\Lambda }\mathrm{exp}(\pi v_Fv_\mathrm{\Delta }/g)`$, where $`\mathrm{\Lambda }`$ is a ultraviolet cut-off. This solution gives rise to an exponentially small density of state at the band-center. Now it is easy to identify the saddle point manifold as H=Sp($`n`$): The chiral transformation $`g(g_+,g_{})`$ in Eqs. (25) and (28) is divided into two types. One is $`g_\mathrm{v}=(g_1,g_1^{})`$ under which $`M_0`$ is still invariant, and the other is $`g_\mathrm{a}=(g_2,g_2^\mathrm{t})`$ under which $`M_0`$ is no longer invariant.
Nonlinear sigma model is derived as small fluctuation around the saddle point manifold H by considering $`g_\mathrm{a}`$ type local Sp($`n`$) rotation. Let us parameterize $`M=\xi ^{}H\xi =\stackrel{~}{H}U`$ with $`\xi `$ Sp($`n`$), $`\stackrel{~}{H}=\xi ^{}H\xi ^{}`$, and $`U=\xi ^2`$. The field $`\xi `$ describes the massless fluctuation around the saddle point manifold H, whereas $`H`$ describes massive longitudinal modes. In what follows, we take only the leading order for the latter mode, setting $`H=M_0(=\stackrel{~}{H})`$. It should be noted that nonlinear sigma model on G/H has a global G symmetry as well as a hidden local H symmetry. Though the field $`\xi `$ itself is not invariant under local H transformation, the composite field $`U`$ is invariant.
To derive an effective action for the transverse mode, let us come back to the Lagrangian (27), since we should be careful in the integration over fermi fields. It is practical to firstly integrate out the fermi field of node-1. Then, the contribution from node-2 can be obtained by replacing $`xy`$. To carry out the former integration, make the transformation $`\overline{\chi }_+=\overline{\psi }_+U^{}`$ and $`\chi _+=U\psi _+`$, whereas $`\overline{\chi }_{}=\overline{\psi }_{}1_n\sigma _2`$ and $`\chi _{}=1_n\sigma _2\psi _{}`$. The Lagrangian (27) is then converted into
$`_{\mathrm{F1}}=\overline{\chi }_+^1i\overline{)}D\chi _+^1+\overline{\chi }_{}^1i\overline{)}\chi _{}^1+iv_0\left(\overline{\chi }_+^1\chi _{}^1+\overline{\chi }_{}^1\chi _+^1\right),`$ (29)
where $`D_\mu `$ is defined by $`D_\mu =_\mu +L_\mu `$ with $`L_\mu =U_\mu U^{}`$. It is convenient to scale $`x=v_Fx^{}`$ and $`y=v_\mathrm{\Delta }y^{}`$ working in the node-1 sector. Integration over the fermi field of node-1 yields
$`Z_{\mathrm{F1}}=e^{\frac{1}{2}\mathrm{\Gamma }_1(U)}`$ (30)
$`\times \mathrm{Det}_{V_1W_{\mathrm{rs}}}^{\frac{1}{2}}1_2\left(\begin{array}{cc}iv_0& i(i_1_2)\\ i(iD_1D_2)& iv_0\end{array}\right),`$ (33)
where the identity matrix $`1_2`$ belongs to the spin space of $`V`$, $`V_1`$ means the node-1 sector of $`V`$, the derivatives are with respect to the scaled coordinates $`x^{}`$ and $`y^{}`$, and $`\mathrm{\Gamma }_1(U)`$ is the Jacobian for the chiral transformation, which can be calculated by using the Fujikawa method . We simply present the final answer $`Z_{\mathrm{F1}}e^{S_1}`$, where the effective action $`S_1`$ associated with the node-1 fermion is composed of the principal chiral action of $`U`$ with a coupling constant $`\lambda =4\pi `$ and of the WZW term
$`\mathrm{\Gamma }_{\mathrm{WZW}}={\displaystyle \frac{i}{12\pi }}{\displaystyle d^3xϵ_{\mu \nu \tau }\mathrm{tr}_{W_{\mathrm{rs}}}_\mu UU^{}_\nu UU^{}_\tau UU^{}}.`$ (34)
Therefore, integration over fermion of node-1 actually yields the WZW term.
Next let us compute the contribution from the node-2. The procedure is the rescaling $`x^{}=x/v_F`$ and $`y^{}=y/v_\mathrm{\Delta }`$ and the exchange $`xy`$. It should be noted that the WZW term is invariant under the rescaling but is odd under the exchange, and hence it cancels out. The total effective action ends up with
$$S=d^2x\mathrm{tr}_{W_{\mathrm{rs}}}\left[\frac{1}{2\lambda }_\mu U_\mu U^{}ϵ\left(U+U^{}\right)\right]+k\mathrm{\Gamma }_{\mathrm{WZW}}$$
(35)
with $`\frac{1}{\lambda }=\frac{1}{4\pi }\frac{v_F^2+v_\mathrm{\Delta }^2}{v_Fv_\mathrm{\Delta }}`$ and $`k=0`$. This is just the action derived by Senthil et al. Although the WZW term disappears in the ordinary $`d`$-wave superconductors, we are tempted to expose it, since the WZW term exists potentially. This is indeed possible: If the lattice model includes the symmetry breaking term (3) and therefore the Dirac fermion for the node-2 has a mass (10), we can neglect the node-2 fermion in the lower energy than the mass gap. In this case, we have the same action (35) but with $`\frac{1}{\lambda }=\frac{1}{4\pi }`$ and $`k=1`$ in the scaled coordinates $`x^{}`$ and $`y^{}`$.
The renormalization group equations of the action (35) are calculated at the one-loop order as
$`{\displaystyle \frac{d\lambda }{d\mathrm{ln}L}}=\epsilon \lambda +{\displaystyle \frac{\lambda ^2}{4\pi }}\left[1\left({\displaystyle \frac{k\lambda }{4\pi }}\right)^2\right],`$ (36)
$`{\displaystyle \frac{dϵ}{d\mathrm{ln}L}}=\left(d{\displaystyle \frac{\lambda }{8\pi }}\right)ϵ,`$ (37)
where $`d=2`$, $`\epsilon =d2`$, and the replica limit $`n0`$ has been taken. In the case where $`m=0`$, we reach the same conclusion as Senthil et al. Namely, one-quasiparticle states are localized, since the spin conductance against weak magnetic fields is related with $`\lambda `$ as $`\sigma =2/(\pi \lambda )`$, which is calculated via diffusion constant in the diffusion propagator (13). It should be stressed that the cancellation of the WZW term is due to the four-fold symmetry of the $`d`$-wave Hamiltonian. Actually, the latent WZW term can emerge via symmetry breaking mass term for the Dirac fermions, and in that case the coupling constant $`\lambda `$ flows to the strong-coupling fixed point value $`\lambda _\mathrm{c}=4\pi /k`$. This fixed-point is conformal invariant described by Sp($`n`$) WZW model. From the scaling dimension of the energy in Eq. (37), it turns out that the density of state near this fixed-point obeys the scaling law
$$\rho (E)=E^{\frac{1}{4k1}}.$$
(38)
Therefore, if the pure model has the breaking term (3), we suggest that $`\rho (E)=E^{\frac{1}{3}}`$.
In this paper, we have taken the breaking term of type (3) into account. More detailed phase diagram will be published elsewhere. Numerical check of the present conjecture does not seem difficult. It is quite interesting to expose the hidden WZW term which exists potentially but conceals itself in the four-fold symmetry of the $`d`$-wave superconductors.
The author would like to thank Y. Hatsugai, Y. Morita, C. Mudry, and Y. Kato for helpful discussions and comments. He is deeply indebted to M. R. Zirnbauer and A. Altland for valuable discussions in the early stage of this work.
|
warning/0002/quant-ph0002034.html
|
ar5iv
|
text
|
# A model for ensemble NMR quantum computer using antiferromagnetic structure
## Introduction
The fundamental obstacle, preventing experimentalists from extending the number of qubits to $`L1`$ in an individual molecule of the liquid-state NMR quantum computer, is the difficulty of distinguishing $`L`$ unique set of two-state cells. To remove this obstacle it was already proposed several models for solid-state quantum computers with both individual and ensemble control of qubits. One of such potentially realizable model based on a one-dimensional cellular automaton, using an one-dimensional periodic array ABCABC… of three types of two-state quantum-mechanical cells (they may be heteronuclear system of spins $`I=1/2`$) with distinct resonant frequencies and local interaction between near neighbors, was first considered by S.Lloyd . The effect of the interaction contains in a shift of the each cell energy levels depending on states of its neighbors. After using the resonant $`\pi `$pulse all cells of type A, for instance, invert their state if, and only if, the left neighbor C is in ground state and the B on its right is in excited state. In , it was represented algorithm, which was applied globally to all cells, so that there is no need to address cells individually. This model was recently developed by S.Lloyd .
The more general model of a solid state ensemble NMR quantum computer was described in , where it was considered periodic structure of ABCABCABC… type in two or three dimensions with the nuclear spins 1/2 only of three distinguish types A, B, C. It was supposed that the nuclei are embedded in a crystal lattice of some solid state compound with spinless nuclei and all spins are initialized to the ground state $`|0`$. Each ABC-unit of this superlattice can be used to store quantum information by setting one of spin up or down. This information can be moved around via some quantum cellular state shifting mechanism. Cascading unitary quantum SWAP operations of $`A\mathrm{B}`$, $`B\mathrm{C}`$, $`C\mathrm{A}`$, $`A\mathrm{B}`$,… is used for this process. An ancillary dopand nucleus $`D`$ with spin 1/2 in the proximity of an A-site can serve as the input/output port. A local environment region near dopand nucleus provides a large quantum system with a wealth of qubits and only three types of nuclear spins. Therefore impurity doping may induce large-scale quantum automata in a single crystal and the whole crystal contains a huge ensemble of such identical NMR quantum computers — large artificial ”molecules”.
One-dimensional scheme that was based only on two different A and B types of cell in a periodic array without the ability to distinguish the left neighbor from the right was described in . Each two-state cell of the scheme has ground $`|`$ and excited $`|`$ internal eigenstates and can represent any quantum superposition of these states. All cells are initially in the same ground states $`|`$ and the state of the all array is $`|\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\mathrm{}`$, similar to an one-dimensional two-sublattice ferromagnetic. Each qubit of quantum information in the state is represented by four consecutive units: the qubit basis state ”0” is represented by unit $`|\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}`$, whilst the state ”1” is represented by $`|\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}`$.
The model of array described below could be realized by using a linear artificial ”molecule” with A and B cells alternating along its length in antiferromagnetic-type structure. As the cells in this array are used only identical nuclear spins $`I=1/2`$. The neighbor nuclear spins in the ground state of antiferromagnetic structure are opposite orientated and have distinct resonant frequencies determined by hyperfine interaction constant, by applied magnetic field value and by interaction with the left and right nuclear neighbor spins. The major advantage of this variant over the ferromagnetic structure is that the antiferromagnet doesn’t have the total spontaneous magnetization and the nuclear resonance frequency doesn’t depend on the sample shape.
## 1 The one-dimensional antiferromagnetic model on atoms $`{}_{}{}^{31}𝐏`$.
In it was suggested a bulk-ensemble generalization of the silicon quantum computer model proposed by Kane previously . In ensemble case, unlike the individual Kane’s model, two-type electrodes $`𝐀`$ and $`𝐉`$ form a set of narrow $`(l_\mathrm{A}10\mathrm{nm})`$ and long (several micrometers) strips. The distance between neighbors $`𝐀`$ gates was assumed $`l_\mathrm{x}l_\mathrm{A}`$. Along the gates $`𝐀`$, donor $`{}_{}{}^{31}\mathrm{P}`$ atoms $`l_\mathrm{y}`$ distant from each other are placed. If exchange interaction constant for localized electronic spins along the strip gates is more than for electronic spins between neighboring strips and more than Zeeman energy $`J(l_\mathrm{y})J(l_\mathrm{x}),2\mu _\mathrm{B}B`$ ($`B`$ is the induction of the applied magnetic field), it produces an artificial one-dimensional antiferromagnetically ordered state of electronic spins. At the temperatures well below the critical temperature (Neel temperature) $`T_{\mathrm{NS}}J(l_\mathrm{y})/k`$ ($`k`$ — the Boltzmann constant) we will have a pure macroscopic electronic ground quantum state. Due to hyperfine interaction nuclear spins will be oriented according to the electronic spin direction in the resultant field and will form array with the alternating orientation of nuclear spins. Note, this state is not the true pure nuclear antiferromagnetic state so as long as the phases of distinct nuclear spins at macroscopic distances are not correlated at temperature of order or higher then critical temperature of nuclear magnetic dipole ordering, that is $`T>T_{\mathrm{NI}}(10^610^7)K`$ . However, the phase correlations of near neighbor nuclear spins of course exist.
The nuclear resonant frequencies $`\nu _{\mathrm{A},\mathrm{B}}`$ of neighbor nuclear spins are different for each of the magnetic quasi-one-dimensional subarrays A and B in the chain and depend on the states of neighboring spins. We will take it in the form:
$`\nu _{\mathrm{A},\mathrm{B}}(m_<+m_>)|g_\mathrm{N}\mu _\mathrm{N}B\pm A/2I_\mathrm{n}(m_<+m_>)|/2\pi \mathrm{},`$ (1)
where $`\mu _\mathrm{N}=5.0510^{27}\mathrm{J}/\mathrm{T}`$ is the nuclear magneton, $`A`$ is hyperfine interaction constant, (for $`{}_{}{}^{31}\mathrm{P}`$: $`g_\mathrm{N}=2.26`$, $`A=7.7610^{26}\mathrm{J})`$, $`I_\mathrm{n}`$ — the constant of two neighbor nuclear indirect spin-spin interaction, $`m_<`$ and $`m_>`$ are the magnetic quantum numbers for the left and right spins. The nonsecular part of interaction is neglected here taking in to account that $`g_\mathrm{N}\mu _\mathrm{N}B`$, $`A/2I_\mathrm{n}`$.
Thus we have the one-dimensional homonuclear periodic array of nuclear spins $`I=1/2`$, formed in the one-dimensional antiferromagnet at the applied magnetic field, owing to hyperfine interaction of nuclear spins with the electronic magnetic moments. At magnetic fields $`BA/2g_\mathrm{N}\mu _\mathrm{N}3.5\mathrm{T}`$ and at temperatures $`T10^3\mathrm{K}`$ the nuclear spins $`{}_{}{}^{31}\mathrm{P}`$ have in each subarray almost 100% orientation (2$`\pi \mathrm{}\nu _{\mathrm{A},\mathrm{B}}/kT1)`$, that is they are in ground state. Note, that the using of dynamic methods, such as optical pumping, makes possible the high orientation of nuclear spins also at more large temperatures.
We will estimate here the exchange interaction constant $`J>2\mu _\mathrm{B}B6.510^{23}\mathrm{J}`$, the critical temperature $`T_{\mathrm{NS}}J/k4.5\mathrm{K}`$ and the nuclear spin critical temperature, that is due mainly to the Suhl-Nakamura indirect spin-spin interaction, $`T_{\mathrm{NI}}I_\mathrm{n}/kA^2/Jk10^5\mathrm{K}`$.
Here we shall use for the organization of logic operations the addressing to spin states and qubits, analogously to consideration . We shall consider at first the simple one-dimensional model of the antiferromagnet, in which each cell is represented by the magnetic atom and has one electronic and one nuclear spin 1/2 with hyperfine interaction, similar to the mentioned above artificial molecule of antiferromagnetically ordered donors $`{}_{}{}^{31}P`$ in silicon substrate.
Nuclear spins of identical atoms at $`g_\mathrm{N}\mu _\mathrm{N}B<A/2`$ are oriented according to the electronic spin direction in the resultant field and will form a periodic ground state array of ABAB… type: $`\mathrm{}`$, where $``$ marks the ground state of nuclear spin in an A-site and $``$ — the ground state of nuclear spin in a B-site, that is we have here homonuclear system of spins at two distinct ground states. Each nuclear spin in A-site of this scheme, has two internal eigenstates — ground $`|`$ and excited $`|`$ and in B-site, accordingly, — $`|`$ and $`|`$. We take into account that the life time (the longitudinal or spin-lattice relaxation time $`T_1)`$ of excited states at low temperatures is very long. Each qubit of quantum information in this state will be represented here, similar to , by the four consecutive cells: the logical qubit basis state ”0” will be represented by unit $`|`$, whilst the state ”1” — by $`|`$. It is important here that the resonant frequencies of nuclear spins depend on neighbor spins states.
The input and output of the information in the array of ground states spins could be performed at the ends of the array, where the nuclear spin (say in A-site at the left end) has only one neighbor spin and distinguishing resonant frequency $`\nu _{\mathrm{A},1/2}`$ $`(m_<+m_>=1/2)`$. The corresponding selective resonance RF $`\pi _{\mathrm{A},1/2}`$pulse inverts only one nuclear spin (in A-site) at the end of array and doesn’t influence on any ones. Then the new selective RF $`\pi _{\mathrm{B},0}`$pulse will invert next nuclear spin (in B-site), which has the opposite orientation of ground and exited neighbor nuclear spins ($`m_<+m_>=0`$ in A-site) and consequently the new resonant frequency, distinguished from the frequency of spins with the neighbor nuclear spin in ground states $`(m_<+m_>=1)`$. Thus the qubit state ”0”, that is $`|`$, is formed in the following way (the pulses act on underlined spins):
$$\underset{}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\mathrm{}\stackrel{\pi _{\mathrm{A},1/2}\mathrm{pulse}}{}\underset{\mathrm{A}}{}\underset{}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\mathrm{}\stackrel{\pi _{\mathrm{B},0}\mathrm{pulse}}{}\stackrel{\mathrm{"}0\mathrm{"}}{\overline{\overline{\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}}}}\mathrm{}$$
The qubit state ”1” at the edge of array is formed by means of still three pulses: at first $`\pi _{\mathrm{A},0}`$, then $`\pi _{\mathrm{A},1/2}`$ and $`\pi _{\mathrm{B},0}`$pulses:
$$\overline{\overline{\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{}{}\underset{\mathrm{B}}{}}}\mathrm{}\stackrel{\pi _{\mathrm{A},0}\mathrm{pulse}}{}\underset{}{}\overline{\underset{}{}\underset{}{}}\mathrm{}\stackrel{\pi _{\mathrm{A},1/2}\mathrm{pulse}}{}\overline{\underset{}{}\underset{}{}}\mathrm{}$$
$$\stackrel{\pi _{\mathrm{B},0}\mathrm{pulse}}{}\stackrel{\mathrm{"}1\mathrm{"}}{\overline{\overline{\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}}}}\mathrm{}$$
The states $`|\overline{}`$ and $`|\overline{}`$ may be called as the reversed states relative to the states $`|\overline{\overline{}}`$ and $`|\overline{\overline{}}`$.
Note that a random inversion of only one spin will result in completely destruction of the qubit. However, to form, for example, the error of ”0” $`\mathrm{"}1\mathrm{"}`$ type in the coding of stored quantum information it is essential to invert simultaneously four spins. Therefore, it may be concluded that the considered way of qubit coding ensures a better fault-tolerance with respect to this type of errors.
Authors of have considered also another scheme of the four-spin encoding two logical qubits, which are represented by the two zero-total states of four spins, generated by the pairs respectively of the singlet and triplet states. This scheme leads in the collective decoherence conditions to the fault-tolerant implementation of quantum computations. The collective decoherence conditions can be attained in coupled spins at very low temperatures, where all collective but the longest wavelength acoustic phonon modes are quenched.
The further shift-loading of qubit states into the array is implemented by means of pulse sequence $`\pi _{\mathrm{A},0}`$, $`\pi _{\mathrm{B},0}`$, $`\pi _{\mathrm{A},0}`$, $`\pi _{\mathrm{B},0}\mathrm{}`$, which is represented by following SWAP operation:
$$\stackrel{\mathrm{"}1\mathrm{"}}{\overline{\overline{\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{}{}\underset{\mathrm{B}}{}}}}\mathrm{}\stackrel{\pi _{\mathrm{A},0}\mathrm{pulse}}{}\overline{\underset{}{}}\mathrm{}\stackrel{\pi _{\mathrm{B},0}\mathrm{pulse}}{}\overline{\overline{\underset{}{}}}\mathrm{}\stackrel{\pi _{\mathrm{A},0}\mathrm{pulse}}{}$$
$$\overline{\underset{}{}}\underset{}{}\mathrm{}\stackrel{\pi _{\mathrm{B},0}\mathrm{pulse}}{}\stackrel{\mathrm{"}1\mathrm{"}}{\overline{\overline{\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}\underset{\mathrm{A}}{}\underset{\mathrm{B}}{}}}}\mathrm{}$$
and so on.
The role of the atoms at array ends can play here, as it was discussed in , also dopand nuclei D at the certain place of the array with distinct resonant frequency or a defect that modifies the resonant frequency of the nearest nuclear spin in the array. Starting from the perfectly initialized states inputting the information can be performed by setting the dopant D-spin to desired state by means of pulse at his resonant frequency. The nuclear spin state of cell nearest to the dopand is created by SWAP operation mentioned above. After the required information is loaded, D-spin is reset to the ground state $`|0`$. Upon completion of computation, the state of any qubits can be measured by moving it to the A-site nearest to D, then swapping $`A\mathrm{D}`$ and finally measuring the state of D-spin.
## 2 One-qubit operations in one dimension
As in , we introduce still $`\pi _{\mathrm{A},1}`$ and $`\pi _{\mathrm{B},1}`$pulses and operators U<sub>A,1</sub> and U<sub>B,-1</sub>. The last means, that each spin in A- and B-site is subjected to a unitary transform U, which acts on spins in A- and B-sites with resonant frequency corresponding to the both neighbor nuclear spin in the same excited states ($`m_<+m_>=\pm 1`$, sign ”$`+`$” is for excited neighbor spins in B-sites $`|`$, and sign ”$``$” — for A-sites $`|`$, see Table). Note, that when operator $`\mathrm{U}`$ is a simple inversion, the actions of U<sub>A,1</sub> and U<sub>B,-1</sub> are identical to $`\pi _{\mathrm{A},1}`$ and $`\pi _{\mathrm{B},1}`$pulse.
Table. The $`\pi `$pulses for spins in A- and B-sites
| Neighbor spin states. A-site | $`\underset{\mathrm{A}}{}`$ | $`\underset{\mathrm{A}}{}`$ | $`\underset{\mathrm{A}}{}`$ | $`\underset{\mathrm{A}}{}`$ | $`\underset{\mathrm{A}}{}`$ |
| --- | --- | --- | --- | --- | --- |
| Resonance frequency | $`\nu _\mathrm{A}(1/2)`$ | $`\nu _\mathrm{A}(1)`$ | $`\nu _\mathrm{A}(0)`$ | $`\nu _\mathrm{A}(0)`$ | $`\nu _\mathrm{A}(1)`$ |
| $`\pi `$pulses | $`\pi _{\mathrm{A},1/2}`$ | $`\pi _{\mathrm{A},1}`$ | $`\pi _{\mathrm{A},0}`$ | $`\pi _{\mathrm{A},0}`$ | $`\pi _{\mathrm{A},1}`$ |
| Neighbor spin states. B-site | $`\underset{\mathrm{B}}{}`$ | $`\underset{\mathrm{B}}{}`$ | $`\underset{\mathrm{B}}{}`$ | $`\underset{\mathrm{B}}{}`$ | $`\underset{\mathrm{B}}{}`$ |
| --- | --- | --- | --- | --- | --- |
| Resonance frequency | $`\nu _\mathrm{B}(1/2)`$ | $`\nu _\mathrm{B}(1)`$ | $`\nu _\mathrm{B}(0)`$ | $`\nu _\mathrm{B}(0)`$ | $`\nu _\mathrm{B}(1)`$ |
| $`\pi `$pulses | $`\pi _{\mathrm{B},1/2}`$ | $`\pi _{\mathrm{B},1}`$ | $`\pi _{\mathrm{B},0}`$ | $`\pi _{\mathrm{B},0}`$ | $`\pi _{\mathrm{B},1}`$ |
Let us construct now the logical gates of quantum computer. At first we shall investigate at first the scheme for one-qubit gate. The considered section of the array (Fig. 1) contains three qubits in states ”1”, ”0” and ”1”, each being separated by number multiple four of spacer cells -nuclear spins in ground state. Therefore each qubit requires a total of eight physical spins in the array (four for the encoding plus four spacers) (Fig. 1). As it was made in , we will also introduce here one control unit, (not to be confused with control qubit in case of CNOT gates), which is represented here by six consecutive cells in the pattern $`\overline{\overline{}}`$. The control unit (CU) exists only in one place along the array and is separated by odd number of spacer cells — spins (the scheme at Fig. 1 has three).
Fig. 1. The scheme of the SWAP update pulse sequence.
The applying SWAP update sequence of pulses $`\pi _{\mathrm{A},0}`$, $`\pi _{\mathrm{B},0}`$, $`\pi _{\mathrm{A},0}`$, $`\pi _{\mathrm{B},0}`$, $`\pi _{\mathrm{A},0}\mathrm{}`$ moves the qubits to the right and CU to the left relative to the qubits, yet the form of the qubits and the CU are preserved. The CU passes through qubit in state ”1” and ”0”, leaving it unchanged and continues further (Fig. 1).
To implement the one-qubit logical gate the additional computing updates six pulse sequence $`\pi _{\mathrm{A1}}`$, $`\pi _{\mathrm{B1}}`$, $`\pi _{\mathrm{B0}}`$, $`\pi _{\mathrm{A1}}`$, $`\pi _{\mathrm{B0}}`$, U<sub>A,1</sub> is applied when CU reaches the mid-way through passing the qubit $`\mathrm{Y}`$ (”1” or ”0”, marked * and \** at Fig. 1). The effect of the additional sequence is to apply a unitary transform $`\mathrm{U}`$ only to the spin representing the qubit $`\mathrm{Y}`$: $`\mathrm{t}=\mathrm{UY}`$ (Fig. 2). The scheme of the additional computing update pulse sequence after stage * is shown at Fig. 2. The scheme of sequence after stage \** is shown in Appendix A1.
Fig. 2. The scheme of the additional computing update pulse sequence after $`\mathrm{stage}`$
Let us take now the one-cell unitary operation at the end of additional sequence:
$`\mathrm{U}_{\mathrm{A},1}|=a|+b|,|a|^2+|b|^2=1.`$ (2)
Re-applying the update pulses after unitary operation in reverse order the CU moves away from transformed cell and is returned to its initial state (see Appendix A2). We will have in the result the superposition of states:
$`|\psi =a|\stackrel{\mathrm{"}1\mathrm{"}}{\overline{\overline{}}}+b|\stackrel{\mathrm{"}0\mathrm{"}}{\overline{\overline{}}}.`$ (3)
The CU and additional computing updates pulse sequence together ensure the computing operations with qubits. Note, such one-qubit gate requires seventeen elementary pulses.
Not let us consider the state of quantum register, shown at the first line on Fig. 1 and apply after stage marked \** the pulse $`\pi _{\mathrm{A},1}`$. The result is presented on Fig. 3.
Fig. 3. The result of the SWAP pulse sequence finished by pulse $`\pi _{\mathrm{A},1}`$.
We see that the CU moves transparently past the qubit ”1” and continues until mid-way through passing qubit ”0”. Now the CU itself is subject to a transformation: it is altered from $`\overline{\overline{}}`$ to $`\stackrel{}{}\stackrel{}{}\stackrel{}{}\stackrel{}{}\stackrel{}{}\stackrel{}{}`$ (only for passing the qubit ”0”!) and qubit ”0” itself will be destroyed (Fig. 3).
Now we will apply again the SWAP pulse sequence and will become:
Fig. 4. The scheme of the SWAP pulse sequence following sequence Fig. 3.
We see here, that by passing qubit ”1” the altered CU is preserved his form.
## 3 Two-qubit operations in one dimension
To implement the two-qubit gate such as CNOT it should be applied the another update pulse sequence.
Let now the qubit ”0” will be as control qubit of CNOT gate. The CU transforms in altered form by passing the control qubit and then we extend the sequence Fig. 5 after stage marked \***** by following pulses with the end inversion pulse U$`{}_{\mathrm{A},1}{}^{}\pi _{\mathrm{A},1}`$:
$`\pi _{\mathrm{A},1},\pi _{\mathrm{B},1},\pi _{\mathrm{B},0},\pi _{\mathrm{A},0},\pi _{\mathrm{A},1},\pi _{\mathrm{B},0},\pi _{\mathrm{A},0},\pi _{\mathrm{B},1},\pi _{\mathrm{A},1}.`$ (4)
We will obtain:
Fig. 5. The scheme of the update pulse sequence (4)
The last inversion operation doesn’t have effect on target qubit ”1”, as it must be for CNOT gate. The reverse sequence returns CU and qubits to their initial states.
Let us return then to sequence Fig. 1 and continue it after stages marked \**, when CU passes mid-way through qubit ”1” by following sequence:
Fig. 6. The scheme of the update pulse sequences after stage \** when CU passes the qubit ”1”
The scheme of the update pulse sequences after stage \**** when CU also passes the qubit ”1” is shown in Appendix A3.
We see, that the last inversion operation has effect on target qubit, as it must be for CNOT gate only when the control qubit is ”1”.
We can consider a large set of quasi-one-dimensional antiferromagnetically ordered weakly coupled at $`J(l_\mathrm{y})J(l_\mathrm{x})`$ arrays donors $`{}_{}{}^{31}P`$ in silicon substrate as an ensemble of artificial molecules. In this case, there is no need to address qubits individually. We suppose that for the determination of nuclear spin states in ensemble of those identical artificial molecules, as in liquid-state bulk-ensemble quantum computer , there are no need to fulfill the electrical measurements and consequently any electrodes. Since the read-out signal in this case will be proportional to the numbers of artificial molecule in the ensemble, it may be used the NMR or fluorescence techniques for ensemble measurement of spin states.
## 4 Two- and three-dimensional antiferromagnetic structures
Instead of generalization to the parallel model employing ”sub-computers” with one-dimensional structure, which was considered in , our approach allows to use also two and three-dimensional structures. The coupled antiferromagnetically ordered chains model can be extended to a two-dimensional antiferromagnetic chess-type ordering. Let the electronic spins of the neighboring chains are setting for $`J(l_\mathrm{y})J(l_\mathrm{x})>2\mu _\mathrm{B}B`$. The electronic spins of two neighbor chains will be in the singlet ground state. The subarray of nuclear spins will have the opposite orientation of nuclear spins relative to the subarray of neighboring chain. The electron subsystem of two neighbor chains is in antiphase state, that is have the half period shift of antiferromagnetically ordered electronic spins in the one chain relative to the other. The nuclear subsystem of the both chains becomes corresponding chess-type ordering (Fig. 7).
Let us suppose that an initial state containing some number of cells is loaded in the two-dimensional structure of many coupled chains. The inputting the information into the cell nearest to dopant atoms D (D-spin) is performed by means of corresponding $`\pi `$pulse: $``$ or $``$. Then, the resonant frequencies of neighbor spins in the same chain and of spins in neighbor chain is altered and another $`\pi `$pulse will invert one of they or both according to the values $`I_1`$ and $`I_2`$ of indirect nuclear spins interaction inside and between the chains. Therefore, the excited nuclear spin states and accordingly the qubits may be passed to any place in all two-dimensional structure. We will suppose as before that the qubits state will be represented by four spin states in chains. The computing operation can be fulfilled analogous to the above-considered one-dimensional scheme.
Fig. 7. The scheme of two-dimensional nuclear spin ordering in antiferromagnetic structure. It is showed the different ways that connect the D-spin (marked $``$) and a certain qubit.
For everyone CU we have a two-dimensional section of array with large enough number of qubits and one dopant atom. It is defined as a single-domain antiferromagnetic sample. There are many ways that connect the D-spin and the qubit (Fig. 7). This section plays role of many-qubit artificial molecule and the whole structure represent a large ensemble of such molecules, which work simultaneously and ensure the parallelism of quantum operations.
The structures with two and three-dimensional antiferromagnetic and ferrimagnetic order may be found perhaps among the natural rare earth or transition element dielectric compounds.
The electron magnetization of a one subarray of antiferromagnet is defined by expression :
$`2\mu _\mathrm{B}NS_{\mathrm{jz}}=2\mu _\mathrm{B}N(1P(T)\psi )/2,`$ (5)
where for low temperatures in spin-wave approximation
$`P(T)={\displaystyle \frac{1}{(2\pi )^\mathrm{d}}}{\displaystyle \frac{d^\mathrm{d}k}{\mathrm{exp}(ϵ(k)/kT1}}`$ (6)
is the contribution of thermal and $`\psi `$ — of quantum fluctuations, $`ϵ(k)=\sqrt{ϵ_0^2+(Jak)^2}`$ — spin-wave spectrum, $`a`$ — lattice period, Z — number of near neighbors, $`d`$ — dimension of structure.
For antiferromagnetic state of easy-axis type, when the interaction Hamiltonian of two electron spin $`j`$ and $`g`$ for single-axis crystal has the form:
$`H_{\mathrm{j},\mathrm{g}}=J𝐒_\mathrm{j}𝐒_\mathrm{g}J_\mathrm{A}(S_{\mathrm{jx}}S_{\mathrm{gx}}+S_{\mathrm{jy}}S_{\mathrm{gy}}),`$ (7)
where $`J>J_\mathrm{A}>0`$, $`J_\mathrm{A}`$ — anisotropy constant, the contribution of quantum fluctuation, as shown in , $`\psi =0`$. In addition, for $`k0`$, $`Tϵ_0`$, where $`ϵ_0\mathrm{Z}\sqrt{JJ_\mathrm{A}}`$,
$`P(T)\mathrm{Const}(kTϵ_0/J^2)^{\mathrm{d}/2}\mathrm{exp}(ϵ_0/kT)0,`$ (8)
that is in this state the thermal fluctuations in electron system also are not essential for the ground electron spin states.
Note, that in the case of easy-flat state, $`\psi 0`$, but the NMR resonance frequency depends on the neighbor nuclear states to a greater extent than in case of easy-axis state .
The quantum state decoherence at low temperatures is defined on the one hand by the active role of electron spin-wave effects . They generate the fluctuated local field due to Raman process of the electron spin wave scattering on individual nuclear spins. The decoherence time or transverse relaxation time $`T_2`$ of NMR in antiferromagnet for the low temperatures $`(ϵ_0/kT1)`$ then is determined by expression
$`1/T_2(A^2/J)(kT/J)^3(ϵ_0/kT)\mathrm{exp}(ϵ_0/kT)/\pi ^2\mathrm{}0,`$ (9)
value $`T_2`$ rapidly grows.
On the other hand decoherence is defined by inhomogeneity of the local magnetic fields and spread in resonance frequencies. The nuclear spin-spin interaction in natural dielectric antiferromagnets is defined mainly by the Suhl-Nakamura indirect mechanism of interaction through exchange of spin waves and is typically greater than the value, determined by the direct nuclear spin-spin dipole interaction. This interaction of nuclear spins could play a large role in the case of high spin concentration. Both of these decoherence mechanisms can be, in principle, suppressed by some NMR many-pulse methods using the stroboscopic observation of spin dynamics .
The general requirements for natural antiferromagnetic structures, required for the construction of NMR quantum computers, can be formulated in the following way:
1) The operating temperature $`T`$ must correspond to the fully ordered antiferromagnet $`T_{\mathrm{NS}}TT_{\mathrm{NI}}`$ and to fully polarized nuclear spins $`T_{\mathrm{NS}}A/JA/k>TT_{\mathrm{NI}}`$. From where we will have the value $`T10^3\mathrm{K}`$.
2) The two-dimensional and tree-dimensional magnetic structure must have chess-type order (see Fig. 7).
3) The magnetic structure must have the easy-axis state of antiferromagnetism in single-axis crystals.
4) The atoms must have nuclear spins $`I=1/2`$. Electron spins may be $`S1/2`$.
There are the rare earth compounds of unique thulium stable isotope $`{}_{}{}^{169}\mathrm{Tm}`$, that has nuclear spin $`I=1/2`$, $`g_\mathrm{N}=0.458`$ and makes up 100% of naturally occurring elements with stable spinless isotopes of other elements. They can be: $`\mathrm{Tm}_2\mathrm{O}_3`$, $`\mathrm{TmSi}_2`$, $`\mathrm{TmGe}_2`$ and $`\mathrm{TmSe}`$ . The ground electronic state of magnetic ions $`\mathrm{Tm}^{3+}`$ corresponds to $`S=1`$. The natural elements O, Si, Ge and Se have, accordingly, nuclear spin containing isotopes (in brackets it is shown the isotope occurrence) $`{}_{}{}^{17}\mathrm{O}`$ $`I=5/2`$ (0.04%), $`{}_{}{}^{29}\mathrm{Si}`$ $`I=1/2`$ (4.7%), $`{}_{}{}^{73}\mathrm{Ge}`$ $`I=9/2`$ (7.76%), $`{}_{}{}^{77}\mathrm{Se}`$ $`I=1/2`$ (7.78%). For $`\mathrm{Yb}_2\mathrm{O}_3`$ $`T_{\mathrm{NS}}2.3\mathrm{K}`$, isotope $`{}_{}{}^{171}\mathrm{Yb}`$ (14.31%) has $`I=1/2`$ and $`S^{}=1/2`$ (ground state is Kramers doublet). It is known, that compound $`\mathrm{TmSe}`$ has the critical temperature for antiferromagnetic transition $`T_{\mathrm{NS}}2K\text{[17]}`$. The antiferromagnets with two different nuclear spin $`I=1/2`$, for example $`\mathrm{FeF}_2`$ with rutile-type and $`\mathrm{TmAg}`$ with $`\mathrm{CsCl}`$type structure, which have critical temperature 79 K and 9.5 K, may be also of interest to the considered questions. Isotopes $`{}_{}{}^{57}\mathrm{Fe}`$ (2.19%), $`{}_{}{}^{19}\mathrm{F}`$ (100%), $`{}_{}{}^{107+109}\mathrm{Ag}`$ (100%) have according values $`g_\mathrm{N}=0.182`$, $`g_\mathrm{N}=5.26`$ and $`g_\mathrm{N}=0.24`$.
In conclusion, we will point out the several advantages of the considered model: it uses the antiferromagnetic structure containing only one type of atoms with nuclear spin 1/2, it is not needed to have any gate electrodes, the decoherence associated with noise voltage is absent, the considered way of qubit coding ensures a better fault-tolerance with respect to the generation of wrong qubits, the model admits an ensemble address qubits, it may be used as base for development of bulk-ensembles three-dimensional solid-state NMR quantum computer.
The author is grateful to K.A.Valiev for critical reading of the article and useful remarks and V.A.Kokin for the help in preparation of this text.
## Appendixes
### A1. The scheme of the additional computing update pulse sequence after stage \** at Fig. 1:
### A2. The scheme of the reverse update pulse sequence after one-qubit operation U<sub>A,1</sub>:
### A3. The scheme of the update pulse sequences when CU passes the qubit ”1” after stage \****:
|
warning/0002/astro-ph0002057.html
|
ar5iv
|
text
|
# The Warp of the Galaxy and the Orientation of the LMC Orbit
## 1 Introduction
The disk of the Milky Way is remarkably flat out to 10 kpc, where it starts to bends in opposite directions in the southern and northern parts. The cause of it is still a puzzle: for a review, see Binney \[Binney 1992\].
One possibility is that the Magellanic Clouds distort the Galactic disk in the observed way. The fact that the direction of the maximum warping lies very close to the galactocentric longitude of the Clouds makes this hypothesis tempting.
The problem with this scenario is that the Clouds are not massive enough to generate the warp amplitudes that we observe at their present distance. This was noticed by the first researches in this field \[Burke 1957, Kerr 1957\], and later by Hunter & Toomre (1969). A remedy which might allow this scenario to work was to suppose that the Clouds are currently not at the pericenter of their orbit, so that they have been much closer to the Galactic disk in a previous epoch ($`20`$ kpc is what was needed). However, later work by Murai & Fujimoto (1986) determined the orbit of the Clouds, and proved that the Clouds are actually at their pericenter, with an apocenter close to 100 kpc, so the problem of the small amplitude still remains if the Clouds are to be blamed as the cause of the Galactic warp.
Recently a mechanism for amplifying the effect of a satellite has been proposed by Weinberg \[Weinberg 1998\]. He describes a calculation in which a disk galaxy surrounded by a dark halo is perturbed by a massive satellite, similar to the LMC. By means of a linear perturbation analysis, he follows the perturbation (wake) created by the satellite in the halo, including its self-gravity. He finds that the torque exerted on the disk is several times larger than that due directly to the satellite: the latter is amplified because (i) the satellite-induced wake in the halo itself exerts a torque, roughly in phase with that from the satellite; and (ii) the wake itself further perturbs the halo, resulting in a torque that is larger again. Under circumstances in which the satellite orbital frequency is close to the natural precession frequency of the disk (i.e., the warp mode frequency of Sparke & Casertano \[Sparke & Casertano 1988\]), a significant amplitude can result. A calculation by Lynden-Bell \[Lynden-Bell 1985\] of a similar scenario gives comparable results, as does a simple model described by Kuijken \[Kuijken 1997\].
In this paper, we focus on the orientation of a warp generated by a massive orbiting satellite with less emphasis on the amplitude of the warp. In §2 we discuss a simple analytic model in which the disk and halo are rigid: this establishes the baseline response of a disk to satellite tides, and its orientation with respect to the satellite orbital plane. As we show, this orientation is different from that of the Galactic warp to the LMC orbit. §3 contains a description of the N-body code used, and §4 to §6 the results of our N-body simulations, showing that the orientation problem remains. In §7 we give our conclusions.
## 2 Analytic results with a simplified model
A simple model serves as a reference for the warp response of the disk to an orbiting satellite. Consider a rigid disk, embedded in a rigid halo potential, and subjected to the potential of an orbiting satellite. The evolution of the disk is governed by the combined torque from halo and satellite. A stellar or gaseous disk is floppy, and so will warp when tilted, since it is not able to generate the stresses that would be required to keep it flat; however the overall re-alignment of the disk angular momentum should be comparable between the rigid and floppy cases.
The angles used in this paper related to the satellite, and the definition of our coordinate system are illustrated in Figure 1. The tilting of the disk is measured by the angle between the $`z`$ axis and the angular momentum of the disk. The longitude of this vector is the same as the longitude of the maximum of the warp when looking at the disk from the North Galactic Pole.
The Lagrangian for a rigidly spinning, axisymmetric object is
$$=\frac{1}{2}I_1(\dot{\theta }^2+\dot{\varphi }^2\mathrm{sin}^2\theta )+\frac{1}{2}I_3(\dot{\varphi }\mathrm{cos}\theta +\dot{\psi })^2V(\theta ,\varphi )$$
(1)
where $`(\theta ,\varphi ,\psi )`$ are the Euler angles, and $`I_3`$ and $`I_1`$ are the moments of inertia of the object about its symmetry axis and about orthogonal directions. $`V`$ is the potential energy of the body in the halo plus satellite potential. The $`\psi `$-equation of motion leads to the conserved quantity $`S=I_3(\dot{\varphi }\mathrm{cos}\theta +\dot{\psi })`$, the spin, and the other two equations of motion then become
$$I_1\ddot{\theta }I_1\dot{\varphi }^2\mathrm{sin}\theta \mathrm{cos}\theta +S\dot{\varphi }\mathrm{sin}\theta +\frac{V}{\theta }=0$$
(2)
and
$$I_1\frac{\mathrm{d}}{\mathrm{d}t}(\dot{\varphi }\mathrm{sin}^2\theta )+\frac{V}{\varphi }=0.$$
(3)
For small deviations from the equator ($`\theta =0`$), we can expand these equations in terms of $`x=\mathrm{sin}\theta \mathrm{cos}\varphi \theta \mathrm{cos}\varphi `$, $`y=\mathrm{sin}\theta \mathrm{sin}\varphi \theta \mathrm{sin}\varphi `$. In these terms the equations of motion become
$$I_1\ddot{x}+S\dot{y}+\frac{V}{x}=0,$$
(4)
$$I_1\ddot{y}S\dot{x}+\frac{V}{y}=0.$$
(5)
For small $`x,y`$, the potential energy of the disk due to the flattened halo will have the form $`\frac{1}{2}V_\mathrm{H}(x^2+y^2)`$, and that due to the satellite at position $`\theta _\mathrm{S},\varphi _\mathrm{S}`$ will be $`V_\mathrm{S}(\mathrm{sin}^2\theta _\mathrm{S}x\mathrm{sin}2\theta _\mathrm{S}\mathrm{cos}\varphi _\mathrm{S}y\mathrm{sin}2\theta _\mathrm{S}\mathrm{sin}\varphi _\mathrm{S})`$ where $`V_\mathrm{H}`$ and $`V_\mathrm{S}`$ are constants representing the strengths of the halo torque and of the quadrupole of the tidal field from the satellite, respectively. Hence we find
$$I_1\ddot{x}+S\dot{y}+V_\mathrm{H}x+V_\mathrm{S}\mathrm{sin}2\theta _\mathrm{S}\mathrm{cos}\varphi _\mathrm{S}=0,$$
(6)
$$I_1\ddot{y}S\dot{x}+V_\mathrm{H}y+V_\mathrm{S}\mathrm{sin}2\theta _\mathrm{S}\mathrm{sin}\varphi _\mathrm{S}=0.$$
(7)
If furthermore the satellite orbit is circular and polar in the $`x=0`$ plane, $`\theta _\mathrm{S}=\mathrm{\Omega }_\mathrm{S}t`$, $`\varphi _\mathrm{S}=90`$, and the solution to the equations of motion is
$$x=\frac{2\mathrm{\Omega }_\mathrm{S}S}{\mathrm{\Delta }}V_\mathrm{S}\mathrm{cos}2\mathrm{\Omega }_\mathrm{S}t;y=\frac{4I_1\mathrm{\Omega }_\mathrm{S}^2V_\mathrm{H}}{\mathrm{\Delta }}V_\mathrm{S}\mathrm{sin}2\mathrm{\Omega }_\mathrm{S}t$$
(8)
plus free precession and nutation terms, where $`\mathrm{\Delta }=(V_\mathrm{H}4I_1\mathrm{\Omega }_\mathrm{S}^2)^24\mathrm{\Omega }_\mathrm{S}^2S^2`$. (A more general quasiperiodic satellite orbit yields a solution which can be written as a sum of such terms.) Notice that the satellite provokes an elliptical precession about the halo symmetry axis, with axis ratio dependent on the halo flattening and on the satellite orbit frequency. For example, for an exponential disk of mass $`M`$, scale length $`h`$ and with a flat rotation curve of amplitude $`v`$, $`I_3=2I_1=6Mh^2`$ and $`S=2hvM`$. For such a disk in a spherical (or absent) halo ($`V_\mathrm{H}=0`$), a satellite orbiting at radius $`r_\mathrm{S}`$ has frequency $`\mathrm{\Omega }_\mathrm{S}=v/r`$, and hence the axis ratio of the forced precession is $`(x:y)=r_\mathrm{S}/3h`$. Hence the response of the disk to a distant satellite is mainly to nod perpendicular to the satellite orbit plane. This result can be understood as the classic orthogonal response of a gyroscope to an external torque: a distant satellite has a sufficiently low orbital frequency that the disk responds as if the torque were static.
For a slightly flattened potential of the form $`\frac{1}{2}v^2\mathrm{ln}[R^2+(z/(1ϵ))^2]`$, $`V_\mathrm{H}=Mv^2ϵ`$. With non-zero $`ϵ`$, the axis ratio of the precession cone becomes $`[(4h/r_\mathrm{S})/(ϵ12h^2/r_\mathrm{S}^2)]`$: again the oscillation in $`x`$ is larger than that in $`y`$ except for very flattened halos.
The amplitudes generated by tidal perturbation from a satellite such as the LMC are small, less than a degree. The largest amplitude of oscillation is in the $`y`$-direction. The potential energy of the disk due to the tidal field of the satellite can be shown to be (see Appendix)
$$V_\mathrm{S}=\frac{3GM_\mathrm{S}I_1}{2r_\mathrm{S}^3}.$$
(9)
Hence equation 8 yields, to leading order in $`h/r_\mathrm{S}`$, an $`x`$-amplitude of
$$\frac{9}{8}\frac{GM_\mathrm{S}}{v^2r_\mathrm{S}}\frac{h}{r_\mathrm{S}}0.09^{}$$
(10)
for the LMC (orbital radius about $`50\mathrm{k}pc`$, and $`r_\mathrm{S}/h15`$). This number increases only slightly (a factor 2) for halo flattenings up to 0.2 (see Figure 2).
It is clear from this calculation that simple tidal tilting of a disk by an LMC-like satellite does not provide a good model for the warping in the Galaxy, because the orientation of the warp is not perpendicular to the orbital plane of the LMC. This constraint is independent of the strength of the perturbation $`V_\mathrm{S}`$.
The amplitudes are much too small, but we have only considered the tilting of a rigid disk, and the situation can change when the floppiness of the disk is considered.
## 3 Simulation details
To test this scenario, and in particular to get beyond the rigid tilting considered above, we have performed some N-body simulations. We assume the halo to be a background potential which does not respond to the disk or the satellite.
The N-body code used for this work models the disk as a set of concentric spinning rings embedded in a spherical, rigid halo. This description allows warps to be described, but not in-plane distortions of the disk such as bars or lopsidedness.
### 3.1 Initial conditions
We have performed simulations with two types of disks: a rigid disk and a exponential disk. The rigid disk run tells us how good the analytic predictions are, and the exponential disk is used later for a more realistic approach.
We use a King model for the halo, in order to obtain a reasonable flat rotation curve (Figure 3). This is accomplished with a model of $`\mathrm{\Psi }_0/\sigma _{0}^{}{}_{}{}^{2}=6`$, a tidal radius of 44 (200 kpc. for a 4.5 kpc scale-length disk), and mass of 10 disk masses.
Each of the disks is made of 1000 rings, each of them consisting of 36 particles. Various runs where made with more rings and more particles per ring, without important changes in the results described below.
The satellite is modelled as having a Plummer distribution. To avoid relative movements of the galaxy with respect to the satellite we have used two satellites instead of one, symmetrically placed with respect to the centre of the halo-disk system. This causes the dipole term of the tidal field to be zero, avoiding relative movements of the galaxy with respect to the satellites. It is equivalent to only keeping the even-$`l`$ terms in the potential of the satellite, neglecting the dipole, $`l=1`$, terms in the potential, and concentrating on the warp (which result from the quadrupole, $`l=2`$ terms).
The first run was made with a satellite in a circular orbit, to try to reproduce the predictions in §2. Later a non-circular orbit is considered, and the difference between both simulations analysed. The non-circular orbit has a pericenter at 50 kpc and apocenter at 100 kpc, consistent with recent determination of the orbit of the Clouds \[Lin, Jones & Klemola 1995\]. In the non-circular simulations the satellite starts at its apocenter at the beginning of the simulation, where the perturbation on the disk is the smallest possible.
The units of the model translate to the Galaxy (disk scale-length $`4.5\mathrm{k}pc`$, and the rotation velocity at $`8.5\mathrm{k}pc`$ of $`220\mathrm{k}ms^1`$) as follows: length unit = $`4.5\mathrm{k}pc`$, velocity unit = 340 $`\mathrm{k}ms^1`$, time unit = $`1.30\times 10^7`$years, mass unit= $`1.20\times 10^{11}\mathrm{M}_{}`$. With these numbers, the disk mass of our model is $`6.1\times 10^{10}M_{}`$, and the satellite (LMC) has a mass of $`1.5\times 10^{10}M_{}`$, the biggest current mass estimate for the Clouds \[Schommer et al. 1992\]. In the coordinate system of the simulations, the $`z=0`$ is the disk plane, and the orbit of the satellite lies in the $`x=0`$ plane.
### 3.2 Code used to evolve the system
The disk is modelled as a system of pivoted spinning rings, fixed at the centre of the halo. Each ring is realized as 36 azimuthally-spaced particles, and the potential generated by the rings is calculated with a tree code \[Barnes & Hut 1986\]. The forces on the individual “ring-particles” are used to calculate the torque on each ring. The force exerted by a satellite on the ring particles was evaluated directly using the Plummer law.
The Euler equations for rings and axisymmetric bodies can be rewritten in a form so that the time derivatives of the instantaneous angular velocities about the body axes are linear combinations of the angular velocities, torques and body normal vector components. This allows the derivation of a second order explicit time-centred leapfrog integration scheme that can be used to solve the coupled equations for the rings and make it easy to merge with an N-body code \[Dubinski 2000\].
## 4 Rigid Disk
As a first approach, we have evolved a rigid disk and analysed its evolution under the influence of an orbiting satellite. The result of our simulation is in good agreement with the analytic predictions. The disk wobbles under the influence of the satellite, describing an ellipse elongated in the direction perpendicular to the satellite’s orbital plane. The path followed by the disk is plotted in Figure 4, where it can be seen that most of the time the maximum of the warp is located in the direction perpendicular to the orbital plane of the satellite ($`l=0^{}`$ and $`l=180^{}`$). The ellipse isn’t as regular as in Figure 2 for two reasons: the assumption that the disk is much smaller than the orbital radius of the satellite is not completely fulfilled; and there are some transient terms present because of the initial conditions of the simulation. This are also the cause for the precession ellipse of the disk not to be centred in the origin.
The position of the warp when the satellite is at the location of the LMC is indicated by the dots in Figure 4, and the location of them resembles the predicted one in Figure 2 (for $`ϵ_{halo}=0`$) remarkably well.
## 5 Exponential self-gravitating Disk
We now consider a more realistic disk: an exponential disk model, in which we have considered also the disk’s self-gravity. The first thing that draws our attention in this simulation is a peak we see in the inclination at around 6.5 scale-lengths. Simulations done with a different rotation curve showed that this peak occurs at the locations on the disk that satisfy $`\mathrm{\Omega }_s/w_z=2,3,\mathrm{}`$, that are caused by resonances with the satellite’s orbital frequency. This happens because the non-linear behaviour of the outer parts of the disk, where the assumption $`r_sr_{disk}`$ is worse than it is further in.
This is not the kind of warp we are looking for, due to the fact that it is the result of a satellite with a single frequency, and in the real case the eccentric orbit of the satellite will wash out this peak. Looking at the evolution of the disk it is clear that the warp looses its coherence at a radius about 4.5 scale-lengths (at larger radius the line of nodes winds up), so we will measure the warp properties considering that the disk finishes there.
In the case of a floppy disk it is not straight-forward to define a single inclination and position angle. We have separated the disk into two components: the inner disk and the outer (warped) disk. The inner disk consists on the first 2 scale-lengths, and remains practically flat along the simulation. The warping angle is then calculated as the angle between the inner and outer disk vectors. We have chosen to use the disk vectors and not the angular momentum, for example, not to penalise the outer less massive rings. The results presented here do not change significantly when the definition of the inner disk is altered.
It has to be kept in mind that the warping angles quoted here are different than the maximum amplitude of the warp, who usually are larger by a factor not greater than 5.
Using this method we obtain a plot similar to Figure 4 for the exponential disk, which is shown in Figure 5. Only the path after t=160 is shown, that is the moment when the disk behaviour reaches an equilibrium.
Note that the predictions for the Galactic Warp don’t really change with the floppiness of the disk: it is clearly close to $`l=0^{}`$, as chapter §2 predicted, and not at $`l=90^{}`$, as we observe it in the Galaxy.
## 6 Non-circular orbit, and flattened halos
We also considered non-circular orbits, to allow for the fact that the orbit derived for the Clouds has a pericenter of 50 kpc and an apocenter of 100 kpc \[Murai & Fujimoto 1980\]. The changing radius of the satellite causes a fluctuating tidal field amplitude, which could be important for the dynamics of the disk. Here we show that in fact the effect does not change our conclusions materially.
First, to have an idea of what to expect, we integrated the analytic equations of section §2 with a satellite in this kind of orbit. The result was, as before, that the disk’s precession path was contained within an ellipse, elongated along the direction perpendicular to the satellite’s plane. This causes the warp maxima to be most of the time close to the direction perpendicular to the satellite’s orbit.
We then performed simulations with this type of orbits. The first thing we observe in these simulations is that the resonance peak we found in the circular orbit simulation has disappeared. Now the satellite doesn’t have a single frequency, so the result is not surprising. The energy of the resonance gets distributed along different parts of the disk now, and no coherent pattern can be maintained across the disk, winding up the outer parts of the disk. When we look at the inner 4.5 scale-lengths as before, the precession pattern remains similar to the simulation with the circular orbit, so does the prediction of the warp’s longitude at LMC’s actual orbital phase. So our conclusions are not modified by the non-circularity of the orbit.
The halos considered in all these simulations are spherical, which means that they don’t contribute to the generation of torques on the disk. We know that halos are not spherical, which creates a preferred plane in which the disk settles. Ellipticities of the order of 0.05 in the potential make the precession paths described before yet more elongated, which would make the chances of finding the warp maxima in the satellites’ direction even more unlikely.
## 7 Conclusion
We show by means of analytic work and N-body simulations that the precession path of a warp generate by an orbiting satellite galaxy is elongated along the direction perpendicular to the satellite’s orbital plane. Applying our result to the Milky Way, if the Galactic Warp is generated by the Magellanic Clouds, the direction of maximum amplitude of the warp would lie close to $`l0^{}`$, as compared to the observed direction of $`90^{}`$. Even if the halo’s self-gravitating tidal response to the satellite amplifies the effect of the satellite \[Weinberg 1998\], this response will be mostly in phase with the satellite, and the alignment problems will persist. Possibly the Sagittarius dwarf galaxy, whose orbit lies at 90 to that of the LMC, can be the cause of the warp instead \[Ibata & Razoumov 1998\].
A limitation of the present work is that the halo has not been considered as a live, self-gravitating component. It has been shown \[Dubinski & Kuijken 1995, Nelson & Tremaine 1995\] that the back-reaction of the halo on a re-aligning disk can have important consequences. Such effects will be the subject of a further paper.
## Appendix A Potential of axisymmetric disk due to a satellite
The potential energy of an disk of surface density $`\mathrm{\Sigma }(r)`$ and in the gravitational field due to a satellite at position $`𝒓_\mathrm{S}`$ is given by
$$V=\mathrm{d}^2𝒓G\mathrm{\Sigma }(r)\frac{M_\mathrm{S}}{|𝒓𝒓_\mathrm{S}|}.$$
(11)
Choosing spherical coordinates for the satellite’s position (see Figure 1), and Cartesian coordinates in the disk plane so that the satellite has $`x=0`$, we have
$$V=GM_\mathrm{S}\mathrm{\Sigma }dxdy(r_\mathrm{S}^22yr_\mathrm{S}\mathrm{sin}\theta _\mathrm{S}+x^2+y^2)^{1/2}.$$
(12)
Assuming that the disk is small compared to $`r_\mathrm{S}`$, we can expand the integrand in $`x`$ and $`y`$. For an axisymmetric disk the second-order terms are the first ones that generate a potential gradient: they are
$$V=\frac{GM_\mathrm{S}}{r_\mathrm{S}^3}\mathrm{\Sigma }dxdy[\frac{1}{2}(13\mathrm{sin}^2\theta _\mathrm{S})y^2\frac{1}{2}x^2]$$
(13)
which results in
$$V=\frac{3GM_\mathrm{S}I_1}{2r_\mathrm{S}^3}\mathrm{sin}^2\theta _\mathrm{S}+\text{constant}.$$
(14)
|
warning/0002/nlin0002033.html
|
ar5iv
|
text
|
# References
§1. Introduction
In this paper, we consider an immune network dynamical system model with small degrees of freedom.
First, we explain the present understanding of immune systems briefly.
The main constituents of an immune system are B-lymphocytes(B-cells) produced in the Bone marrow, T-lymphocytes(T-Cells) produced in the Thymus and free antibodies produced by B-cells. B-cells and T-cells have protein molecules called receptors on their surfaces. The receptors of B-cells are antibodies(Immunoglobulin, Ig), and antibodies recognize and connect to antigens to neutralize them. On the other hand, T-cell receptors(TcR) cannot recognize antigens, but they recognize pieces of antigens which appear on the surfaces of antigen presenting cells. When this happens, as a result, the helper T-cell expedites the immune response and the suppressor T-cell suppresses it. The killer T-cell attacks and kills a cell which is infected by viruses et.al. The receptors of B- and T-cells have proper 3-dimensional structures and these are called ’Idiotypes’. A family of B-cells which are generated from a B-cell are called ’clones’. Therefore, a clone and antibodies produced by the clone have the same idiotype.
In a human body, the total number of clones which are generated from a single B-cell are about 10 to $`10^4`$, the total number of clones amounts to $`10^8`$ and the number of antibodies is about $`10^{20}`$. Thus, the repertoire of antibodies are enough to bind to any antigen. This diversity is due to the recombination and the mutation of genes.
The response to the invasion by antigens is considered as follows. When antigens enter into a body, clones which can recognize the antigen bind to it, and maturate by the help of the helper T-cells, and a part of them proliferate. Others become antibody forming cells. In the antibody forming cells, many antibodies are produced and secreted. As a result, many antibodies appear and neutralize antigens. When the neutralization completes, by the action of the suppressor T-cells, the proliferation of B-cells is suppressed and the immune response ends.
Further, in the course of the division of B-cells a part of each B-cells is preserved as a memory B-cell. When the same antigens enter into the body again, these cells rapidly differentiate into antibody forming cells and produce many antibodies in a short time. This phenomenon is called ’the secondary immune response’.
In the mechanism of the immune response explained above, a clone of the B-cells which can recognize an antigen is selected. Thus, this theory is called ’clonal selection theory’ and has been confirmed experimentally.
B-cells, T-cells and antibodies die if they are not stimulated. As is mentioned above, in a human body there are a huge number of these cells. To explain this, in 1974 N. K. Jerne proposed the so-called network view of the immune systems. In his theory, these cells interact with and activate each other organizing a network.
However, in 10 years after Jerne’s theory appeared, the network theory was considered to fail to live up to its initial promise. This is because the theory cannot explain the correct direction of change of the system when antigens invade a body. Another reason is that since T-cells and their actions were discovered, to include these elements the original theory would lose the simplicity which attracted many researchers.
However, there are several experiments to support the network theory. For example, in new born mice activated lymphocytes are retained although mice are isolated from any antigen. That is, it seems that the immune system is activated by itself to prepare for the invasion of external antigens.
Thus, taking into account T-cells and their roles F. J. Varela et.al. proposed the second generation immune network model in 1991. Since then, this theory has been developing .
We would like to investigate the effects of the interaction among lymphocytes and antigens, and to analyze what kind of states and structures the network can have, and to see which directions the network moves to when antigens invade the system. Also, we have interest in mathematical structures of network systems from the view point of dynamical systems.
In this paper, we study the dynamical system model of immune networks introduced by Varela et.al. In this model the essential characteristics of the real immune systems are taken into account. That is, not only antibodies, but also B-lymphocytes which produce antibodies, and the roles of T-lymphocytes, i.e., the activation and the suppression of B-lymphocytes, are included.
In reality, an immune system has a huge number of degrees of freedom. However, in this paper we focus on the Varela model with small number of degrees of freedom. Our objective is to investigate the possible states in the network and the change of these states when antigens invade the system in small systems, as a necessary step before studying the states of immune network and the immune response in large systems. Let us explain the model we treat in this paper. The constituents of the network, free antibodies and B-lymphocytes(B-cells), interact with each other through idiotypes. Let us distinguish idiotypes by index $`i`$. Between two different idiotypes $`i`$ and $`j`$, there may occur an affinity, which is represented by the connectivity $`m_{ij}`$. We assume $`m_{ij}=1`$ if there is an affinity between $`i`$ and $`j`$ and $`m_{ij}=0`$ if not. $`m_{ij}`$ is measurable by experiments. Let us denote the concentration of B-lymphocytes with the $`i`$-th idiotype by $`b_i`$ and that of free antibodies produced by the B-lymphocytes by $`f_i`$. These antibodies have the same idiotype as the B-lymphocytes. The sensitivity of the network for the $`i`$-th idiotype is defined as follows;
$`\sigma _i={\displaystyle \underset{j=1}{\overset{N}{}}}m_{ij}f_j,`$ (1)
where $`N`$ is the number of idiotypes. It represents the strength of the influence by other antibodies to the $`i`$-th antibody. The number of B-lymphocytes and antibodies change in time by the following causes. Free antibodies are removed from the constituents of the network because they have a natural lifetime and also they interact with other idiotypes and are neutralized. On the other hand they are produced by B-cells as a result of the maturation of B-cells. The probability of the maturation is assumed to depend on their sensitivity $`\sigma `$. This effect is expressed by the function $`Mat(\sigma )`$. In the beginning of immune response, antibodies which interact with antigens maturate with help of T-cells. Then, it is natural to assume that the function $`Mat(\sigma )`$ is increasing with respect to $`\sigma `$ when $`\sigma `$ is small. If the number of antibodies becomes large, and the immune response comes to end, the creation of antibodies will be suppressed. Thus, for large values of $`\sigma `$ $`Mat(\sigma )`$ should be decreasing with respect to $`\sigma `$. Thus, $`Mat(\sigma )`$ is assumed to have the convex profile illustrated in Fig.1.
Then, a differential equation describing the change in time of the concentration $`f_i`$ of the $`i`$-th antibody can be written as
$`{\displaystyle \frac{df_i}{dt}}`$ $`=`$ $`K_1\sigma _if_iK_2f_i+K_3Mat\left(\sigma _i\right)b_i,`$ (2)
where $`K_1`$ is the rate of the neutralization by other antibodies, $`K_2`$ is the rate of the death of the antibody and $`K_3`$ is the rate of the creation of the antibodies by B-cells. Correspondingly B-cells carrying $`i`$-th idiotype on their surfaces decay at a given rate and proliferate when they maturate. The probability of the proliferation of B-cells is represented by the function $`Prol(\sigma )`$. When B-cells maturate, they begin to proliferate. Again, $`Prol(\sigma )`$ is assumed to be increasing with respect to $`\sigma `$ when $`\sigma `$ is small. When the neutralization of antigens completes, the proliferation of B-cells is suppressed by T-cells. Therefore, we assume that $`Prol(\sigma )`$ is decreasing with respect to $`\sigma `$ when $`\sigma `$ is large. Thus, $`Prol(\sigma )`$ also has the convex shape. Further, it seems that the proliferation of B-cells ends after their maturation ends, it is a reasonable assumption that $`Prol(\sigma )`$ is shifted to right from $`Mat(\sigma )`$(Fig.1).
Then, the evolution equation for the concentration $`b_i`$ of the B-cells with $`i`$-th idiotype can be written as
$`{\displaystyle \frac{db_i}{dt}}`$ $`=`$ $`K_4b_i+K_5Prol\left(\sigma _i\right)b_i+K_6,`$ (3)
where $`K_4`$ is the death rate of the B-cells and $`K_5`$ is the rate of production of the B-cells. Further, the term $`K_6`$ is added to take into account the cells that are recruited into the active network from the bone marrow.
Here, let us see in detail how $`Mat(\sigma )`$ and $`Prol(\sigma )`$ work. First, let us consider $`Mat(\sigma )`$.
When the sensitivity $`\sigma `$ is small, the B-cells are inactive. If the value of $`\sigma `$ becomes large, they are activated to maturate by helper T-cells and begin to produce antibodies. If the value of $`\sigma `$ increases further, the production of antibodies by B-cells is suppressed by the suppressor T-cells. As for the behavior of $`Prol(\sigma )`$, its behavior is similar to that of $`Mat(\sigma )`$. If sensitivity $`\sigma `$ becomes large, the B-cells are proliferated by the T-cells and then they produce many antibodies. If $`\sigma `$ increases further, this action is suppressed also by the T-cells.
Next, we describe the version of the Varela model introduced by H. Bersini and B. Calenbuhr, which we investigate and modify in this paper.
H. Bersini and V. Calenbuhr have investigated a dynamical system model of immune networks in the above framework using the following functions of the maturation and the proliferation in small degrees of freedom(Fig.2).
$`Mat\left(\sigma _i\right)`$ $`=`$ $`\mathrm{exp}\left[\left\{{\displaystyle \frac{ln\left(\sigma _i/\mu _m\right)}{S_m}}\right\}^2\right]`$ (4)
$`Prol\left(\sigma _i\right)`$ $`=`$ $`\mathrm{exp}\left[\left\{{\displaystyle \frac{ln\left(\sigma _i/\mu _p\right)}{S_p}}\right\}^2\right]`$ (5)
In their model, B-cells and antibodies with the same idiotype are considered to form a unit. By the interaction between two units the two-unit system oscillates and the phase of one unit is opposite to that of the other.
For a three-unit network, to begin with, they considered the connection between two units in an open chain fashion(Fig.3a). Then, three units can be constrained with opposite phases each other.
Next, they considered the closed network by connecting three units as shown in Fig.3b. The connectivity matrix they used is as follows,
$`M=\left[\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 0\end{array}\right]`$ (9)
Although each pair of units must independently comply with the imposed constraint that they oscillate in opposite phases, it isn’t possible for all units to satisfy this constraint. This phenomenon has been designated by the term ”frustration” and occurs in a network with a closed loop composed of an odd number of units. In general, frustration induces instability. As a result of this instability, although the time evolution of each unit resembles the motion in an open chain network, this motion does not continue more than several oscillations, and the network behaves in random way. In the following section, we investigate the characteristics of the behaviors in the original model by Bersini and Calenbuhr. In §3, we modify the original model by adopting simpler functions of $`Mat(\sigma )`$ and $`Prol(\sigma )`$ and see the effects of the choice of these functions. In the modified model, we consider a threshold over which each antibody can recognize others. This is introduced in §4. In §5, we consider the networks with more than 3 units and investigate the effect of degrees of freedom. Then the invasion of antigens is investigated in §6. Finally, §7 is devoted to summary and discussions.
§2. The characteristics of the original model
As is mentioned in the introduction, the system exhibits periodic oscillations for both the cases of two units and of the three-units open chain. On the other hand, for the case of the three-units closed chain, the system exhibits stable chaotic oscillations.
In this system, there always exists the following fixed point
$$(f_i,b_i)=(0,K_6/K_4),(i=1,\mathrm{},N),$$
(10)
and this point is stable. However, this solution has no meaning as a network because the interaction between units does not exist.
First, to investigate the instability of the system, we calculated Lyapunov characteristic exponents in the original model and obtained the following result for the Lyapunov spectrum,
$$(\lambda _1,\mathrm{},\lambda _6)=(+,+,0,,,).$$
This implies that the resultant strange attractor of this system is rather complicated in structure because there are two positive exponents.
Since this model has permutational symmetry, it should be checked whether the chaos is robust with respect to symmetry breaking perturbations. To investigate this, we gave random values around 1 to $`m_{ij}`$ for $`ij`$ with $`m_{ii}`$ fixed to 0, and studied the time evolution of the system. We found that chaotic behaviors could also be obtained for asymmetric systems. Thus the chaos in this model is robust and the cause of appearance of chaos is not the symmetry of the system.
Next, to see how chaos appears in this system, we changed the strength of connectivity retaining the permutational symmetry in the system. To be specific, we set the connectivity matrix as follows,
$`M=s\times \left[\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 0\end{array}\right]`$ (14)
and lowered $`s`$ from 1 to 0. Then, we obtained the following results. See Fig.4.
Until a rather small value of $`s_00.4`$, the system exhibits chaotic motion. When $`s`$ is less than $`s_0`$, a trajectory converges to the fixed point. Thus, when the connectivity is small, this model is not adequate to describe the immune network.
On the other hand, when we increased $`s`$ from 1, at $`s1.7`$ a limit cycle appears and this breaks the permutational symmetry of numbering of units.
As is shown in Fig.5(a), the phase portraits of two of the units are the same but they oscillate in opposite phases to each other. We call these units long-pulse units. One unit has smaller amplitude than those of the other two units. We call it the short-pulse unit. Let us see the characteristic feature of the oscillation of this limit cycle. Let us take notice of the time series of antibodies (Fig.5(b)). At almost all times, one of the long-pulse units has a large value of the concentration but others have small values. When the largest concentration of a long-pulse unit, say $`f_1`$, becomes small, the other two $`f_2`$ and $`f_3`$ increase taking similar values initially, then at some value of the concentrations, the concentration of the other long-pulse unit, say $`f_2`$, becomes large and the short-pulse unit $`f_3`$ becomes small. Next, when $`f_2`$ becomes small, this time $`f_1`$ and $`f_3`$ increase taking similar value and $`f_1`$ becomes large and $`f_3`$ becomes small. Thus, it seems that the short-pulse unit plays a role of ’switching’ of the activation of the long-pulse units. We call this state a clustering state.
As $`s`$ is increased further, at $`s4.2`$. the permutational symmetry breaks completely and the medium-pulse unit appears. This state also can be called a clustering state and the role of switching is played by the short-pulse unit.
At $`s4.7`$, the concentrations $`f`$ and $`b`$ in one unit are nearly equal to 0. Since this unit merely affects other units, we do not regard this state as a clustering state.
As for the route to chaos at $`s1.7`$, it is considered to be Intermittency by heteroclinic intersections <sup>1</sup><sup>1</sup>1It is reported by other authors that when some parameter is changed, this system shows Intermittency.. The sudden disappearance of chaos at $`s0.4`$ seems to be crisis, that is, the basin of chaos intersects with that of stable fixed point.
There are several characteristic features of this model by Bersini and Calenbuhr, those are, the topological dimension of the resultant strange attractor is three, the chaos is hyperchaos, and the switching in the clustering state. Since in this model the functions $`Mat(\sigma )`$ and $`Prol(\sigma )`$ are rather complicated, it is interesting to investigate whether these features are due to the special choice of these functions. In order to see the effect of these functions, we modify the above model by choosing simpler functions $`Mat(\sigma )`$ and $`Prol(\sigma )`$. In the next section, we go into this study.
§3. Modified model
We change the functions of the maturation and the proliferation as follows,
$`Mat\left(\sigma _i\right)`$ $`=`$ $`U_1\times [\mathrm{tanh}\{U_2\times (\sigma _iT_{lm})\}`$
$`\mathrm{tanh}\{U_3\times (\sigma _iT_{um})\}],`$
$`Prol\left(\sigma _i\right)`$ $`=`$ $`U_4\times [\mathrm{tanh}\{U_5\times (\sigma _iT_{lp})\}`$
$`\mathrm{tanh}\{U_6\times (\sigma _iT_{up})\}],`$
where $`U_1U_6,T_{lm},T_{um},T_{lp}`$ and $`T_{up}`$ are constants. In Fig.6 we show the graphs of these functions. They look rather similar to those of the previous functions.
When the above functions are adopted as $`Mat(\sigma )`$ and $`Prol(\sigma )`$ , in the case of two units, the system converges to a fixed point and in the case of the three-unit closed chain, a strange attractor appears. See Fig.7.
The Lyapunov spectrum for chaos becomes as follows.
$$(\lambda _1,\mathrm{},\lambda _6)=(+,0,,,,).$$
Then, chaos in this system has lower dimension than that in the original system.
To investigate the onset mechanism of chaos, we drew the bifurcation diagram decreasing the magnitude of connection matrix in the same manner as in the previous section. See Fig.8.
When $`s`$ is decreased from 1 or increased from 1, the transition from the strange attractor to a limit cycle takes place suddenly at $`s0.94`$ or $`s1.5`$, respectively. For the limit cycle state which appears below $`s0.94`$, two of three units oscillate in opposite phases, and the other unit takes negligibly small values. See Fig.9.
On the other hand, for the limit cycle state which appears above $`s1.5`$, two of three units oscillate in opposite phases, and the other unit oscillates with smaller values. That is, the permutational symmetry is broken in these states. The limit cycle state which appears above $`s1.5`$ is considered to be a clustering state as discussed in §2. See Fig.5 and Fig.10.
To see the bifurcation phenomena below $`s=1`$ in detail, we calculated the first Lyapunov characteristic exponent while decreasing the strength $`s`$ of the connection matrix. See Fig.11.
There exists a definite critical point. In the chaos region, time series of $`b_i`$ exhibit regular oscillations interrupted by irregular motion. Thus the route to chaos from the limit cycle is Intermittency. In fact, we confirmed that in both transitions to chaos taking place at larger and smaller values of $`s`$ than $`s=1`$, the routes to chaos are Intermittency by heteroclinic intersections as in the original model.
As for the robustness, we checked that the system is robust with respect to permutational symmetry breaking perturbation as in the original model.
From the results in this section, we conclude that except for the topological dimensionality of the chaos, which is two in this model and three in the original one, all features are the same as those in the original model. Thus, the modified model is simpler than the original model.
§4. The modified model with threshold
In this section, we introduce a threshold over which antibodies can recognize antibodies and antigens. There are several reasons to take the threshold into account. One is that it seems that there exists some threshold for the concentration of antibodies to recognize antigens. Another reason is that we would like to consider the situation in which the concentration of an antigen can become large without being recognized by antibodies for some reason. As such a situation we can consider the case that the ability of detecting antigens in the immune system becomes weak. Further, as a technical reason, introducing thresholds makes it possible to define states clearly.
When the threshold is introduced, the system exhibits various interesting behaviors.
Threshold
In the three-unit closed chain system, we introduce a threshold over which the $`i`$-th antibody can recognize other antibodies. For simplicity we take the common value $`f_0`$ of thresholds for all antibodies as follows,
$`\{\begin{array}{cc}m_{ij}(t)=1\text{ for any }j(i)\hfill & \text{ when }f_j(t)f_0,\hfill \\ m_{ij}(t)=0\text{ for any }j(i)\hfill & \text{ when }f_j(t)<f_0,\hfill \end{array}`$
That is,
$$m_{ij}(t)=m_{ij}\mathrm{\Theta }(f_j(t)f_0),$$
(18)
where $`\mathrm{\Theta }(x)`$ is the Heaviside function, i.e., $`\mathrm{\Theta }(x)=1`$ for $`x0`$ and 0 for $`x<0`$. Hereafter, $`m_{ij}`$ in the right hand side of the above equation is fixed to the value in eq.(6). We scanned the value of the threshold $`f_0`$ every 5 values and obtained the following behaviors of the system.
$`\begin{array}{cc}f_0=515\hfill & \{\begin{array}{c}\text{ Limit cycles. The concentrations of B-cells of two units are}\hfill \\ \text{ always greater than the threshold and the other is always}\hfill \\ \text{ less than it.(Fig.12)}\hfill \end{array}\hfill \\ f_020\hfill & \text{ Limit cycle with period two.}\hfill \\ f_0=2540\hfill & \text{ Chaos. Fig.13 }\hfill \\ f_0=4550\hfill & \text{ Limit cycles. Clustering state.(Fig.14).}\hfill \\ f_055\hfill & \text{ Fixed point.}\hfill \end{array}`$
For $`f_0=550`$, the system is in the clustering state which is defined in §2. In this state, each of three units oscillates taking values below and over the threshold. Two long-pulse units have longer duration over the threshold and the short-pulse unit has shorter duration.
For the time series analysis, we define the on-off time series as follows. Let us associate 0 or 1 with each unit according to the value of $`f_i`$, that is, 0 for $`f_i<f_0`$ and 1 for $`f_if_0`$. We call the former the off-state and the latter the on-state, and the sequence of 0 and 1 as a function of time the on-off time series. As shown in Fig.15, at almost all times, there exists only one on-state, and whenever the long-pulse unit changes from the on-state to the off-state, the short-pulse unit becomes the on-state. That is, the short-pulse unit plays a role of switching. This is the same phenomenon as observed in §2 and §3.
Thus, this phenomenon takes place in all models we studied. The cause of this phenomenon is ascribed to the nature of the interaction. We discuss this in the final section.
If we take an initial state such that all units are less than the threshold, the system converges to the fixed point. On the other hand, if at least one unit exceeds the threshold initially, then the system goes to the clustering state.
We investigated the bifurcation structure when the system is in the clustering state for $`f_0=50`$. To see the bifurcation structure clearly, we introduce the strength of effective interaction $`<I_{ij}>`$. $`<I_{ij}>`$ is defined by the following relation,
$`<I_{ij}>`$ $`=`$ $`\underset{T\mathrm{}}{lim}{\displaystyle \frac{1}{T}}{\displaystyle _0^{\mathrm{}}}𝑑tI_{ij}(t).`$ (25)
$`I_{ij}(t)=m_{ij}(t)+m_{ji}(t).`$
For example, if $`f_1(t)f_0,f_2(t)f_0`$ and $`f_3(t)<f_0`$, then $`m_{21}(t)=m_{31}(t)=1,m_{12}(t)=m_{32}(t)=1`$ and $`m_{13}(t)=m_{23}(t)=0`$. Thus, $`I_{12}(t)=I_{21}(t)=2,I_{13}(t)=I_{31}(t)=1`$ and $`I_{23}(t)=I_{32}(t)=1`$. See Fig.16.
As $`s`$ is lowered, the initial periodic state for $`s=1`$ becomes chaos at $`s0.93`$, and the permutational symmetry is recovered. See Fig.17. When $`s`$ is lowered further, the periodic state appears again at $`s0.52`$(Fig.18).
We show the bifurcation diagram (Fig.19) and the $`s`$ dependence of $`<I_{ij}>`$ (Fig.20).
Here, we summarize the result.
1. $`s_1(0.93)s1`$. Limit cycle. (Clustering state.) Fig.14.
There occurs clustering and there are two groups of $`<I_{ij}>`$. That is, the interaction between two long-pulse units, say $`I_{LL}`$, and the interaction between the long-pulse unit and the short-pulse unit , say $`I_{LS}`$. Typical values are
$$I_{LL}0.83,I_{LS}0.8.$$
2. $`s_2(0.52)ss_1`$. Chaos. Fig.17.
All $`<I_{ij}>`$ take almost the same values. The typical value is
$$I_{LL}I_{LS}0.85.$$
3. $`s_3(0.4)ss_2`$. Limit cycle. Fig.18.
As for the usual time series, one unit takes negligibly small values and the other two units oscillate in opposite phases. Then, the symmetry of this state is broken and two units always exceed the threshold and the other never exceeds the threshold. Reflecting these behaviors, the values of $`<I_{ij}>`$ are divided into two groups, - in one group $`<I_{ij}>`$ tends to 2 and in the other it tends to 1 as $`ss_3`$. Since one unit does not affect other units, it is not appropriate to call this a clustering state.
4. $`0<ss_3`$. Fixed point.
There is no meaning of a network.
From these observations, we note that $`<I_{ij}>`$ can be used to distinguish the clustering state from other behaviors. In particular, in the state of chaos, all the $`<I_{ij}>`$ take almost the same value.
The Lyapunov spectrum for chaos is
$$(\lambda _1,\mathrm{},\lambda _6)=(+,0,,,,),$$
and is the same as in the modified model.
The routes to chaos at $`s0.93`$ and $`s0.52`$ are also Intermittency by the heteroclinic intersections.
§5. The effect of the degree of freedom
Here we investigate the behaviors of the network when the number of units is increased for the modified model with threshold. We assume that there exists interactions between any two units and the connectivity matrix is symmetric, i.e., $`m_{ij}=m_{ji}=1`$ for any $`i`$ and $`j`$ $`(ij)`$.
We investigated the cases of $`N=3,4,\mathrm{},10`$. In Fig.21 and 22, we show phase portraits for several cases.
The clustering state occurs in all cases. However, it seems that it takes place more frequently in the case of the odd number of units in the parameter ranges which we investigated. (Fig.21, 22)
Indeed, for $`N=3,5,7`$ and 9 and when the threshold is the same for all units, clustering takes place and for $`N=4,6,8`$ and 10, it does not. In the figures 21(b) and 22(b), we show the effective interactions schematically. In these figures thick lines denote large strength, thin lines represent medium strength and dotted lines represent small strength. When the clustering occurs, the system is in a limit cycle state, and the number of the long-pulse units is larger by one than that of the short-pulse units. There are three values of the strength of the effective interaction in the system. See Fig.22(b). The strength of the effective interaction takes the largest value $`I_{LL}`$ between two long-pulse units, the smallest $`I_{SS}`$ between two short-pulse units and intermediate value between the long-pulse unit and the short-pulse unit. When the clustering does not occur in the system, the strengths of the effective interactions are almost the same. See Fig.21(b).
We show the on-off time series for $`N=3`$(Fig.15), $`N=5`$(Fig.23), $`N=4`$(Fig.24(a)) and $`N=6`$(Fig.24(b)).
From these figures we note that for $`N=3`$ and 5, there are only two types of units, i.e., long-pulse units and short-pulse units. On the other hand, for $`N=4`$ and 6, in any unit, the duration time of the on-state varies in time. To clarify this we calculate the histogram of the duration time. See Fig.25, 26, 27, 28. In the case of odd number of units, we notice that the duration time $`T_L`$ for the long pulse unit is nearly twice as the duration time $`T_S`$ for the short-pulse unit. On the other hand, in the case of even member of units, although there are various duration times, we can find two peaks of the long duration time and the short one. That is, in the chaotic state, the role of each unit changes dynamically.
§6. The invasion of antigen
Hereafter, we consider the response of the system to the invasion of antigens in the modified model with threshold.
Case 1
We consider the clustering state in the 3-units closed network. Suppose that external antigens similar to antibodies $`f_1`$ invade the system. Let us denote the concentration of the antigen by $`a_1`$ and that of the corresponding antibodies by $`f_1`$. Then, antibodies $`f_2`$ and $`f_3`$ recognize the antigens, because $`a_1`$ resemble $`f_1`$. However, in general the antibodies $`f_1`$ cannot recognize the antigens. See Fig.29.
Further, we assume that the antigen does not proliferate by itself <sup>2</sup><sup>2</sup>2This restricts the type of antigens. For example, pollen are one candidate.. Thus, the differential equation for the antigen is given by
$`{\displaystyle \frac{da_1}{dt}}=K_1\sigma _a(t)a_1+K_7,`$ (26)
where $`\sigma _a(t)=m_{12}(t)f_2(t)+m_{13}(t)f_3(t)`$. Here, we assume that the antigens enter into the system at a rate $`K_7`$ per unit time. On the other hand, since antibodies $`f_2`$ and $`f_3`$ interact with the antigens, $`\sigma _2`$ and $`\sigma _3`$ become $`\sigma _2=m_{21}(f_1+a_1)\mathrm{\Theta }(f_1+a_1f_0)+m_{23}(t)f_3`$ and $`\sigma _3=m_{31}(f_1+a_1)\mathrm{\Theta }(f_1+a_1f_0)+m_{32}(t)f_2`$, respectively. Then using these $`\sigma `$s the equations for antibodies and B-cells are the same as the previous ones.
Now, let us see what happens in this system. The behavior of the system depends on the increase rate of the antigen $`K_7`$. If $`K_7`$ is large enough, say $`K_7>K_7^u(1.2)`$, the concentration of the antigen $`a_1`$ increases infinitely and the system is completely invaded and destroyed by the antigen<sup>3</sup><sup>3</sup>3 The system tends to the fixed point. If $`K_7`$ is less than some value, say $`K_7<K_7^l(0.7)`$, the system copes with the antigens completely. The number of antigens finally become small and the system settles near to a clustering state depending on the initial condition. Thus, in this case there is no memory of the invasion by antigens.
If we set $`K_7`$ between $`K_7^l`$ and $`K_7^u`$, e.g., $`K_7=0.7`$, $`a_1`$ does not increase infinitely but oscillates in some range of concentration. The time series of $`a_1`$ and the phase portrait of the system are drawn in Fig.30 and 31, respectively.
Although the system is modified because of the invasion by the antigen, for $`K_7^l<K_7<K_7^u`$, it still keeps the nature of the network as a whole. In the resultant state, the duration time of the on-state for $`f_2`$ and $`f_3`$ are longer, i.e., the units 2 and 3 are long-pulse units.
Now, let us study the response time of the system when $`K_7=0`$. To do this, starting from $`a_1=50`$ we calculated the relaxation time in which the concentration of the antigen becomes negligibly small. See Figure 32.
From this figure, we note that the response time is shorter in the state in which the units 2 and 3 are in the long-pulse units. In the resultant attractor the units 2 and 3 are the long-pulse units. This result implies that the system modifies itself so as to neutralize the antigen as effectively as possible. Thus, it seems that the resultant state can respond much better than other states.
Therefore, we can state that the resultant attractor is viewed as a kind of memory of the invasion by the antigen.<sup>4</sup><sup>4</sup>4 We use the term ’memory’ to express the change to the positive direction of the system under the invasion by the antigens. It should not be confused with creation of memory B-cells.
Further, we scanned the initial value of $`A_7`$ with $`K_7`$ fixed to 0.7. For $`A_1<288`$, the system settles to a clustering state. On the other hand, for $`A_1>288`$ the system tends to a fixed point and the network collapses.
Case 2
Next, we consider the case that an antigen $`A_1`$ interacts only with the antibody $`f_1`$ in the 3-unit closed network. (Fig.33).
In this case, we set the thresholds depending on units in order to obtain clustering states. Here, we introduce new notations of thresholds, $`f_{i,0}(i=13)`$, $`g_{1,0}`$ and $`g_{A,0}`$ as follows. $`f_{i,0}`$ is the threshold over which the $`i`$-th antibody recognizes other antibodies. $`g_{1,0}`$ is the threshold over which the antibody $`f_1`$ recognizes the antigen and $`g_{A,0}`$ is the threshold over which the antigen $`A_1`$ recognizes the antibody $`f_1`$. Then the equation for the antigen $`A_1`$ is
$`{\displaystyle \frac{dA_1}{dt}}`$ $`=`$ $`K_1\sigma _A(t)A_1+K_7,`$
$`\sigma _A(t)=\mathrm{\Theta }(f_1g_{1,0})f_1.`$
The first term expresses the decrease of the antigen by neutralization by the antibody 1 and the second term represents the continuous invasion of the antigen. The sensitivity $`\sigma _i`$ of the $`i`$-th unit is modified to
$$\sigma _i=\underset{j=1}{\overset{n}{}}m_{ij}\mathrm{\Theta }(f_jf_{j,0})f_j+l_iA_1\mathrm{\Theta }(A_1g_{A,0}),$$
(28)
where $`l_i`$ is the strength of the interaction between $`f_i`$ and $`A_1`$. We put $`l_i=s_A\delta _{i,1}`$, where $`\delta _{i,1}`$ is the Kronecker’s delta. The behavior of this network strongly depends on the thresholds, the value of connectivity $`s_A`$, the initial value of the antigen $`A_1`$ and the rate of invasion of the antigen $`K_7`$. If these values are taken appropriately, the concentration of the antigen oscillates in some range. See Fig.34.
As in the case 1, the unit 1 which can interact with the antigen is activated and finally settles near to the long-pulse state. See Fig.35.
It turns out that the relaxation time of the concentration of the antigen is comparable in this state with in the state where the unit 1 is in the short-pulse state. See Fig.36.
When we compare this result with the result in the case 1, it is considered that the number of units which can interact with antigens is important for the relaxation time.
Case 3
Now, let us consider the network with 4-units. We assume that the antigen ($`a_1`$) has a similar three-dimensional structure to the antibody 1 and then can interact with the unit 2, 3 and 4, but cannot interact with the unit 1. See Fig.37.
The differential equation for the antigen is
$`{\displaystyle \frac{da_1}{dt}}`$ $`=`$ $`K_1\sigma _a(t)a_1+K_7,`$ (30)
$`\sigma _a(t)=(m_{12}(t)f_2+m_{13}(t)f_3+m_{14}(t)f_4).`$
For the unit $`i`$, the differential equations are
$`{\displaystyle \frac{df_i}{dt}}`$ $`=`$ $`K_1\sigma _i(t)f_iK_2f_i+K_3Mat(\sigma _i(t))b_i,`$ (31)
$`{\displaystyle \frac{db_i}{dt}}`$ $`=`$ $`K_4b_i+K_5Prol(\sigma _i(t))b_i+K_6,`$ (34)
$`\sigma _1(t)={\displaystyle \underset{j1}{}}m_{1j}(t)f_j,`$
$`\sigma _i(t)=m_{i1}(f_1+a_1)\mathrm{\Theta }(f_1+a_1f_0)+{\displaystyle \underset{j1}{}}m_{1j}(t)f_j,i1.`$
If we break the symmetry of the connectivity matrix by lowering the threshold of one of the four units, for some initial condition this unit stays in the short-pulse state and other units stay in the long-pulse states, and so the clustering takes place(Fig.38). For some other initial conditions, chaotic states appear. That is, in this case chaos and the clustering states coexist.
From the result of the case 1, it is expected that the relaxation time of the system is short when the unit which resembles the antigen is in the short-pulse state. Further, from the result of the case 2, it is expected that the more is the number of units which interact with the antigens, the shorter is the relaxation time. Thus, here we investigate the relaxation time for the following three cases.
Case a. The unit 1 is in the short-pulse state.
Case b. The unit 1 is in the long-pulse state.
Case c. The system is chaos.
We show the result in Fig.39.
The relaxation time $`\tau _a`$ for the case a is shortest and $`\tau _b`$ for the case b is longest. $`\tau _c`$ for the case c is in between $`\tau _a`$ and $`\tau _b`$. From this result, it is considered that chaos is more effective than the clustering states to prepare for various types of antigens.
§7. Summary and discussions
In this paper, we studied the three models of the immune network for small number of degrees of freedom $`N10`$, the original model introduced by Bersini et.al., the modified model with different functions of the maturation and the proliferation of B-cells from those of the original model, and the modified model with a threshold over which antibodies can recognize other antibodies and antigens. First, we summarize common characteristics of these models.
In these models there exist limit cycle states and chaotic states. We investigated bifurcation structures obtained by changing several parameters and found that the transitions to chaos are Intermittency by heteroclinic intersections.
There is a peculiar type of limit cycle. In this limit cycle, the permutational symmetry of the system is broken and concentrations of antibodies and B-cells oscillate in a characteristic manner. We call this the clustering state.
In the clustering state, for example in the three-unit system, there are two long-pulse units in which the concentrations of antibodies are large and one short-pulse unit in which the concentration of the antibody is small. Two long-pulse units oscillate in anti-phase to each other. At almost all times, only one unit has a large concentration of the antibody. The short-pulse unit plays a role of ”switching” the long-pulse units which take the large concentrations.
On the other hand, in the state of chaos, no such explicit division of roles exist, but each unit changes its role dynamically. That is, in chaos, a dynamical change of roles takes place.
We investigated the invasion by antigens when the clustering takes place. We found that when the number of antigens is not too many and interaction between antigens and antibodies continues for an appropriate period, the unit which can interact with antigens settles near to the long-pulse unit after the concentration of antigens is reduced to small values in some range of a parameter.
By investigating the relaxation time, we found that in the clustering state the relaxation time depends on what state the system stays in. In the case 1 that the two units interact with antigens the relaxation time is short or long when the unit which interacts with antigens is in the long-pulse unit or short-pulse unit, respectively. On the other hand, in the case 2 that only one unit interacts with antigens the relaxation time takes similar value both in the long-pulse unit and in the short-pulse unit. Thus, if many units can interact with antigens by the invasion of the antigens the system moves to the state in which the response to antigens is most efficient. Thus, the clustering state is considered to be a memory of the invasion by antigens.
Further, we checked that the system is robust against the symmetry breaking perturbations. Chaotic and periodic states change only a little by these perturbations.
As for the modified model with thresholds, we observed a interesting feature. The feature is the positive aspect of chaos in response to the invasion by antigens. In this model there is the coexistence of chaos and clustering states when the thresholds are chosen appropriately depending on units. In the response to the antigen invasion, we found that the relaxation time in chaotic state takes an intermediate value between the large and the small response times in the two types of clustering states. This suggests a positive aspect of chaos in immune networks, that is, the possibility that chaos may cope with the invasion by any kinds of antigens equally well.
Now, let us discuss the cause of the appearance of clustering state. This is related to the mechanism of the switching of long-pulse units or the dynamical role change.
Let us consider the 3-unit closed chain. See Fig.5(b), 10(b) and 40. In the present interaction, any two units tend to oscillate in opposite phases. If the concentration of the antibody in one unit, say $`f_1`$, increases, then the sensitivities of other two units become large. By this, the second terms of the differential equations for $`f_2`$ and $`f_3`$ become large and $`f_2`$ and $`f_3`$ decrease. This is a kind of ’winner takes all’ mechanism. Then, the sensitivity $`\sigma _1`$ becomes small. Thus, when $`f_1`$ becomes large, the third term of the differential equation for $`f_1`$ becomes small compared to the other two terms. Thus, next $`f_1`$ begins decreasing. This causes the decrease of $`\sigma _2`$ and $`\sigma _3`$ and the increase of $`f_2`$ and $`f_3`$. Since $`f_2`$ and $`f_3`$ begin to increase from small values, these take similar values. However, when these become rather large, two units are forced to stay in states with opposite phases each other. Thus, one increases further and the other decreases. At this stage, for the clustering state, the increasing unit is fixed to the long-pulse unit, but for chaos, it is not fixed and the dynamical change of the switching role takes place. Therefore, the behavior of the switching is due to the ’winner takes all’ mechanism and the ’anti-ferro’ type interaction between any two units.
Next, let us discuss the causes of different behaviors between the original model and the modified model without threshold.
We studied influences of each of the terms in the 3-unit closed chain system. We pay attention to the unit 1 only, ($`f_1`$,$`b_1`$), because we consider the case that the system has the permutational symmetry. The differential equations of $`f_1`$ and $`b_1`$ are
$`{\displaystyle \frac{df_1}{dt}}`$ $`=`$ $`K_1\sigma _1f_1K_2f_1+K_3Mat\left(\sigma _1\right)b_1`$ (35)
$`{\displaystyle \frac{db_1}{dt}}`$ $`=`$ $`K_4b_1+K_5Prol\left(\sigma _1\right)b_1+K_6`$ (36)
Here, we denote each term in the above equations as follows.
$`F_1`$ $`=`$ $`K_1\sigma _1f_1,`$
$`F_2`$ $`=`$ $`K_2f_1,`$
$`F_3`$ $`=`$ $`K_3Mat\left(\sigma _1\right)b_1,`$ (37)
$`B_1`$ $`=`$ $`K_4b_1,`$
$`B_2`$ $`=`$ $`K_5Prol\left(\sigma _1\right)b_1,`$
$`B_3`$ $`=`$ $`K_6.`$
We compare the time sequence in the original model with that in the modified model. See Fig.41,42.
In these figures, solid lines denote $`F_1`$ and $`B_1`$, large dotted lines represent $`F_2`$ and $`B_2`$, small dotted lines represent $`F_3`$ and $`B_3`$, respectively. We note that the terms concerning the maturation and the proliferation give the biggest influence on the behaviors of the system in both the models. In the original model these terms increase and decrease more rapidly than in the modified model.
The chaos in the original model is hyperchaos but that in the modified model is not hyperchaos. The complexity and the large topological dimensionality for chaos in the original model seem to be due to sharp behaviors of these functions. However, to clarify the influence of the choice of these functions, it is necessary to perform further investigation. This is left to a future study.
We considered the threshold over which the antibodies can recognize other antibodies. We can also consider a threshold over which the antigen can be recognized by other antibodies. This is another situation to be investigated.
In this study, we investigated three cases of invasions by antigens. In these cases, we have interest in the generic behaviors of the system when the several parameters are changed, e.g., the initial values of antigens or the input rate of the antigen $`K_7`$, etc. We found that in some cases the system changes in a positive way, that is, in the resultant state the relaxation time becomes shorter if there are plural units which interact with antibodies. There exists a kind of memory in the system. From the perspective of the real immune network, the existence of memory states and the response to the invasion by antigens in a system with large number of degrees of freedom are very interesting problems. These will be studied in the future.
Acknowledgements
The authors are grateful to S. Tasaki, S. Kitsunezaki and P. Davis for valuable discussions. One of the authors(S. I.) would like to thank Professor Y. Kuramoto and the members of his research group in Kyoto University for valuable discussions.
Figure captions
Fig.1 Schematic figures for $`Mat(\sigma )`$ and $`Prol(\sigma )`$.
Fig.2 $`Mat(\sigma )`$ and $`Prol(\sigma )`$ used in the original model.
Fig.3 Type of connection. open and black circle are in the opposite phase.
(a) 3-units open chain, (b) 3-units closed chain.
Fig.4 Bifurcation diagram in the original model.
(a) $`s1`$, (b) $`s1`$.
Fig.5 Periodic solution at $`s=1.7`$.
(a) Phase portraits. (b) Time series of $`b_i`$.
Fig.6 $`Mat(\sigma )`$ and $`Prol(\sigma )`$ for the modified model.
Fig.7 Phase portrait of a strange attractor in the modified model.
Fig.8 Bifurcation diagram of the modified model.
Fig.9 Phase portrait of a limit cycle at $`s=0.7`$.
Fig.10 Limit cycles at $`s=1.57`$.
(a) Phase portrait. (b) Time series.
Fig.11 The first Lyapunov exponents.
Fig.12 Phase portrait of a limit cycle at $`f_0=10`$.
Fig.13 Phase portrait of chaos at $`f_0=30`$.
Fig.14 Phase portrait of limit cycle at $`f_0=50`$.
Fig.15 On-off time series in a clustering state. $`s=1,f_0=50`$.
Fig.16 Effective interactions.
Fig.17 Chaos at s=0.8.
(a) Phase portrait. (b) On-off time series.
Fig.18 Limit cycle at s=0.4
(a) Phase portrait. (b) On-off time series.
Fig.19 Bifurcation diagram of the modified model with threshold.
Fig.20 $`s`$ dependence of $`<I_{ij}>`$.
Fig.21 Attractor in 4-units system in the modified model with threshold.
(a) Phase portrait. (b) Strength of effective interactions.
Fig.22 Attractor in 5-units system in the modified model with threshold.
(a) Phase portrait. (b) Strength of effective interactions.
Fig.23 On-off time series for $`N=5`$.
Fig.24 On-off time series. (a) $`N=4`$. (b) $`N=6`$
Fig.25 Histogram of duration time. $`N=3`$.
Fig.26 Histogram of duration time. $`N=5`$.
Fig.27 Histogram of duration time. $`N=4`$.
Fig.28 Histogram of duration time. $`N=6`$.
Fig.29 Schematic figure of interaction between antibodies
and antigen in Case 1.
Fig.30 Time series of antigen $`a_1`$.
Fig.31 Phase portrait of resultant attractor.
Fig.32 Relaxation times in clustering states.
Fig.33 Schematic figure of interaction between antibodies
and antigen in Case 2.
Fig.34 Time series of antigen $`A_1`$.
Fig.35 Phase portrait of resultant attractor.
Fig.36 Relaxation times in clustering states.
Fig.37 Schematic figure of interaction between antibodies
and antigen in Case 3.
Fig.38 Clustering state.
Fig.39 Relaxation times in clustering states and in chaos.
Fig.40 Time series of limit cycle at $`f_0=50`$ in the modified model with threshold.
Fig.41 Time series of each term in the differential equations of the original model.
(a) $`f_1`$. (b) $`b_1`$.
Fig.42 Time series of each term in the differential equations
of the Modified model.(a) $`f_1`$. (b) $`b_1`$.
|
warning/0002/hep-th0002239.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The Casimir effect is among the most interesting consequences of quantum field theory and is essentially the only macroscopic manifestation of the nontrivial properties of the physical vacuum. These properties may be determined from the responce of the vacuum state to classical external fields or constraints. The simplest case is realized by boundary conditions on quantized fields. Such conditions modify the zero point mode spectrum and as a result can change the energy of the vacuum. This change is manifest as an observable Casimir energy. Since the original work by Casimir in 1948 many theoretical and experimental works have been done on this problem, including various types of boundary geometry and non-zero temperature effects (see, e.g., and references therein). Many different approaches have been used: mode summation method, Green function formalism, multiple scattering expansions, heat-kernel series, zeta function regularization technique, etc.. From the general theoretical point of view the main point here is the unique separation and subsequent removing of the divergences. Within the framework of the mode summation method in calculations of the expectation values for physical observables, such as energy-momentum tensor, one often needs to sum over the values of a certain function at integer points, and then subtract the corresponding quantity for unbounded space (usually presented in terms of integrals). Practically, the sum and integral, taken separately, diverge and some physically motivated procedure, to handle finite result, is needed. For a number of geometries one of the most convenient methods to obtain such regularized values of the mode sums is based on the using of the Abel-Plana formula (APF) . In this formula have been used to regularize scalar field energy momentum-tensor on backgrounds of various Friedmann cosmological models. Further applications to the Casimir effect for flat boundary geometries with corresponding references can be found in . Abel-Plana formula allows (i) to extract by cutoff independent way the Minkowski vacuum part and (ii) to obtain for the regularized part strongly convergent integrals, useful, in particular, for numerical calculations. However the applications of APF in usual form is restricted by the flat boundary cases when the eigenmodes have simple dependence on quantum numbers.
In the APF was generalized (see also ). The generalized version contains two meromorphic functions. Choosing one of these functions in specific form APF in usual form is obtained. By applying the generalized formula to Bessel functions in summation formulae are obtained over the zeros of various combinations of these functions. In particular, formulae for Fourier-Bessel and Dini series are derived. From these formulae by specifying the constants and choosing the order of Bessel function equal to 1/2 one obtains a simple generalization of APF for the case of a function having poles. It have been shown that from generalized formula interseting results can be derived for infinite integrals involving Bessel functions. Further the obtained summation formulae are applied to regularize the vacuum expectation values for the energy-momentum tensor components of the electromagnetic field in the Casimir effect with spherically and cylindrically symmetric boundaries. As in the case of flat boundaries the using of generalised Abel-Plana formula allow to extract in manifestly cutoff independent way the contribution of the unbounded space and to present regularized values in terms of exponentially converging integrals.
The present paper reviews these results and is organized as follows. In section 2 the generalized Abel-Plana formula is derived and as a special case usual APF is obtained. It is indicated how to generalize this formula for the functions having poles. The applications of generalized formula to Bessel functions are considered in the next section. We derive two formulae for the sums over zeros of $`AJ_\nu (z)+BzJ_\nu ^{}(z)`$. Specific examples of applications of the general formulae are considered. For $`\nu =1/2,B=0`$ and for analytic function $`f(z)`$ the APF is obtained. In section 4 from generalized Abel-Plana formula by special choice of function $`g(z)`$ summation formulae are derived for the series over zeros of the function $`J_\nu (z)Y_\nu (\lambda z)J_\nu (\lambda z)Y_\nu (z)`$ and similar combinations with Bessel functions derivatives. Special examples are considered. The applications to the integrals involving Bessel functions and some their combinations are discussed in section 5. A number of interesting results for these integrals are presented. Specific examples of applying these general formulae are described in the next section. In section 7 by using generalized Abel-Plana formula two theorems are proved for the integrals involving the function $`J_\nu (z)Y_\mu (\lambda z)J_\mu (\lambda z)Y_\nu (z)`$ and their applications are considered. The following sections are devoted to the applications of generalized formula for the calculations of the regularized vacuum expectation values of the electromagnetic energy-momentum tensor inside (section 8) and outside (section 9) a perfectly conducting spherical shell, and for the region between two perfectly conducting spherical surfaces (section 10). In sections 11-13 the similar problems for the cylindrical surfaces are considered. The section 14 concludes the main results considered in this paper.
## 2 Generalized Abel-Plana formula
Let $`f(z)`$ and $`g(z)`$ be meromorphic functions for $`axb`$ in the complex plane $`z=x+iy`$. Let us note by $`z_{f,k}`$ and $`z_{g,k}`$ the poles of $`f(z)`$ and $`g(z)`$ in region $`a<x<b`$, respectively. Assume that $`\mathrm{Im}z_{f,k}0`$ (see however the Remark to Lemma).
Lemma. If functions $`f(z)`$ and $`g(z)`$ satisfy condition
$$\underset{h\mathrm{}}{lim}_{a\pm ih}^{b\pm ih}\left[g(z)\pm f(z)\right]𝑑z=0,$$
(2.1)
then the following formula takes place
$$_a^bf(x)𝑑x=R[f(z),g(z)]\frac{1}{2}_i\mathrm{}^{+i\mathrm{}}\left[g(u)+\mathrm{sgn}(\mathrm{Im}z)f(u)\right]_{u=a+z}^{u=b+z}𝑑z,$$
(2.2)
where
$$R[f(z),g(z)]=\pi i\left[\underset{k}{}\mathrm{Res}_{z=z_{g,k}}g(z)+\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}>0}f(z)\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}<0}f(z)\right].$$
(2.3)
Proof. Let us consider a rectangle $`C_h`$ with vertices $`a\pm ih`$, $`b\pm ih`$ described in the positive sense. In accordance to the residue theorem
$$_{C_h}g(z)𝑑z=2\pi i\underset{k}{}\mathrm{Res}_{z=z_{g,k}}g(z),$$
(2.4)
where rhs contains sum over poles within $`C_h`$. Let $`C_h^+`$ and $`C_h^{}`$ denote the upper and lower halfs of this contour. Then one has
$$_{C_h}g(z)𝑑z=_{C_h^+}[g(z)+f(z)]𝑑z+_{C_h^{}}[g(z)f(z)]𝑑z_{C_h^+}f(z)𝑑z+_{C_h^{}}f(z)𝑑z.$$
(2.5)
By the same residue theorem
$$_{C_h^{}}f(z)𝑑z_{C_h^+}f(z)𝑑z=2_a^bf(x)𝑑x+2\pi i\left[\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}<0}f(z)\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}>0}f(z)\right].$$
(2.6)
Then
$$_{C_h^\pm }[g(z)\pm f(z)]𝑑z=\pm _0^{\pm ih}[g(u)\pm f(u)]_{u=a+z}^{u=b+z}𝑑z_{a\pm ih}^{b\pm ih}[g(z)\pm f(z)]𝑑z.$$
(2.7)
Combining these results and allowing in (2.4) $`h\mathrm{}`$ one obtains the formula (2.2).
If the functions $`f(z)`$ and $`g(z)`$ have poles with $`\mathrm{Re}z_{i,k}=a,b`$ ($`i=f,g`$) the contour have to pass round these points on the right or left, correspondingly.
Remark. The formula (2.2) is valid also when the function $`f(z)`$ has real poles $`z_{f,n}^{(0)}`$, $`\mathrm{Im}z_{f,n}^{(0)}=0`$ in the region $`a<\mathrm{Re}z<b`$ if the main part of its Laurent expansion near of these poles does not contain even powers of $`zz_{f,n}^{(0)}`$. In this case on the left of the formula (2.2) the integral is meant in the sense of the principal value, which exists as a consequence of the abovementioned condition. For brevity let us consider the case of a single pole $`z=z_0`$. One has
$`{\displaystyle _{C_h^{}}}f(z)𝑑z{\displaystyle _{C_h^+}}f(z)𝑑z=2\left[{\displaystyle _a^{z_0\rho }}f(z)𝑑z+{\displaystyle _{z_0+\rho }^b}f(z)𝑑z\right]+`$
$`+2\pi i\left[{\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_{f,k}<0}f(z){\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_{f,k}>0}f(z)\right]+{\displaystyle _{\mathrm{\Gamma }_\rho ^+}}f(z)𝑑z+{\displaystyle _{\mathrm{\Gamma }_\rho ^{}}}f(z)𝑑z,`$ (2.8)
with contours $`\mathrm{\Gamma }_\rho ^+`$ and $`\mathrm{\Gamma }_\rho ^{}`$ being the upper and lower circular arcs (with center at $`z=z_0`$) joining the points $`z_0\rho `$ and $`z_0+\rho `$. By taking into account that for odd negative $`l`$
$$_{\mathrm{\Gamma }_\rho ^+}(zz_0)^l𝑑z+_{\mathrm{\Gamma }_\rho ^{}}(zz_0)^l𝑑z=0,$$
(2.9)
in the limit $`\rho 0`$ we obtain the required result.
In the following on the left of (2.2) we will write $`\mathrm{p}.\mathrm{v}._a^bf(x)𝑑x`$, assuming that this integral converges in the sense of the principal value. As a direct consequence of Lemma one obtains :
Theorem 1. If in addition to the conditions of Lemma one has
$$\underset{b\mathrm{}}{lim}_b^{b\pm i\mathrm{}}\left[g(z)\pm f(z)\right]𝑑z=0,$$
(2.10)
then
$$\underset{b\mathrm{}}{lim}\{\mathrm{p}.\mathrm{v}._a^bf(x)dxR[f(z),g(z)]\}=\frac{1}{2}_{ai\mathrm{}}^{a+i\mathrm{}}[g(z)+\mathrm{sgn}(\mathrm{Im}z)f(z)]dz,$$
(2.11)
where on the left $`R[f(z),g(z)]`$ is defined as (2.3), $`a<\mathrm{Re}z_{f,k},\mathrm{Re}z_{g,k}<b`$, and summation goes over poles $`z_{f,k}`$ and $`z_{g,k}`$ arranged in order $`\mathrm{Re}z_{i,k}\mathrm{Re}z_{i,k+1}`$, $`i=f,g`$.
Proof. To proof it is sufficient to insert in the general formula (2.2) $`b\mathrm{}`$ and to use the condition (2.10). The order of summation in $`R[f(z),g(z)]`$ is determined by the choice of the integration contour $`C_h`$ and by limiting transition $`b\mathrm{}`$.
We will call the formula (2.11) as Generalized Abel-Plana Formula (GAPF) as for $`b=n+a`$, $`0<a<1`$, $`g(z)=if(z)\mathrm{cot}\pi z`$ and analytic functions $`f(z)`$ from (2.11) follows the Abel-Plana formula (APF)
$$\underset{n\mathrm{}}{lim}\left[\underset{1}{\overset{n}{}}f(s)_a^{n+a}f(x)𝑑x\right]=\frac{1}{2i}_a^{ai\mathrm{}}f(z)(\mathrm{cot}\pi zi)𝑑z\frac{1}{2i}_a^{a+i\mathrm{}}f(z)(\mathrm{cot}\pi z+i)𝑑z.$$
(2.12)
The useful form of (2.12) may be obtained performing the limit $`a0`$. By taking into account that the point $`z=0`$ is a pole for integrands and therefore have to be around by arcs of the small circle $`C_\rho `$ on the right and performing $`\rho 0`$ one obtains
$$\underset{n=0}{\overset{\mathrm{}}{}}f(n)=_0^{\mathrm{}}f(x)𝑑x+\frac{1}{2}f(0)+i_0^{\mathrm{}}\frac{f(ix)f(ix)}{e^{2\pi x}1}𝑑x.$$
(2.13)
Note that now the condition (2.1) is satisfied if
$$\underset{y\mathrm{}}{lim}e^{2\pi |y|}|f(x+iy)|=0$$
(2.14)
uniformly in any finite interval of $`x`$. The (2.13) is the most frequently used form of APF in its physical applications. Another useful form (in particular for fermionic field calculations) to sum over the values of an analytic function at half of an odd integer points can be obtained from (2.13) :
$$\underset{n=0}{\overset{\mathrm{}}{}}f(n+1/2)=_0^{\mathrm{}}f(x)𝑑xi_0^{\mathrm{}}\frac{f(ix)f(ix)}{e^{2\pi x}+1}𝑑x$$
(2.15)
By adding to the rhs of (2.13) the term
$$\pi i\left\{\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}>0}f(z)\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_{f,k}<0}f(z)i\underset{k}{}\mathrm{Res}_{z=z_{f,k}}\left[f(z)\mathrm{cot}\pi z\right]\right\}$$
(2.16)
the APF may be generalized for the case when the function $`f(z)`$ has poles $`z_{f,k}`$, $`\mathrm{Re}z_{f,k}>0`$, $`z_{f,k}1,2,\mathrm{}`$.
As a next consequence of (2.11) a summation formula can be obtained over the points $`z_n,\mathrm{Re}z_n>0`$ at which the analytic function $`s(z)`$ takes integer values, $`s(z_n)`$ is an integer, and $`s^{}(z_n)0`$. Taking in (2.11) $`g(z)=if(z)\mathrm{cot}\pi s(z)`$ one obtains the following formula
$$\frac{f(z_n)}{s^{}(z_n)}=w+_0^{\mathrm{}}f(x)𝑑x+_0^{\mathrm{}}\left[\frac{f(ix)}{e^{2\pi is(ix)}1}\frac{f(ix)}{e^{2\pi is(ix)}1}\right]𝑑x,$$
(2.17)
where
$$w=\{\begin{array}{cc}0,\hfill & \text{if}s(0)0,\pm 1,\pm 2,\mathrm{}\hfill \\ f(0)/[2s^{}(0)],\hfill & \text{if}s(0)=0,\pm 1,\pm 2,\mathrm{}\hfill \end{array}$$
(2.18)
For $`s(z)=z`$ we return to APF in usual form. An example of applications of this formula to the Casimir effect is given in .
## 3 Applications to Bessel functions
The formula (2.11) contains two meromorphic functions and is too general. To obtain more special consequences we have to specify the one of them. As we have seen in previous section the one of the possible ways leads to APF. Here we will consider another choices of the function $`g(z)`$ and will obtain useful formulae for the sums over zeros of Bessel function and their combinations, as well as some formulae for integrals involving these functions.
First of all to simplify the formulae let us introduce the notation
$$\overline{F}(z)AF(z)+BzF^{}(z)$$
(3.1)
for a given function $`F(z)`$, where the prime denotes derivative with respect to the argument of function, $`A`$ and $`B`$ are constants. As a function $`g(z)`$ in GAPF let us choose
$$g(z)=i\frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}f(z),$$
(3.2)
where $`J_\nu (z)`$ and $`Y_\nu (z)`$ are Bessel functions of the first and second (Neumann function) kind. For the sum and difference on the right of (2.11) one obtains
$$f(z)(1)^kg(z)=\frac{\overline{H}_\nu ^{(k)}(z)}{\overline{J}_\nu (z)}f(z),k=1,2$$
(3.3)
with $`H_\nu ^{(1)}`$ and $`H_\nu ^{(2)}`$ being Bessel functions of the third kind or Hankel functions. For such a choice the integrals (2.1) and (2.10) can be estimated by using the asymptotic formulae for Bessel functions for fixed $`\nu `$ and $`|z|\mathrm{}`$ (see, for example, ). It can be easily seen that conditions (2.1) and (2.10) are satisfied if the function $`f(z)`$ is restricted by the one of the following constraints
$$|f(z)|<\epsilon (x)e^{c|y|}\text{ or }|f(z)|<\frac{Me^{2|y|}}{|z|^\alpha },z=x+iy,|z|\mathrm{},$$
(3.4)
where $`c<2`$, $`\alpha >1`$ and $`\epsilon (x)0`$ for $`x\mathrm{}`$. Indeed, from the asymptotic expressions for Bessel functions it follows that
$`\left|{\displaystyle _{a\pm ih}^{b\pm ih}}\left[g(z)\pm f(z)\right]𝑑z\right|`$ $`=`$ $`\left|{\displaystyle _a^b}{\displaystyle \frac{\overline{H}_\nu ^{(1,2)}(x\pm ih)}{\overline{J}_\nu (x\pm ih)}}f(x\pm ih)𝑑x\right|<\{\begin{array}{c}M_1e^{(c2)h}\hfill \\ M_1^{}/h^\alpha \hfill \end{array}`$ (3.7)
$`\left|{\displaystyle _b^{b\pm i\mathrm{}}}\left[g(z)\pm f(z)\right]𝑑z\right|`$ $`=`$ $`\left|{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{H}_\nu ^{(1,2)}(b\pm ix)}{\overline{J}_\nu (b\pm ix)}}f(b\pm ix)𝑑x\right|<\{\begin{array}{c}N_1\epsilon (b)\hfill \\ N_1^{}/b^{\alpha 1}\hfill \end{array}`$ (3.10)
with constants $`M_1,M_1^{},N_1,N_1^{}`$, and $`H_\nu ^{(1)}`$ ($`H_\nu ^{(2)}`$) corresponds to the upper (lower) sign.
Let us denote by $`\lambda _{\nu ,k}0`$, $`k=1,2,3\mathrm{}`$ the zeros of $`\overline{J}_\nu (z)`$ in the right half-plane, arranged in ascending order of the real part, $`\mathrm{Re}\lambda _{\nu ,k}\mathrm{Re}\lambda _{\nu ,k+1}`$, (if some of these zeros lie on the imaginary axis we will take only zeros with positive imaginary part). All these zeros are simple. Note that for real $`\nu >1`$ the function $`\overline{J}_\nu (z)`$ has only real zeros, except the case $`A/B+\nu <0`$ when there are two purely imaginary zeros . By using the Wronskian $`W[J_\nu (z),Y_\nu (z)]=2/\pi z`$ for (2.3) one finds
$$R[f(z),g(z)]=2\underset{k}{}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})+r_{1\nu }[f(z)],$$
(3.11)
where we have introduced the notations
$`T_\nu (z)`$ $`=`$ $`{\displaystyle \frac{z}{\left(z^2\nu ^2\right)J_\nu ^2(z)+z^2J_\nu ^2(z)}}`$ (3.12)
$`r_{1\nu }[f(z)]`$ $`=`$ $`\pi i{\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k>0}f(z){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}}\pi i{\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k<0}f(z){\displaystyle \frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}}`$ (3.13)
$`\pi {\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k=0}f(z){\displaystyle \frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}}.`$
Here $`z_k`$ ($`\lambda _{\nu ,i}`$) are the poles for the function $`f(z)`$ in the region $`Rez>a>0`$. Substituting (3.11) into (2.11) we obtain that for the function $`f(z)`$ meroporphic in the half-plane $`\mathrm{Re}za`$ and satisfying the condition (3.4) the following formula takes place
$`\underset{b+\mathrm{}}{lim}\{2{\displaystyle \underset{k=m}{\overset{n}{}}}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})+r_{1\nu }[f(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _a^b}f(x)dx\}=`$
$`{\displaystyle \frac{1}{2}}{\displaystyle _a^{a+i\mathrm{}}}f(z){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}}𝑑z{\displaystyle \frac{1}{2}}{\displaystyle _a^{ai\mathrm{}}}f(z){\displaystyle \frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}}𝑑z,`$ (3.14)
where $`\mathrm{Re}\lambda _{\nu ,m1}<a<\mathrm{Re}\lambda _{\nu ,m}`$, $`\mathrm{Re}\lambda _{\nu ,n}<b<\mathrm{Re}\lambda _{\nu ,n+1}`$, $`a<\mathrm{Re}z_k<b`$. We will apply this formula to the function $`f(z)`$ meromorphic in the half-plane $`\mathrm{Re}z0`$ taking $`a0`$. Let us consider separately two cases.
### 3.1 Case (a)
Let $`f(z)`$ have no poles on the imaginary axis, except possibly at $`z=0`$, and
$$f(ze^{\pi i})=e^{2\nu \pi i}f(z)+o(z^{\beta _\nu }),z0$$
(3.15)
(this condition is trivially satisfied for the function $`f(z)=o(z^{\beta _\nu })`$), with
$$\beta _\nu =\{\begin{array}{cc}2|\mathrm{Re}\nu |1\hfill & \text{ for integer }\nu \hfill \\ \mathrm{Re}\nu +|\mathrm{Re}\nu |1\hfill & \text{ for noninteger }\nu \hfill \end{array}$$
(3.16)
Under this condition for values $`\nu `$ for which $`\overline{J}_\nu (z)`$ have no purely imaginary zeros the rhs of Eq.(3.14) in the limit $`a0`$ can be presented in the form
$$\frac{1}{\pi }_\rho ^{\mathrm{}}\frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}\left[e^{\nu \pi i}f(xe^{\pi i/2})+e^{\nu \pi i}f(xe^{\pi i/2})\right]𝑑x+_{\gamma _\rho ^+}f(z)\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}𝑑z_{\gamma _\rho ^{}}f(z)\frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}𝑑z,$$
(3.17)
with $`\gamma _\rho ^+`$ and $`\gamma _\rho ^{}`$ being upper and lower halfs of the semicircle in the right half-plane with radius $`\rho `$ and with center at point $`z=0`$, described in the positive sense with respect to this point. In (3.17) we have introduced modified Bessel functions $`I_\nu (z)`$ and $`K_\nu (z)`$ . It follows from (3.15) that for $`z0`$
$$\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}f(z)=\frac{\overline{H}_\nu ^{(2)}(ze^{\pi i})}{\overline{J}_\nu (ze^{\pi i})}f(ze^{\pi i})+o(z^1).$$
(3.18)
From here for $`\rho 0`$ one finds
$$D_\nu _{\gamma _\rho ^+}f(z)\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}𝑑z_{\gamma _\rho ^{}}f(z)\frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}𝑑z=\pi \mathrm{Res}_{z=0}f(z)\frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}$$
(3.19)
Indeed,
$`D_\nu `$ $`=`$ $`{\displaystyle _{\gamma _\rho ^+}}f(z){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}}𝑑z+{\displaystyle _{\gamma _{1\rho }^+}}f(ze^{\pi i}){\displaystyle \frac{\overline{H}_\nu ^{(2)}(ze^{\pi i})}{\overline{J}_\nu (ze^{\pi i})}}𝑑z={\displaystyle _{\gamma _\rho ^++\gamma _{1\rho }^+}}f(z){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}}𝑑z+`$ (3.20)
$`+{\displaystyle _{\gamma _{1\rho }^+}}o(z^1)𝑑z=i{\displaystyle _{\gamma _\rho ^++\gamma _{1\rho }^+}}f(z){\displaystyle \frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}}𝑑z+{\displaystyle _{\gamma _{1\rho }^+}}o(z^1)𝑑z,`$
where $`\gamma _{1\rho }^+`$ ($`\gamma _{1\rho }^{}`$, see below) is the upper (lower) half of the semicircle with radius $`\rho `$ in the left half-plane with center at $`z=0`$ (described in the positive sense). In the last equality we have used the condition that integral $`\mathrm{p}.\mathrm{v}._0^bf(x)𝑑x`$ converges at lower limit. By similar way it can be seen that
$$D_\nu =i_{\gamma _\rho ^{}+\gamma _{1\rho }^{}}f(z)\frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}𝑑z+_{\gamma _{1\rho }^{}}o(z^1)𝑑z.$$
(3.21)
Combining the last two results we obtain (3.19) in the limit $`\rho 0`$. By using (3.14), (3.17) and (3.19) we have :
Theorem 2. If f(z) is a single valued analytic function in the half-plane $`\mathrm{Re}z0`$ (with possible branch point at $`z=0`$) except the poles $`z_k`$ ($`\lambda _{\nu ,i}`$), $`\mathrm{Re}z_k>0`$ (for the case of function $`f(z)`$ having purely imaginary poles see Remark after Theorem 3), and satisfy conditions (3.4) and (3.15), then in the case of $`\nu `$ for which the function $`\overline{J}_\nu (z)`$ has no purely imaginary zeros, the following formula is valid
$`\underset{b+\mathrm{}}{lim}\{2{\displaystyle \underset{k=1}{\overset{n}{}}}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})+r_{1\nu }[f(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _0^b}f(x)dx\}=`$
$`={\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}f(z){\displaystyle \frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}}{\displaystyle \frac{1}{\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}}\left[e^{\nu \pi i}f(xe^{\pi i/2})+e^{\nu \pi i}f(xe^{\pi i/2})\right]𝑑x,`$ (3.22)
where on the left $`\mathrm{Re}\lambda _{\nu ,n}<b<\mathrm{Re}\lambda _{\nu ,n+1}`$, $`0<\mathrm{Re}z_k<b`$, and $`T_\nu (\lambda _{\nu ,k})`$ and $`r_{1\nu }[f(z)]`$ are determined by relations (3.12) and (3.13).
Under the condition (3.15) the integral on the right converges at lower limit. Recall that we assume the existence of the integral on the left as well (see section 2). The formula (3.22) and analog ones given below are especially useful for numerical calculations of the sums over $`\lambda _{\nu ,k}`$ as under the first conditions in (3.4) the integral on the right converges exponentially fast at the upper limit.
Remark. Deriving the formula (3.22) we have assumed that the function $`f(z)`$ is meromorphic in the half-plane $`\mathrm{Re}z0`$ (except possibly at $`z=0`$). However this formula is valid also for some functions having branch points on the imaginary axis, for example,
$$f(z)=f_1(z)\underset{l=1}{\overset{k}{}}\left(z^2+c_l^2\right)^{\pm 1/2},$$
(3.23)
with meromorphic function $`f_1(z)`$. The proof for (3.22) in this case is similar to the given above with difference that branch points $`\pm ic_l`$ have to be around on the right along contours with small radii. In view of the further applications to the Casimir effect (see below) let us consider the case $`k=1`$. By taking into account that
$$\left(z^2+c^2\right)^{1/2}=\{\begin{array}{ccc}\left|z^2+c^2\right|^{1/2}\hfill & \text{if}\hfill & |z|<c\hfill \\ \left|z^2+c^2\right|^{1/2}e^{i\pi /2}\hfill & \text{if}\hfill & \mathrm{Im}z>c\hfill \\ \left|z^2+c^2\right|^{1/2}e^{i\pi /2}\hfill & \text{if}\hfill & \mathrm{Im}z<c\hfill \end{array}$$
(3.24)
from (3.22) one obtains
$`\underset{b+\mathrm{}}{lim}\{2{\displaystyle \underset{k=1}{\overset{n}{}}}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})+r_{1\nu }[f(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _0^b}f(x)dx\}={\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}f(z){\displaystyle \frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}}`$
$`{\displaystyle \frac{1}{\pi }}{\displaystyle _0^c}{\displaystyle \frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}}\left[e^{\nu \pi i}f_1(xe^{\pi i/2})+e^{\nu \pi i}f_1(xe^{\pi i/2})\right]\left(c^2x^2\right)^{\pm 1/2}𝑑x`$
$`{\displaystyle \frac{i}{\pi }}{\displaystyle _c^{\mathrm{}}}{\displaystyle \frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}}\left[e^{\nu \pi i}f_1(xe^{\pi i/2})e^{\nu \pi i}f_1(xe^{\pi i/2})\right]\left(x^2c^2\right)^{\pm 1/2}𝑑x,`$ (3.25)
where $`f(z)=f_1(z)\left(z^2+c^2\right)^{\pm 1/2},c>0`$. In Section 11 we apply this formula with analytic function $`f_1(z)`$ to derive the expressions for the regularized values of the energy-momentum tensor components in the region inside the perfectly conducting cylindrical shell.
For an analitic function $`f(z)`$ the formula (3.22) yields
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{2\lambda _{\nu ,k}f(\lambda _{\nu ,k})}{\left(\lambda _{\nu ,k}^2\nu ^2\right)J_\nu ^2(\lambda _{\nu ,k})+\lambda _{\nu ,k}^2J_\nu ^2(\lambda _{\nu ,k})}}={\displaystyle _0^{\mathrm{}}}f(x)𝑑x+{\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}f(z){\displaystyle \frac{\overline{Y}_\nu (z)}{\overline{J}_\nu (z)}}`$
$`{\displaystyle \frac{1}{\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}}\left[e^{\nu \pi i}f(xe^{\pi i/2})+e^{\nu \pi i}f(xe^{\pi i/2})\right]𝑑x.`$ (3.26)
By taking in this formula $`\nu =1/2`$, $`A=1,B=0`$ (see the notation (3.1)) as a particular case we immediately receive the APF in the form (2.13). In like manner substituting $`\nu =1/2`$, $`A=1`$, $`B=2`$ we obtain APF in the form (2.15). Consequently the formula (3.22) is a generalization of APF for general $`\nu `$ (with restrictions given above) and for functions $`f(z)`$ having poles in the right half-plane.
Having in mind the further applications to the Casimir effect in Sections 8 and 11 let us choose in (3.26)
$$f(z)=F(z)J_{\nu +m}^2(zt),t>0,\mathrm{Re}\nu 0$$
(3.27)
with $`m`$ being an integer. Now the conditions (3.4) formulated in terms of $`F(z)`$ are in form
$$|F(z)|<|z|\epsilon e^{(c2t)|y|}\text{ or }|f(z)|<\frac{Me^{2(1t)|y|}}{|z|^{\alpha 1}},z=x+iy,|z|\mathrm{}$$
(3.28)
with the same notations as in (3.4). In like manner from the condition (3.15) for $`F(z)`$ one has
$$F(ze^{\pi i})=F(z)+o(z^{2m1}),z0.$$
(3.29)
Now as a consequence of (3.26) we obtain that if the conditions (3.28) and (3.29) are satisfied, then for the function $`F(z)`$ analytic in the right half-plane, the following formula takes place
$`2{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}T_\nu (\lambda _{\nu ,k})F(\lambda _{\nu ,k})J_{\nu +m}^2(\lambda _{\nu ,k}t)={\displaystyle _0^{\mathrm{}}}F(x)J_{\nu +m}^2(xt)𝑑x`$
$`{\displaystyle \frac{1}{\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{K}_\nu (x)}{\overline{I}_\nu (x)}}I_{\nu +m}^2(xt)\left[F(xe^{\pi i/2})+F(xe^{\pi i/2})\right]𝑑x`$ (3.30)
for $`\mathrm{Re}\nu 0`$ and $`\mathrm{Re}\nu +m0`$.
### 3.2 Case (b)
Let $`f(z)`$ be a function satisfying the condition
$$f(xe^{\pi i/2})=e^{2\nu \pi i}f(xe^{\pi i/2})$$
(3.31)
for real $`x`$. It is clear that if $`f(z)`$ have purely imaginary poles, then they are complex conjugate: $`\pm iy_k`$, $`y_k>0`$. By (3.31) the rhs of (3.14) for $`a0`$ and $`\mathrm{arg}\lambda _{\nu ,k}=\pi /2`$ may be written as
$$\left(_{\gamma _\rho ^+}+\underset{\sigma _k=iy_k,\lambda _{\nu ,k}}{}_{C_\rho (\sigma _k)}\right)\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}f(z)dz\left(_{\gamma _\rho ^{}}+\underset{\sigma _k=iy_k,\lambda _{\nu ,k}}{}_{C_\rho (\sigma _k)}\right)\frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}f(z)dz,$$
(3.32)
where $`C_\rho (\sigma _k)`$ denotes the right half of the circle with center at the point $`\sigma _k`$ and radius $`\rho `$, described in the positive sense, and the contours $`\gamma _\rho ^\pm `$ are the same as in (3.17). We have used the fact the purely imaginary zeros of $`\overline{J}_\nu (z)`$ are complex conjugate numbers, as $`\overline{J}_\nu (ze^{\pi i})=e^{\nu \pi i}\overline{J}_\nu (z)`$. We have used also the fact that on the right of (3.14) the integrals (with $`a=0`$) along straight segments of the upper and lower imaginary semiaxes are canceled, as in accordance of (3.31) for $`\mathrm{arg}z=\pi /2`$
$$\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}f(z)=\frac{\overline{H}_\nu ^{(2)}(ze^{\pi i})}{\overline{J}_\nu (ze^{\pi i})}f(ze^{\pi i}).$$
(3.33)
Let us show that from (3.33) for $`z_0=x_0e^{\pi i/2}`$ it follows that this relation is valid for any $`z`$ in a small enough region including this point. Namely, as the function $`f(z)\overline{H}_\nu ^{(p)}(z)/\overline{J}_\nu (z)`$, $`p=1,2`$ is meromorphic near the point $`(1)^{p+1}x_0e^{\pi i/2}`$, there exists a neighbourhood of this point where this function is presented as a Laurent expansion
$$\frac{\overline{H}_\nu ^{(p)}(z)}{\overline{J}_\nu (z)}f(z)=\underset{n=n_0}{\overset{\mathrm{}}{}}\frac{a_n^{(p)}}{\left[z(1)^{p+1}x_0e^{\pi i/2}\right]^n}.$$
(3.34)
From (3.33) for $`z=xe^{\pi i/2}`$ one concludes
$$\underset{n=n_0}{\overset{\mathrm{}}{}}\frac{a_n^{(1)}e^{n\pi i/2}}{\left(xx_0\right)^n}=\underset{n=n_0}{\overset{\mathrm{}}{}}\frac{(1)^na_n^{(2)}e^{n\pi i/2}}{\left(xx_0\right)^n},$$
(3.35)
and hence $`a_n^{(1)}=(1)^na_n^{(2)}`$. Our statement follows directly from here. By this it can be seen that
$$_{C_\rho (\sigma _k)}\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}f(z)𝑑z_{C_\rho (\sigma _k)}\frac{\overline{H}_\nu ^{(2)}(z)}{\overline{J}_\nu (z)}f(z)𝑑z=2\pi i\mathrm{Res}_{z=\sigma _k}\frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}f(z)$$
(3.36)
where $`\sigma _k=iy_k,\lambda _{\nu ,k}`$, $`\mathrm{arg}\lambda _{\nu ,k}=\pi /2`$. Now by taking into account (3.19) and letting $`\rho 0`$ we get :
Theorem 3. Let $`f(z)`$ be meromorphic function in the half-plane $`\mathrm{Re}z0`$ (except possibly at $`z=0`$) with poles $`z_k,\mathrm{Re}z_k>0`$ and $`\pm iy_k,y_k>0`$, $`k=1,2,\mathrm{}`$ ($`\lambda _{\nu ,p}`$). If this function satisfy the conditions (3.4) and (3.31) then
$`\underset{b+\mathrm{}}{lim}\{2{\displaystyle \underset{k=1}{\overset{n}{}}}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})+r_{1\nu }[f(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _0^b}f(x)dx\}=`$
$`={\displaystyle \frac{\pi i}{2}}{\displaystyle \underset{\eta _k=0,iy_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}f(z){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}},`$ (3.37)
where on the left $`0<\mathrm{Re}z_k<b`$, $`\mathrm{Re}\lambda _{\nu ,n}<b<\mathrm{Re}\lambda _{\nu ,n+1}`$ and $`r_{1\nu }`$ is defined by (3.13).
Note that the residue terms in (3.36) with $`\sigma _k=\lambda _{\nu ,k}`$, $`\mathrm{arg}\lambda _{\nu ,k}=\pi /2`$ are equal to $`4T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})`$ and are included in the first sum on the left of (3.37).
Remark. Let $`\pm iy_k,y_k>0`$ and $`\pm \lambda _{\nu ,k}`$, $`\mathrm{arg}\lambda _{\nu ,k}=\pi /2`$ are purely imaginary poles of function $`f(z)`$ and purely imaginary zeros of $`\overline{J}_\nu (z)`$, correspondingly. Let function $`f(z)`$ satisfy condition
$$f(z)=e^{2\nu \pi i}f(ze^{\pi i})+o\left((z\sigma _k)^1\right),z\sigma _k,\sigma _k=iy_k,\lambda _{\nu ,k}.$$
(3.38)
Now in the limit $`a0`$ the rhs of (3.14) can be presented in the form (3.32) plus integrals along the straight segments of the imaginary axis between the poles. Using the arguments similar those given above we obtain the relation (3.36) with additional contribution from the last term on the right of (3.38) in the form $`_{C_\rho (\sigma _k)}o\left((z\sigma _k)^1\right)𝑑z`$. In the limit $`\rho 0`$ the latter vanishes and sum of the integrals along the straight segments of the imaginary axis gives the principal value of the integral on the right of (3.22). As a result the formula (3.22) can be generalized for functions having purely imaginary poles and satisfying condition (3.38) writing instead of residue term on the right the sum of residues from the right of (3.37) and taking the principle value of the integral on the right. The latter exists due to the condition (3.38).
It follows from (3.37) an interesting result. Let $`\lambda _{\mu ,k}^{(1)}`$ be zeros of the function $`A_1J_\mu (z)+B_1zJ_\mu ^{}(z)`$ with some real constants $`A_1`$ and $`B_1`$. Let $`f(z)`$ be an analytic function in the right half-plane satisfying condition (3.31) and $`f(z)=o(z^\beta )`$ for $`z0`$, where $`\beta =\mathrm{max}(\beta _\mu ,\beta _\nu )`$ (the definition $`\beta _\nu `$ see (3.16)). For this function from (3.37) we get
$$\underset{k=1}{\overset{\mathrm{}}{}}T_\mu (\lambda _{\mu ,k}^{(1)})f(\lambda _{\mu ,k}^{(1)})=\underset{k=1}{\overset{\mathrm{}}{}}T_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k}),\mu =\nu +m.$$
(3.39)
For the case of Fourier-Bessel and Dini series this result is given in .
Let us consider some applications of the formula (3.37) to the special types of series. Firstly we choose in this formula
$$f(z)=F_1(z)J_\mu (zt),t>0,$$
(3.40)
where the function $`F_2(z)`$ is meromorphic on the right half-plane and satisfy conditions
$$|F_1(z)|<\epsilon _1(x)e^{(ct)|y|}\text{or}|F_1(z)|<M|z|^{\alpha _1}e^{(2t)|y|},|z|\mathrm{},$$
(3.41)
with $`c<2,\alpha _1>1/2`$, $`\epsilon _1(x)=o(\sqrt{x})`$ for $`x+\mathrm{}`$, and condition
$$F_1(xe^{\pi i/2})=e^{(2\nu \mu )\pi i}F_1(xe^{\pi i/2}).$$
(3.42)
From (3.41) it follows that the integral $`\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F_1(x)J_\mu (xt)𝑑x`$ converges at the upper limit and hence in this case the formula (3.37) may be written in the form
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}T_\nu (\lambda _{\nu ,k})F_1(\lambda _{\nu ,k})J_\mu (\lambda _{\nu ,k}t)={\displaystyle \frac{1}{2}}\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}F_1(x)J_\mu (xt)𝑑x{\displaystyle \frac{1}{2}}r_{1\nu }\left[F_1(z)J_\mu (zt)\right]`$
$`{\displaystyle \frac{\pi i}{4}}{\displaystyle \underset{\eta _k=0,iy_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}F_2(z)J_\mu (zt){\displaystyle \frac{\overline{H}_\nu ^{(1)}(z)}{\overline{J}_\nu (z)}},`$ (3.43)
For example, it follows from here that for $`t<1`$, $`\mathrm{Re}\sigma ,\mathrm{Re}\nu >1`$
$$\underset{k=1}{\overset{\mathrm{}}{}}\frac{T_\nu (\lambda _{\nu ,k})}{\lambda _{\nu ,k}^\sigma }J_{\sigma +\nu +1}(\lambda _{\nu ,k})J_\nu (\lambda _{\nu ,k}t)=\frac{1}{2}_0^{\mathrm{}}J_{\sigma +\nu +1}(z)J_\nu (zt)\frac{dz}{z^\sigma }=\frac{\left(1t^2\right)^\sigma t^\nu }{2^{\sigma +1}\mathrm{\Gamma }(\sigma +1)}$$
(3.44)
(for the value of integral see, e.g., ). For $`B=0`$ this result is given in . In a similar manner taking $`\mu =\nu +m`$,
$$F_1(z)=z^{\nu +m+1}\frac{J_\sigma (a\sqrt{z^2+z_1^2})}{\left(z^2+z_1^2\right)^{\sigma /2}},a>0$$
(3.45)
with $`\mathrm{Re}\nu 0`$ and $`\mathrm{Re}\nu +m0`$, from (3.43) for $`a<2t`$, $`\mathrm{Re}\sigma >\mathrm{Re}\nu +m`$ one finds
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}T_\nu (\lambda _{\nu ,k})\lambda _{\nu ,k}^{\nu +m+1}J_{\nu +m}(\lambda _{\nu ,k}t){\displaystyle \frac{J_\sigma (a\sqrt{\lambda _{\nu ,k}^2+z_1^2})}{\left(\lambda _{\nu ,k}^2+z_1^2\right)^{\sigma /2}}}=`$ (3.46)
$`={\displaystyle \frac{1}{2}}{\displaystyle _0^{\mathrm{}}}x^{\nu +m+1}J_{\nu +m}(xt){\displaystyle \frac{J_\sigma (a\sqrt{x^2+z_1^2})}{\left(x^2+z_1^2\right)^{\sigma /2}}}𝑑x={\displaystyle \frac{t^{\nu +1}}{a^\sigma }}\left(z_1\right)^{m+1}{\displaystyle \frac{J_{m+1}(z_1\sqrt{a^2t^2})}{\left(a^2t^2\right)^{(m+1)/2}}},a>t`$
and the sum is zero when $`a<t`$. Here we have used the known value for Sonine integral .
If an addition to (3.41), (3.42) the function $`F_1`$ satisfies conditions
$$F_1(xe^{\pi i/2})=e^{\mu \pi i}F_1(xe^{\pi i/2})$$
(3.47)
and
$$|F_2(z)|<\epsilon _1(x)e^{c_1t|y|}\text{or}|F_2(z)|<M|z|^{\alpha _1}e^{t|y|},|z|\mathrm{},$$
(3.48)
when the formula (5.7)(see below) with $`B=0`$ may be applied to the integral on the right of (3.43). This gives
Corollary 1. Let $`F(z)`$ be meromorphic function in the half-plane $`\mathrm{Re}z0`$ (except possibly at $`z=0`$) with poles $`z_k,\mathrm{Re}z_k>0`$ and $`\pm iy_k,y_k>0`$ ($`\lambda _{\nu ,i}`$). If $`F(z)`$ satisfy condition
$$F(xe^{\pi i/2})=(1)^{m+1}e^{\nu \pi i}F(xe^{\pi i/2})$$
(3.49)
with an integer $`m`$, and to one of inequalities
$$|F(z)|<\epsilon _1(x)e^{a|y|}\text{or}|F(z)|<M|z|^{\alpha _1}e^{a_0|y|},|z|\mathrm{},$$
(3.50)
with $`a<\mathrm{min}(t,2t)a_0`$, $`\epsilon _1(x)=o(x^{1/2}),x+\mathrm{}`$, $`\alpha _1>1/2`$, the following formula is valid
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}T_\nu (\lambda _{\nu ,k})F(\lambda _{\nu ,k})J_{\nu +m}(\lambda _{\nu ,k}t)=`$
$`={\displaystyle \frac{\pi i}{4}}{\displaystyle \underset{\eta _k=0,iy_k,z_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}\left\{\left[J_{\nu +m}(zt)\overline{Y}_\nu (z)Y_{\nu +m}(zt)\overline{J}_\nu (z)\right]{\displaystyle \frac{F(z)}{\overline{J}_\nu (z)}}\right\}.`$ (3.51)
Recall that for the imaginary zeros $`\lambda _{\nu ,k}`$, in lhs of (3.51) the zeros with positive imaginary parts enter only. By using the formula (3.51) a number of Furier-Bessel and Dini series can be summarized (see, for instance, below).
Remark. The formula (3.51) may be obtained also by considering the integral
$$\frac{1}{\pi }_{C_h}\left[H_{\nu +m}^{(2)}(zt)\overline{H}_\nu ^{(1)}(z)H_{\nu +m}^{(1)}(zt)\overline{H}_\nu ^{(2)}(z)\right]\frac{F(z)}{\overline{J}_\nu (z)}𝑑z,$$
(3.52)
where $`C_h`$ is an rectangle with vertices $`\pm ih,b\pm ih`$, described in the positive sense (purely imaginary poles of $`F(z)/\overline{J}_\nu (z)`$ and the origin are around by semicircles in the right half-plane with small radii). This integral is equal to the sum of residues over the poles within $`C_h`$ (points $`z_k`$, $`\lambda _{\nu ,k}`$, ($`Rez_k,\mathrm{Re}\lambda _{\nu ,k}>0`$)). On the other hand it follows from (3.49) that integrals along the segments of the imaginary axes cancel each other. The sum of integrals along the conjugate semicircles give the sum of residues over purely imaginary poles in the upper half plane. The integrals along the remained three segments of $`C_h`$ in accordance with (3.50) approach to zero in the limit $`b,h\mathrm{}`$. Equating these expressions for (3.52) one immediately obtains the result (3.51).
From (3.51) for $`t=1,F(z)=J_\nu (zx)`$, $`m=1`$ one obtains
$$\underset{k=1}{\overset{\mathrm{}}{}}T_\nu (\lambda _{\nu ,k})J_\nu (\lambda _{\nu ,k}x)J_{\nu +1}(\lambda _{\nu ,k})=\frac{x^\nu }{2},0x<1.$$
(3.53)
By similar way choosing $`m=0`$, $`F(z)=zJ_\nu (zx)/\left(z^2a^2\right)`$, $`B=0`$ we obtain the Kneser-Sommerfeld expansion :
$$\underset{k=1}{\overset{\mathrm{}}{}}\frac{J_\nu (\lambda _{\nu ,k}t)J_\nu (\lambda _{\nu ,k}x)}{\left(\lambda _{\nu ,k}^2a^2\right)J_{\nu +1}^2(\lambda _{\nu ,k})}=\frac{\pi }{4}\frac{J_\nu (ax)}{J_\nu (a)}\left[J_\nu (at)Y_\nu (a)Y_\nu (at)J_\nu (a)\right],0xt1.$$
(3.54)
In (3.51) as a function $`F(z)`$ one may choose, for example, the following functions
$`z^{\rho 1}{\displaystyle \underset{l=1}{\overset{n}{}}}\left(z^2+z_l^2\right)^{\mu _l/2}J_{\mu _l}(b_l\sqrt{z^2+z_l^2}),`$ (3.55)
$`\text{for}\mathrm{Re}\nu <{\displaystyle \underset{l=1}{\overset{n}{}}}\mathrm{Re}\mu _l+n/2+2p+3/2m\delta _{ba_0},ba_0,b={\displaystyle \underset{l=1}{\overset{n}{}}}b_l;`$
$`z^{\rho 2n1}{\displaystyle \underset{l=1}{\overset{n}{}}}\left[1J_0(b_lz)\right],`$ (3.56)
$`\text{for}\mathrm{Re}\nu <2n+2p+3/2m\delta _{ba_0};`$
$`z^{\rho 1}{\displaystyle \underset{l=1}{\overset{n}{}}}\left(z^2+z_l^2\right)^{\mu _l/2}Y_{\mu _l}\left(b_l\sqrt{z^2+z_l^2}\right),\mu _l>0\text{-half of an odd integer,}`$ (3.57)
$`\text{for}\mathrm{Re}\nu <{\displaystyle \underset{l=1}{\overset{n}{}}}\mu _l+n/2+2p+3/2m\delta _{ba_0};`$
$`z^{\rho 1}{\displaystyle \underset{l=1}{\overset{n}{}}}z^{|k_l|}\left[J_{\mu _l+k_l}(a_lz)Y_{\mu _l}(b_lz)Y_{\mu _l+k_l}(a_lz)J_{\mu _l}(b_lz)\right],k_l\text{- integer,}`$ (3.58)
$`\text{for}\mathrm{Re}\nu <n+2p+3/2m{\displaystyle }|k_l|\delta _{\stackrel{~}{a},a_0},\stackrel{~}{a}{\displaystyle \underset{l=1}{\overset{n}{}}}|a_lb_l|a_0;`$
with $`\rho =\nu +m2p`$ ($`p`$ \- integer), as well as any products between these functions and with $`_l\left(z^2c_l^2\right)^{p_l}`$, provided the condition (3.50) is satisfied. For example, the following formulae take place
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}j_{\nu ,k}^{\nu 2}{\displaystyle \frac{J_\nu (j_{\nu ,k}t)}{J_{\nu +1}^2(j_{\nu ,k})}}{\displaystyle \underset{l=1}{\overset{n}{}}}\left[J_{\mu _l}(a_lj_{\nu ,k})Y_{\mu _l}(b_lj_{\nu ,k})Y_{\mu _l}(a_lj_{\nu ,k})J_{\mu _l}(b_lj_{\nu ,k})\right]=`$
$`={\displaystyle \frac{2^{\nu 2}}{\pi ^nt^\nu }}\left(1t^{2\nu }\right){\displaystyle \underset{l=1}{\overset{n}{}}}{\displaystyle \frac{b_l^{\mu _l}}{\mu _la_l^{\mu _l}}}\left[1\left({\displaystyle \frac{a_l}{b_l}}\right)^{2\mu _l}\right],0<t1,`$ (3.59)
$`c{\displaystyle \underset{l=1}{\overset{n}{}}}|a_lb_l|t,a_l,b_l>0,\mathrm{Re}\mu _l0,\mathrm{Re}\nu <n+3/2\delta _{ct};`$
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{J_\nu (j_{\nu ,k}t)J_{\nu +1}(\lambda j_{\nu ,k})}{j_{\nu ,k}^{2n+3}J_{\nu +1}^2(j_{\nu ,k})}}{\displaystyle \underset{l=1}{\overset{n}{}}}\left[1J_0(b_lj_{\nu ,k})\right]={\displaystyle \frac{\lambda ^{\nu +1}\left(1t^{2\nu }\right)}{4^{n+1}\nu (\nu +1)t^\nu }}{\displaystyle \underset{l=1}{\overset{n}{}}}b_l^2,`$ (3.60)
$`\lambda +{\displaystyle \underset{l=1}{\overset{n}{}}}b_lt1,\lambda ,b_l>0;`$
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{J_\mu (j_{\nu ,k}b)J_{\nu +1}(\lambda j_{\nu ,k})J_\nu (j_{\nu ,k}t)}{\left(j_{\nu ,k}^2a^2\right)j_{\nu ,k}^{\mu +1}J_{\nu +1}^2(j_{\nu ,k})}}={\displaystyle \frac{\pi J_{\nu +1}(a\lambda )}{4a^{\mu +1}}}{\displaystyle \frac{J_\mu (ba)}{J_\nu (a)}}\left[Y_\nu (a)J_\nu (at)J_\nu (a)Y_\nu (at)\right],`$ (3.61)
$`\lambda +bt1,\lambda ,b>0,\mathrm{Re}\mu >7/2+\delta _{\lambda +b,t},`$
where $`j_{\nu ,k}`$ are zeros of $`J_\nu (z)`$.
The examples of the series over zeros of Bessel functions we found in literature (see, e.g., ), when the corresponding sum was evaluated in finite terms, are particular cases of the formulae given in this section.
## 4 Summation formulae over zeros of $`\overline{J}_\nu (z)\overline{Y}_\nu (\lambda z)\overline{Y}_\nu (z)\overline{J}_\nu (\lambda z)`$
In this section we will consider the series over zeros of the function
$$C_\nu ^{AB}(\lambda ,z)\overline{J}_\nu (z)\overline{Y}_\nu (\lambda z)\overline{Y}_\nu (z)\overline{J}_\nu (\lambda z),$$
(4.1)
where the bared quantities are defined as (3.1). Series of this type arise in calculations of the vacuum expectation values in confined regions with boundaries of spherical and cylindrical form. To obtain a summation formula for these series let us substitute in (2.11)
$$g(z)=\frac{1}{2i}\left[\frac{\overline{H}_\nu ^{(1)}(\lambda z)}{\overline{H}_\nu ^{(1)}(z)}+\frac{\overline{H}_\nu ^{(2)}(\lambda z)}{\overline{H}_\nu ^{(2)}(z)}\right]\frac{h(z)}{C_\nu ^{AB}(\lambda ,z)},f(z)=\frac{h(z)}{\overline{H}_\nu ^{(1)}(z)\overline{H}_\nu ^{(2)}(z)},$$
(4.2)
where for definiteness we shall assume that $`\lambda >1`$. The sum and difference of these functions are
$$g(z)(1)^kf(z)=i\frac{\overline{H}_\nu ^{(k)}(\lambda z)}{\overline{H}_\nu ^{(k)}(z)}\frac{h(z)}{C_\nu ^{AB}(\lambda ,z)},k=1,2.$$
(4.3)
The conditions for GAPF written in terms of the function $`h(z)`$ are as follows
$$|h(z)|<\epsilon _2(x)e^{c_2|y|}\text{or}|h(z)|<M|z|^{\alpha _2}e^{2(\lambda 1)|y|},|z|\mathrm{},z=x+iy$$
(4.4)
where $`c_2<2(\lambda 1)`$, $`x^{2\delta _{B0}1}\epsilon _2(x)0`$ for $`x+\mathrm{}`$, $`\alpha _2>2\delta _{B0}`$. Let $`\gamma _{\nu ,k}`$ be zeros for the function $`C_\nu ^{AB}(\lambda ,z)`$ in the right half-plane. In this section we will assume values of $`\nu `$, $`A,B`$ for which all these zeros are real and simple, and the function $`\overline{H}_\nu ^{(1)}(z)`$ ($`\overline{H}_\nu ^{(2)}(z)`$) has no zeros in the right half of the upper (lower) half-plane. As we will see later these conditions are satisfied in physical problems considered below. For real $`\nu `$, $`A`$, $`B`$ the zeros $`\gamma _{\nu ,k}`$ are simple. To see this note that the function $`J_\nu (tz)\overline{Y}_\nu (z)Y_\nu (tz)\overline{J}_\nu (z)`$ is cylinder function with respect to $`t`$. Using the standard result for indefinite integrals containing any two cylinder functions (see ) it can be seen that
$$_1^\lambda t\left[J_\nu (tz)\overline{Y}_\nu (z)Y_\nu (tz)\overline{J}_\nu (z)\right]^2𝑑t=\frac{2}{\pi ^2zT_\nu ^{AB}(\lambda ,z)},z=\gamma _{\nu ,k},$$
(4.5)
where we have introduced the notation
$$T_\nu ^{AB}(\lambda ,z)=z\left\{\frac{\overline{J}_\nu ^2(z)}{\overline{J}_\nu ^2(\lambda z)}\left[A^2+B^2(\lambda ^2z^2\nu ^2)\right]A^2B^2(z^2\nu ^2)\right\}^1.$$
(4.6)
On the other hand
$$\frac{}{z}C_\nu ^{AB}(\lambda ,z)=\frac{2}{\pi T_\nu ^{AB}(\lambda ,z)}\frac{\overline{J}_\nu (\lambda z)}{\overline{J}_\nu (z)},z=\gamma _{\nu ,k}.$$
(4.7)
Combining the last two results we deduce that for real $`\nu `$, $`A`$, $`B`$ the derivative (4.7) is nonzero and hence the zeros $`z=\gamma _{\nu ,k}`$ are simple. By using this it can be seen that
$$\mathrm{Res}_{z=\gamma _{\nu ,k}}g(z)=\frac{\pi }{2i}T_\nu ^{AB}(\lambda ,\gamma _{\nu ,k}).$$
(4.8)
Hence if the function $`h(z)`$ is analytic in the half-plane $`\mathrm{Re}za>0`$ except at the poles $`z_k`$ ($`\gamma _{\nu ,i}`$) and satisfy to the one of two conditions (4.4), the following formula takes place
$`\underset{b+\mathrm{}}{lim}\{{\displaystyle \frac{\pi ^2}{2}}{\displaystyle \underset{k=n}{\overset{m}{}}}T_\nu ^{AB}(\lambda ,\gamma _{\nu ,k})h(\gamma _{\nu ,k})+r_{2\nu }[h(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _a^b}{\displaystyle \frac{h(x)dx}{\overline{J}_\nu ^2(x)+\overline{Y}_\nu ^2(kx)}}\}=`$
$`={\displaystyle \frac{i}{2}}{\displaystyle _a^{a+i\mathrm{}}}{\displaystyle \frac{\overline{H}_\nu ^{(1)}(\lambda z)}{\overline{H}_\nu ^{(1)}(z)}}{\displaystyle \frac{h(z)}{C_\nu ^{AB}(\lambda ,z)}}𝑑z{\displaystyle \frac{i}{2}}{\displaystyle _a^{ai\mathrm{}}}{\displaystyle \frac{\overline{H}_\nu ^{(2)}(\lambda z)}{\overline{H}_\nu ^{(2)}(z)}}{\displaystyle \frac{h(z)}{C_\nu ^{AB}(\lambda ,z)}}𝑑z.`$ (4.9)
Here we assumed that the integral on the left exists, $`\gamma _{\nu ,n1}<a<\gamma _{\nu ,n}`$, $`\gamma _{\nu ,m}<b<\gamma _{\nu ,m+1}`$, $`a<\mathrm{Re}z_k<b`$, $`\mathrm{Re}z_k\mathrm{Re}z_{k+1}`$, and the following notation is introduced
$`r_{2\nu }[h(z)]`$ $`=`$ $`\pi {\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k=0}\left[{\displaystyle \frac{\overline{J}_\nu (z)\overline{J}_\nu (\lambda z)+\overline{Y}_\nu (z)\overline{Y}_\nu (\lambda z)}{\overline{J}_\nu ^2(z)+\overline{Y}_\nu ^2(z)}}{\displaystyle \frac{h(z)}{C_\nu ^{AB}(\lambda ,z)}}\right]+`$ (4.10)
$`+\pi {\displaystyle \underset{k,l=1,2}{}}\mathrm{Res}_{(1)^l\mathrm{Im}z_k<0}\left[{\displaystyle \frac{\overline{H}_\nu ^{(l)}(\lambda z)}{\overline{H}_\nu ^{(l)}(z)}}{\displaystyle \frac{h(z)}{C_\nu ^{AB}(\lambda ,z)}}\right].`$
The general formula (4.9) is a direct consequence of GAPF and will be as starting point for the further applications in this section. In the limit $`a0`$ one has :
Corollary 2. Let $`h(z)`$ be analytic function for $`\mathrm{Re}z0`$ except the poles $`z_k`$ ($`\gamma _{\nu i}`$), $`\mathrm{Re}z_k>0`$ (with possible branch point $`z=0`$). If it satisfies one of two conditions (4.4) and
$$h(ze^{\pi i})=h(z)+o(z^1),z0,$$
(4.11)
and the integral
$$\mathrm{p}.\mathrm{v}._a^b\frac{h(x)dx}{\overline{J}_\nu ^2(x)+\overline{Y}_\nu ^2(x)}$$
(4.12)
exists, then
$`\underset{b+\mathrm{}}{lim}\{{\displaystyle \frac{\pi ^2}{2}}{\displaystyle \underset{k=1}{\overset{m}{}}}h(\gamma _{\nu ,k})T_\nu ^{AB}(\lambda ,\gamma _{\nu ,k})+r_{3\nu }[h(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _0^b}{\displaystyle \frac{h(x)dx}{\overline{J}_\nu ^2(x)+\overline{Y}_\nu ^2(x)}}\}=`$
$`={\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}\left[{\displaystyle \frac{h(z)\overline{H}_\nu ^{(1)}(\lambda z)}{C_\nu ^{AB}(\lambda ,z)\overline{H}_\nu ^{(1)}(z)}}\right]{\displaystyle \frac{\pi }{4}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{K}_\nu (\lambda x)}{\overline{K}_\nu (x)}}{\displaystyle \frac{\left[h(xe^{\pi i/2})+h(xe^{\pi i/2})\right]dx}{\overline{K}_\nu (x)\overline{I}_\nu (\lambda x)\overline{K}_\nu (\lambda x)\overline{I}_\nu (x)}}`$ (4.13)
In the following we shall use this formula to derive the regularized vacuum energy momentum-tensor for the region between two spherical and cylindrical surfaces. Note that (4.13) may be generalized for the functions $`h(z)`$ with purely imaginary poles $`\pm iy_k`$, $`y_k>0`$ satisfying condition
$$h(ze^{\pi i})=h(z)+o\left((ziy_k)^1\right),z\pm iy_k.$$
(4.14)
The corresponding formula is obtained from (4.13) by adding residue terms for $`z=iy_k`$ in the form of (4.16) (see below) and taking the principal value of the integral on the right. The arguments here are similar to those for Remark after Theorem 3.
By the way similar to (3.22) one has another result :
Corollary 3. Let $`h(z)`$ be meromorphic function in the half-plane $`\mathrm{Re}z0`$ (with exception the possible branch point $`z=0`$) with poles $`z_k,\pm iy_k`$ ($`\gamma _{\nu ,i}`$), $`\mathrm{Re}z_k,y_k>0`$. If this function satisfy condition
$$h(xe^{\pi i/2})=h(xe^{\pi i/2})$$
(4.15)
and the integral (4.12) exists then
$`\underset{b+\mathrm{}}{lim}\{{\displaystyle \frac{\pi ^2}{2}}{\displaystyle \underset{k=1}{\overset{m}{}}}h(\gamma _{\nu ,k})T_\nu ^{AB}(\lambda ,\gamma _{\nu ,k})+r_{2\nu }[h(z)]\mathrm{p}.\mathrm{v}.{\displaystyle _0^b}{\displaystyle \frac{h(x)dx}{\overline{J}_\nu ^2(x)+\overline{Y}_\nu ^2(x)}}\}=`$
$`={\displaystyle \frac{\pi }{2}}{\displaystyle \underset{\eta _k=0,iy_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}\left[{\displaystyle \frac{\overline{H}_\nu ^{(1)}(\lambda z)}{\overline{H}_\nu ^{(1)}(z)}}{\displaystyle \frac{h(z)}{C_\nu ^{AB}(\lambda ,z)}}\right],`$ (4.16)
where in the lhs $`\gamma _{\nu ,m}<b<\gamma _{\nu ,m+1}`$.
Let us consider a special applications of the formula (4.16) for $`A=1,B=0`$. The generalizations of these results for general $`A,B`$ under the conditions given above are straightforward.
Theorem 4. Let the function $`F(z)`$ be meromorphic in the right half-plane $`\mathrm{Re}z0`$ (with the possible exception $`z=0`$) with poles $`z_k,\pm iy_k`$ ($`\gamma _{\nu ,i}`$), $`y_k,\mathrm{Re}z_k>0`$. If it satisfy condition
$$F(xe^{\pi i/2})=(1)^{m+1}F(xe^{\pi i/2}),$$
(4.17)
with an integer $`m`$, and to the one of two inequalities
$$|F(z)|<\epsilon (x)e^{a_1|y|}\text{or}|F(z)|<M|z|^\alpha e^{a_2|y|},|z|\mathrm{},$$
(4.18)
with $`a_1<\mathrm{min}(2\lambda \sigma 1,\sigma 1)a_2`$, $`\sigma >0`$, $`\epsilon (x)0`$ for $`x+\mathrm{}`$, $`\alpha >1`$, then
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\gamma _{\nu ,k}F(\gamma _{\nu ,k})}{J_\nu ^2(\gamma _{\nu ,k})/J_\nu ^2(\lambda \gamma _{\nu ,k})1}}\left[J_\nu (\gamma _{\nu ,k})Y_{\nu +m}(\sigma \gamma _{\nu ,k})Y_\nu (\gamma _{\nu ,k})J_{\nu +m}(\sigma \gamma _{\nu ,k})\right]=`$
$`={\displaystyle \frac{1}{\pi }}{\displaystyle \underset{\eta _k=0,iy_k,z_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}{\displaystyle \frac{Y_\nu (\lambda z)J_{\nu +m}(\sigma z)J_\nu (\lambda z)Y_{\nu +m}(\sigma z)}{J_\nu (z)Y_\nu (\lambda z)J_\nu (\lambda z)Y_\nu (z)}}F(z).`$ (4.19)
Proof. As a function $`h(z)`$ in (4.16) let us choose
$$h(z)=F(z)\left[J_\nu (z)Y_{\nu +m}(\sigma z)Y_\nu (z)J_{\nu +m}(\sigma z)\right],$$
(4.20)
which in virtue of (4.18) satisfy condition (4.4). The condition (4.15) is satisfied as well. Hence $`h(z)`$ satisfy conditions for Corollary 3. The corresponding integral in (4.16) with $`h(z)`$ from (4.20) can be calculated by using the formula (7.7) (see below). Putting the value of this integral into (4.16) after some manipulations we receive to (4.19).
Remark. The formula (4.19) may be derived also by applying to the contour integral
$$_{C_h}\frac{Y_\nu (\lambda z)J_{\nu +m}(\sigma z)J_\nu (\lambda z)Y_{\nu +m}(\sigma z)}{J_\nu (z)Y_\nu (\lambda z)J_\nu (\lambda z)Y_\nu (z)}F(z)𝑑z$$
(4.21)
the residue theorem, where $`C_h`$ is a rectangle with vertices $`\pm ih,b\pm ih`$. Here the proof is similar to that for Remark to the Corollary 1.
Formula similar to (4.19) can be obtained also for the series of type $`_{k=1}^{\mathrm{}}G(\gamma _{\nu ,k})J_\mu (\gamma _{\nu ,k}t)`$ by using (4.16).
As a function $`F(z)`$ in (4.19) one can choose, for example,
* function (3.55) for $`\rho =m2p`$, $`_lb_l<a_2`$, $`m<2p+_l\mathrm{Re}\mu _l+n/2+1`$, $`p`$ \- integer;
* function (3.56) for $`\rho =m2p`$, $`_lb_l<a_2`$, $`m<2p+2n+1`$;
* function (3.58) for $`\rho =m2p`$, $`a_l>0`$, $`\mathrm{Re}\mu _l0`$ (for $`\mathrm{Re}\mu _l<0`$, $`k_l>|\mathrm{Re}\mu _l|`$),
$`_{l=1}^n|a_lb_l|<a_2`$, $`m<2p+n_l|k_l|+1`$.
For $`F(z)=1/z,m=0`$ one obtains
$$\underset{k=1}{\overset{\mathrm{}}{}}\frac{J_\nu (\gamma _{\nu ,k})Y_\nu (\sigma \gamma _{\nu ,k})Y_\nu (\gamma _{\nu ,k})J_\nu (\sigma \gamma _{\nu ,k})}{J_\nu ^2(\gamma _{\nu ,k})/J_\nu ^2(\lambda \gamma _{\nu ,k})1}=\frac{\sigma ^\nu }{\pi }\frac{(\lambda /\sigma )^{2\nu }1}{\lambda ^{2\nu }1},\lambda \sigma >1.$$
(4.22)
By similar way it can be seen that
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{\gamma _{\nu ,k}^2\left[J_\nu (\gamma _{\nu ,k})Y_\nu (\sigma \gamma _{\nu ,k})Y_\nu (\gamma _{\nu ,k})J_\nu (\sigma \gamma _{\nu ,k})\right]}{\left(\gamma _{\nu ,k}^2c^2\right)\left[J_\nu ^2(\gamma _{\nu ,k})/J_\nu ^2(\lambda \gamma _{\nu ,k})1\right]}}={\displaystyle \frac{1}{\pi }}{\displaystyle \frac{Y_\nu (\lambda c)J_\nu (\sigma c)J_\nu (\lambda c)Y_\nu (\sigma c)}{J_\nu (c)Y_\nu (\lambda c)J_\nu (\lambda c)Y_\nu (c)}},`$ (4.23)
$`{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{J_\nu (\gamma _{\nu ,k})Y_\nu (\sigma \gamma _{\nu ,k})Y_\nu (\gamma _{\nu ,k})J_\nu (\sigma \gamma _{\nu ,k})}{J_\nu ^2(\gamma _{\nu ,k})/J_\nu ^2(\lambda \gamma _{\nu ,k})1}}{\displaystyle \underset{l=1}{\overset{p}{}}}\gamma _{\nu ,k}^{\mu _l}J_{\mu _l}(b_l\gamma _{\nu ,k})=`$
$`={\displaystyle \frac{\sigma ^\nu }{\pi }}{\displaystyle \frac{(\lambda /\sigma )^{2\nu }1}{\lambda ^{2\nu }1}}{\displaystyle \underset{l=1}{\overset{p}{}}}{\displaystyle \frac{b_l^{\mu _l}}{2^{\mu _l}\mathrm{\Gamma }(\mu _l+1)}},b{\displaystyle \underset{1}{\overset{p}{}}}b_l<\sigma 1,\mathrm{Re}\mu _l+{\displaystyle \frac{p}{2}}+1>\delta _{b,\sigma 1},`$ (4.24)
where $`\mathrm{Re}c0,b_l>0`$, $`\lambda \sigma >1`$, $`\mu _l1,2,\mathrm{}`$.
So far in this section we have considered series over zeros of the function $`C_\nu ^{AB}(\lambda ,z)`$. The similar results can be obtained also for the series containing zeros of the function
$$C_{1\nu }(\lambda ,z)=J_\nu ^{}(z)Y_\nu (\lambda z)Y_\nu ^{}(z)J_\nu (\lambda z),$$
(4.25)
(on properties of zeroes of these function see ). The corresponding furmulae for the zeros $`\gamma _{1\nu ,k}`$ of this function can be obtained from those for $`C_\nu ^{AB}(\lambda ,z)`$ by replacements
$`T_\nu ^{AB}(\lambda ,z){\displaystyle \frac{z}{J_\nu ^2(z)/J_\nu ^2(\lambda z)1+\nu ^2/z^2}}`$ (4.26)
$`\overline{f}(z)f^{}(z),\overline{f}(\lambda z)f(\lambda z),f=J_\nu ,Y_\nu ,H_\nu ^{(1,2)},I_\nu ,K_\nu ,C_\nu ^{AB}C_{1\nu }.`$
The physiacl apllications of the formulae derived in this section will be considered below in Sections 9 and 12.
## 5 Applications to integrals involving Bessel functions
The applications of GAPF to infinite integrals involving some combinations of Bessel functions lead to the interesting results . First of all one can express integrals over Bessel functions through the integrals involving modified functions. Let us substitute in the formula (3.22)
$$f(z)=F(z)\overline{J}_\nu (z).$$
(5.1)
For the function $`F(z)`$ having no poles at $`z=\lambda _{\nu ,k}`$ the sum over zeros of $`\overline{J}_\nu (z)`$ is zero. The conditions (3.4) and (3.15) may be written in terms of $`F(z)`$ as
$$|F(z)|<\epsilon _1(x)e^{c_1|y|}\text{or}|F(z)|<M|z|^{\alpha _1}e^{|y|},|z|\mathrm{},$$
(5.2)
with $`c_1<1`$, $`x^{1/2\delta _{B0}}\epsilon _1(x)0`$ for $`x\mathrm{}`$, $`\alpha _1>\alpha _0=3/2\delta _{B0}`$, and
$$F(ze^{\pi i})=e^{\nu \pi i}F(z)+o\left(z^{|\mathrm{Re}\nu |1}\right).$$
(5.3)
Hence for the function $`F(z)`$ satisfying conditions (5.2) and (5.3) it follows from (3.22) that
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}F(x)\overline{J}_\nu (x)𝑑x=r_{1\nu }\left[F(z)\overline{J}_\nu (z)\right]+{\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}F(z)\overline{Y}_\nu (z)+`$
$`+{\displaystyle \frac{1}{\pi }}{\displaystyle _0^{\mathrm{}}}\overline{K}_\nu (x)\left[e^{\nu \pi i/2}F(xe^{\pi i/2})+e^{\nu \pi i/2}F(xe^{\pi i/2})\right]𝑑x.`$ (5.4)
In expression (3.13) for $`r_{1\nu }`$ the points $`z_k`$ are poles of the meromorpic function $`F(z),\mathrm{Re}z_k>0`$. On the base of Remark after Theorem 3 the formula (5.4) may be generalized for the functions $`F(z)`$ with purely imaginary poles $`\pm iy_k`$, $`y_k>0`$ and satisfying condition
$$F(ze^{\pi i})=e^{\nu \pi i}F(z)+o\left((ziy_k)^1\right),z\pm iy_k.$$
(5.5)
The corresponding formula is obtained from (5.4) by adding residue terms for $`z=iy_k`$ in the form of (5.7) (see below) and taking the principal value of the integral on the right.
The same substitution (5.1) with the function $`F(z)`$ satisfying the conditions (5.2) and
$$F(xe^{\pi i/2})=e^{\nu \pi i}F(xe^{\pi i/2})$$
(5.6)
for real $`x`$, into the formula (3.37) yields the following result
$$\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)\overline{J}_\nu (x)𝑑x=r_{1\nu }\left[F(z)\overline{J}_\nu (z)\right]+\frac{\pi i}{2}\underset{\eta _k=0,iy_k}{}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}F(z)\overline{H}_\nu ^{(1)}(z).$$
(5.7)
In (3.13) now summation is over the poles $`z_k`$, $`\mathrm{Re}z_k>0`$ of the meromorphic function $`F(z)`$, and $`\pm iy_k,y_k>0`$ are purely imaginary poles of this function. Recall that the possible real poles of $`F(z)`$ are such, that integral on the left of (5.7) exists.
For the functions $`F(z)=z^{\nu +1}\stackrel{~}{F}(z)`$, with $`\stackrel{~}{F}(z)`$ being analytic in the right half-plane and even along the imaginary axis, $`\stackrel{~}{F}(ix)=\stackrel{~}{F}(ix)`$, one obtains
$$_0^{\mathrm{}}x^{\nu +1}\stackrel{~}{F}(x)\overline{J}_\nu (x)𝑑x=0.$$
(5.8)
This result for $`B=0`$ (see (3.1)) have been given previously in . The another result of is obtained from (5.7) choosing $`F(z)=z^{\nu +1}\stackrel{~}{F}(z)/(z^2a^2)`$.
Formulae similar to (5.4) and (5.7) can be derived for Neumann function $`Y_\nu (z)`$. Let for the function $`F(z)`$ the integtral $`\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)\overline{Y}_\nu (x)𝑑x`$ exists. Let us substitute in the formula (2.11)
$$f(z)=Y_\nu (z)F(z),g(z)=iJ_\nu (z)F(z)$$
(5.9)
and consider the limit $`a+0`$. The summands containing residues may be presented in the form
$`R[f(z),g(z)]`$ $`=`$ $`\pi {\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k>0}H_\nu ^{(1)}(z)F(z)+\pi {\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k<0}H_\nu ^{(2)}(z)F(z)+`$ (5.10)
$`+`$ $`\pi {\displaystyle \underset{k}{}}\mathrm{Res}_{\mathrm{Im}z_k=0}J_\nu (z)F(z)r_{3\nu }[F(z)],`$
where $`z_k`$ ($`\mathrm{Re}z_k>0`$) are the poles of $`F(z)`$ in the right half-plane. Now the following results can be prooved by using (2.11):
1) If the meromorphic function $`F(z)`$ has no poles on the imaginary axis and satisfy the condition (5.2) then
$$\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)Y_\nu (x)𝑑x=r_{3\nu }[F(z)]\frac{i}{\pi }_0^{\mathrm{}}K_\nu (x)\left[e^{\nu \pi i/2}F(xe^{\pi i/2})e^{\nu \pi i/2}F(xe^{\pi i/2})\right]𝑑x$$
(5.11)
and
2) If the meromorphic function $`F(z)`$ satisfy the conditions
$$F(xe^{\pi i/2})=e^{\nu \pi i}F(xe^{\pi i/2})$$
(5.12)
and (5.2) then one has
$$\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)Y_\nu (x)𝑑x=r_{3\nu }[F(z)]+\pi \underset{k}{}\mathrm{Res}_{z=iy_k}H_\nu ^{(1)}(z)F(z),$$
(5.13)
where $`\pm iy_k,y_k>0`$ are purely imaginary poles of $`F(z)`$.
From (5.13) it directly follows that for $`F(z)=z^\nu \stackrel{~}{F}(z)`$, with $`\stackrel{~}{F}(z)`$ being even along the imaginary axis, $`\stackrel{~}{F}(ix)=\stackrel{~}{F}(ix)`$, and analytic in the right half-plane
$$\mathrm{p}.\mathrm{v}._0^{\mathrm{}}x^\nu \stackrel{~}{F}(x)Y_\nu (x)𝑑x=0,$$
(5.14)
if the condition (5.2) takes place.
Let us consider more general case. Let the function $`F(z)`$ satisfy the condition
$$F(ze^{\pi i})=e^{\lambda \pi i}F(z)$$
(5.15)
for $`\mathrm{arg}z=\pi /2`$. In GAPF as functions $`f(z)`$ and $`g(z)`$ we choose
$`f(z)`$ $`=`$ $`F(z)\left[J_\nu (z)\mathrm{cos}\delta +Y_\nu (z)\mathrm{sin}\delta \right]`$
$`g(z)`$ $`=`$ $`iF(z)\left[J_\nu (z)\mathrm{sin}\delta Y_\nu (z)\mathrm{cos}\delta ,\delta =(\lambda \nu )\pi /2\right],`$ (5.16)
with $`g(z)(1)^kf(z)=H_\nu ^{(k)}(z)F(z)\mathrm{exp}[(1)^ki\delta ]`$, $`k=1,2`$. It can be seen that for such a choice the integral on rhs of (2.11) for $`a0`$ is equal to
$$\pi i\underset{\eta _k=iy_k}{}\mathrm{Res}_{z=\eta _k}H_\nu ^{(1)}(z)F(z)e^{i\delta },$$
(5.17)
where $`\pm iy_k,y_k>0`$, as above, are purely imaginary poles of $`F(z)`$. Substituting (5.16) into (2.11) and using (2.3) we obtain
Corollary 4. Let $`F(z)`$ be meromorphic function for $`\mathrm{Re}z0`$ (except possibly at $`z=0`$) with poles $`z_k,\pm iy_k`$; $`y_k,\mathrm{Re}z_k>0`$. If this function satisfies conditions (5.2) (for $`B=0`$) and (5.15) then
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}F(x)[J_\nu (x)\mathrm{cos}\delta +Y_\nu (x)\mathrm{sin}\delta ]dx=\pi i\{{\displaystyle \underset{z_k}{}}\mathrm{Res}_{\mathrm{Im}z_k>0}H_\nu ^{(1)}(z)F(z)e^{i\delta }`$
$`{\displaystyle \underset{z_k}{}}\mathrm{Res}_{\mathrm{Im}z_k<0}H_\nu ^{(2)}(z)F(z)e^{i\delta }i{\displaystyle \underset{z_k}{}}\mathrm{Res}_{\mathrm{Im}z_k=0}\left[J_\nu (z)\mathrm{sin}\delta Y_\nu (z)\mathrm{cos}\delta \right]F(z)+`$
$`+{\displaystyle \underset{\eta _k=iy_k}{}}\mathrm{Res}_{z=\eta _k}H_\nu ^{(1)}(z)F(z)e^{i\delta }\},`$ (5.18)
where it is assumed that integral on the left exists.
In particular for $`\delta =\pi n,n=0,1,2\mathrm{}`$ the formula (5.7) follows from here in the case $`B=0`$.
One will find a great many particular cases of the formulae (5.7) and (5.18) looking at the standard books and tables of known integrals with Bessel functions (see, e.g., ). Some special examples are given in the next section.
## 6 Integrals involving Bessel functions: Illustrations of general formulae
To illustrate the applications of the general formulae from previous section first of all consider integrals involving the function $`\overline{J}_\nu (z)`$. Let us introduce the functional
$$A_{\nu m}[G(z)]\mathrm{p}.\mathrm{v}._0^{\mathrm{}}z^{\nu 2m1}G(z)\overline{J}_\nu (z)𝑑z$$
(6.1)
with $`m`$ being an integer. Let $`F_1(z)`$ be an analytic function in the right half-plane satisfying condition
$$F_1(xe^{\pi i/2})=F_1(xe^{\pi i/2}),F_1(0)0$$
(6.2)
(the case when $`F_1(z)z^q,z0`$ with an integer $`q`$ can be reduced to this one by redefinitions of $`F_1(z)`$ and $`m`$). From (5.7) the following results can be obtained
$`A_{\nu m}[F_1(z)]`$ $`=`$ $`A_{\nu m}^{(0)}[F_1(z)]{\displaystyle \frac{\pi (1+\mathrm{sgn}m)}{4(2m)!}}\left({\displaystyle \frac{d}{dz}}\right)^{2m}\left[z^\nu \overline{Y}_\nu (z)F_1(z)\right]|_{z=0}`$ (6.3)
$`A_{\nu m}\left[{\displaystyle \frac{F_1(z)}{z^2a^2}}\right]`$ $`=`$ $`{\displaystyle \frac{\pi }{2}}a^{\nu 2m2}\overline{Y}_\nu (a)F_1(a)+A_{\nu m}^{(0)}\left[{\displaystyle \frac{F_1(z)}{z^2a^2}}\right],`$ (6.4)
$`A_{\nu m}\left[{\displaystyle \frac{F_1(z)}{z^4a^4}}\right]`$ $`=`$ $`{\displaystyle \frac{a^{\nu 2m4}}{2}}\left[{\displaystyle \frac{\pi }{2}}\overline{Y}_\nu (a)F_1(a)(1)^m\overline{K}_\nu (a)F_1(ia)\right]+`$ (6.5)
$`+A_{\nu m}^{(0)}\left[{\displaystyle \frac{F_1(z)}{z^4a^4}}\right],`$
$`A_{\nu m}\left[{\displaystyle \frac{F_1(z)}{\left(z^2c^2\right)^{p+1}}}\right]`$ $`=`$ $`{\displaystyle \frac{\pi i}{2^{p+1}p!}}\left({\displaystyle \frac{d}{cdc}}\right)^p\left[c^{\nu 2m2}F_1(c)H_\nu ^{(1)}(c)\right]+A_{\nu m}^{(0)}\left[{\displaystyle \frac{F_1(z)}{\left(z^2c^2\right)^{p+1}}}\right]`$ (6.6)
$`A_{\nu m}\left[{\displaystyle \frac{F_1(z)}{\left(z^2+a^2\right)^{p+1}}}\right]`$ $`=`$ $`{\displaystyle \frac{(1)^{m+p+1}}{2^pp!}}\left({\displaystyle \frac{d}{ada}}\right)^p\left[a^{\nu 2m2}K_\nu (a)F_1(ae^{\pi i/2})\right]+`$ (6.7)
$`+A_{\nu m}^{(0)}\left[{\displaystyle \frac{F_1(z)}{\left(z^2+a^2\right)^{p+1}}}\right],`$
and etc.(note that $`A_{\nu m}^{(0)}=0`$ for $`m<0`$). Here $`a>0`$, $`0<\mathrm{arg}c<\pi /2`$, and we have assumed that $`\mathrm{Re}\nu >0`$. To secure convergence at the origin the condition $`\mathrm{Re}\nu >m`$ is necessary. In the last two formulae we have used the identity
$$\left(\frac{d}{dz}\right)^p\left[\frac{zF(z)}{(z+b)^{p+1}}\right]_{z=b}=\frac{1}{2^{p+1}}\left(\frac{d}{bdb}\right)^pF(b).$$
(6.8)
Note that (6.7) can be obtained also from (6.6) in the limit $`\mathrm{Re}c0`$. For the case $`F_1=1,m=1`$ of (6.7) see, for example, . In (6.3)-(6.7) as a function $`F_1(z)`$ we can choose:
* function (3.55) for $`\rho =1`$, $`\mathrm{Re}\nu <\mathrm{Re}\mu _l+2m+(n+1)/2\delta _{b1}+\delta _{B0}`$, $`b=b_l1,b_l>0`$;
* function (3.56) with $`\rho =1`$, $`\mathrm{Re}\nu <2(m+n)+1/2\delta _{b1}+\delta _{B0}`$, $`b=b_l1`$;
* function (3.57) for $`\rho =1`$, $`\mathrm{Re}\nu <2m\mathrm{Re}\mu _l+(n+1)/2\delta _{b1}+\delta _{B0}`$, $`\mu _l>0`$ is half of an odd integer, $`b=b_l1`$;
* function (3.58) for $`\mathrm{Re}\nu <2m+n|k_l|+1/2+\delta _{B0}\delta _{\stackrel{~}{a}1}`$, $`\stackrel{~}{a}=|a_lb_l|1`$, $`a_l0`$, $`k_l`$ \- integer.
Here we have written the conditions for (6.3). The corresponding ones for (6.4), (6.5), (6.6),(6.7) are obtained from these by adding on the rhs of inequalities for $`\mathrm{Re}\nu `$, respectively 2, 4, $`2(p+1)`$, $`2(p+1)`$. In (6.3)-(6.7) we can choose also any combinations of the functions (3.55)-(3.58) with appropriate conditions.
For concrete evaluations of $`A_{\nu m}^{(0)}`$ in special cases it is useful the following formula
$$\underset{z0}{lim}\left(\frac{d}{dz}\right)^{2m}f_1(z)=(2m1)!!\underset{z0}{lim}\left(\frac{d}{zdz}\right)^mf_1(z),$$
(6.9)
valid for the function $`f_1(z)`$ satisfying condition $`f_1(z)=f_1(z)+o(z^{2m})`$, $`z0`$. From here, for instance, it follows that for $`z0`$
$$\left(\frac{d}{dz}\right)^{2m}\left[z^\nu Y_\nu (bz)F_1(z)\right]=(2m1)!!\frac{2^{\nu m}}{\pi b^{\nu m}}\underset{k=0}{\overset{m}{}}\left(\genfrac{}{}{0pt}{}{m}{k}\right)2^k\frac{\mathrm{\Gamma }(\nu m+k)}{b^{2k}}\left(\frac{d}{zdz}\right)^kF_1(z),$$
(6.10)
where we have used the standard formula for the derivative $`(d/zdz)^n`$ of cylinder functions (see ). From (6.3) one obtains ($`B=0`$)
$`{\displaystyle _0^{\mathrm{}}}z^{\nu 2m1}J_\nu (z){\displaystyle \underset{l=1}{\overset{n}{}}}\left(z^2+z_l^2\right)^{\mu _l/2}J_{\mu _l}(b_l\sqrt{z^2+z_l^2})dz=`$
$`={\displaystyle \frac{\pi }{2^{m+1}m!}}\left({\displaystyle \frac{d}{zdz}}\right)^m\left[z^\nu Y_\nu (z){\displaystyle \underset{l=1}{\overset{n}{}}}\left(z^2+z_l^2\right)^{\mu _l/2}J_{\mu _l}(b_l\sqrt{z^2+z_l^2})\right]_{z=0},`$ (6.11)
for $`m0`$ and the integral is zero for $`m<0`$. Here $`\mathrm{Re}\nu >0`$, $`b_1^nb_l1`$, $`b_l>0`$, $`m<\mathrm{Re}\nu <_{l=1}^n\mathrm{Re}\mu _l+2m+(n+3)/2\delta _{b1}`$. In particular case $`m=0`$ the Gegenbauer integral follows from here . In the limit $`z_l0`$ from (6.11) the value of integral $`_0^{\mathrm{}}z^{\nu 2m1}J_\nu (z)_{l=1}^nz^{\mu _l}J_{\mu _l}(z)dz`$ can be obtained.
By using (5.18) the formulae similar to (6.3)-(6.7) may be derived for the integrals of type
$$B_\nu [G(z)]\mathrm{p}.\mathrm{v}._0^{\mathrm{}}G(z)\left[J_\nu (z)\mathrm{cos}\delta +Y_\nu (z)\mathrm{sin}\delta \right]𝑑z,\delta =(\lambda \nu )\pi /2.$$
(6.12)
It directly follows from Corollary 4 that for function $`F(z)`$ analytic for $`\mathrm{Re}z0`$ and satisfying conditions (5.2) and (5.15) the following formulae take place
$`B_\nu [F(z)]`$ $`=`$ $`0`$ (6.13)
$`B_\nu \left[{\displaystyle \frac{F(z)}{z^2a^2}}\right]`$ $`=`$ $`\pi F(a)\left[J_\nu (a)\mathrm{sin}\delta Y_\nu (a)\mathrm{cos}\delta \right]/2,`$ (6.14)
$`B_\nu \left[{\displaystyle \frac{F(z)}{z^4a^4}}\right]`$ $`=`$ $`{\displaystyle \frac{\pi }{4a^3}}F(a)\left[J_\nu (a)\mathrm{sin}\delta Y_\nu (a)\mathrm{cos}\delta \right]+{\displaystyle \frac{i}{2a^3}}K_\nu (a)F(ia)e^{i\lambda \pi /2},`$ (6.15)
$`B_\nu \left[{\displaystyle \frac{F_1(z)}{\left(z^2c^2\right)^{p+1}}}\right]`$ $`=`$ $`{\displaystyle \frac{\pi i}{2^{p+1}p!}}\left({\displaystyle \frac{d}{cdc}}\right)^p\left[c^1F(c)H_\nu ^{(1)}(c)\right]e^{i\delta }`$ (6.16)
$`B_\nu \left[{\displaystyle \frac{F_1(z)}{\left(z^2+a^2\right)^{p+1}}}\right]`$ $`=`$ $`{\displaystyle \frac{(1)^{p+1}}{2^pp!}}\left({\displaystyle \frac{d}{ada}}\right)^p\left[a^1F(ae^{\pi i/2})K_\nu (a)\right]e^{i\pi \lambda /2},`$ (6.17)
where $`a>0`$, $`0<\mathrm{arg}c\pi /2`$. To obtain the last two formulae we have used the identity (6.8). The formula (6.13) generalizes the result of (the cases $`\lambda =\nu `$ and $`\lambda =\nu +1`$). From the last formula taking $`F(z)=z^{\lambda 1}`$ we obtain result given in . In (6.13) - (6.17) as a function $`F(z)`$ one can choose (the constraints on parameters are written for the formula (6.13); the corresponding constraints for (6.14), (6.15), (6.16), (6.17) are obtained from given ones by adding the summands 2, 4, $`2(p+1)`$, $`2(p+1)`$ to the rhs of inequalities, correspondingly):
* function (3.55) for $`\rho =\lambda `$, $`|\mathrm{Re}\nu |<\mathrm{Re}\rho <\mathrm{Re}\mu _l+(n+3)/2\delta _{b1}`$, $`b=_lb_l1`$;
* function (3.56) for $`\rho =\lambda `$, $`|\mathrm{Re}\nu |<\mathrm{Re}\rho <3/2\delta _{b1}`$, $`b=_lb_l1`$;
* function
$`z^{\rho 1}{\displaystyle \underset{l=1}{\overset{n}{}}}\left[J_{\mu _l+k_l}(a_lz)Y_{\mu _l}(b_lz)Y_{\mu _l+k_l}(a_lz)J_{\mu _l}(b_lz)\right],\lambda =\rho +{\displaystyle \underset{l=1}{\overset{n}{}}}k_l,a_l>0,`$ (6.18)
$`|\mathrm{Re}\nu |+{\displaystyle |k_l|<\mathrm{Re}\rho <n+3/2\delta _{c1},c=|a_lb_l|}1,\mathrm{Re}\mu _l0`$
(for $`\mathrm{Re}\mu _l<0`$ one has $`k_l>|\mathrm{Re}\mu _l|`$).
Any combination of these functions with appropriate conditions on parameters can be choosed as well.
Now consider integrals which can be expressed via series by using (5.7) and (5.18). In (5.7) let us choose the function
$$F(z)=\frac{z^{\nu 2m}F_1(z)}{\mathrm{sinh}\pi z},$$
(6.19)
where $`F_1(z)`$ is the same as in the formulae (6.3) - (6.6). As the points $`\pm i,\pm 2i,\mathrm{}`$ are simple poles for $`F(z)`$ from (5.7) one obtains
$$_0^{\mathrm{}}\frac{z^{\nu 2m}}{\mathrm{sinh}(\pi z)}F_1(z)\overline{J}_\nu (z)𝑑z=A_{\nu m}^{(0)}\left[\frac{zF_1(z)}{\mathrm{sinh}(\pi z)}\right]+\frac{2}{\pi }\underset{k=1}{\overset{\mathrm{}}{}}(1)^{m+k}k^{\nu 2m}\overline{K}_\nu (k)F_1(ik),$$
(6.20)
where $`A_{\nu m}^{(0)}[f(z)]`$ is defined by (6.3) and $`\mathrm{Re}\nu >m`$. The corresponding costraints on $`F_1(z)`$ follow directly from (5.2). The particular case of this formula when $`F_1(z)=\mathrm{sinh}(az)/z`$ and $`m=1`$ is given in . As a function $`F_1(z)`$ here one can choose any of functions (3.55)-(3.58) with $`\rho =1`$ and $`\stackrel{~}{a},_lb_l<1`$. From (6.20) it follows that
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{z^{\nu 2m}}{\mathrm{sinh}(\pi z)}}J_\nu (z){\displaystyle \underset{l=1}{\overset{n}{}}}z^{\mu _l}I_{\mu _l}(b_lz)dz=A_{\nu m}^{(0)}\left[{\displaystyle \frac{z}{\mathrm{sinh}(\pi z)}}{\displaystyle \underset{l=1}{\overset{n}{}}}z^{\mu _l}I_{\mu _l}(b_lz)\right]+`$
$`+{\displaystyle \frac{2}{\pi }}{\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}(1)^{m+k}K_\nu (k){\displaystyle \underset{l=1}{\overset{n}{}}}k^{\mu _l}J_{\mu _l}(b_lk),b_l>0,\pi {\displaystyle \underset{l=1}{\overset{n}{}}}b_l>0,\mathrm{Re}\nu >m.`$ (6.21)
In similar way from (5.18) it can be derived the following formula
$$_0^{\mathrm{}}\frac{zF(z)}{\mathrm{sinh}(\pi z)}\left[J_\nu (z)\mathrm{cos}\delta +Y_\nu (z)\mathrm{sin}\delta \right]𝑑z=\frac{2i}{\pi }e^{i\lambda \pi /2}\underset{k=1}{\overset{\mathrm{}}{}}(1)^kkK_\nu (k)F(ik).$$
(6.22)
Constraints on the function $`F(z)`$ immediately follow from Corollary 4. Instead of this function we can choose the functions (3.55), (3.56), (3.58).
As it have been mentioned above adding residue terms $`\pi i\mathrm{Res}_{z=iy_k}F(z)\overline{H}_\nu ^{(1)}(z)`$ to the rhs of (5.4) this formula may be generalized for the functions having purely imaginary poles $`\pm iy_k`$, $`y_k>0`$, provided the condition (5.5) is satisfied. As an application let us choose
$$F(z)=\frac{z^\nu F_1(z)}{e^{2\pi z/b}1},F_1(z)=F_1(z),b>0$$
(6.23)
with an analytic function $`F_1(z)`$. The function (6.23) satisfy condition (5.5) and have poles $`\pm ikb`$, $`k=0,1,2\mathrm{}`$. The additional constraint directly follows from (5.2). Then one obtains
$$_0^{\mathrm{}}\frac{x^\nu J_\nu (x)}{e^{2\pi x/b}1}F_1(x)dx=\frac{2}{\pi }\underset{k=0}{\overset{\mathrm{}}{}}{}_{}{}^{}(bk)_{}^{\nu }K_\nu (bk)F_1(ibk)\frac{1}{\pi }_0^{\mathrm{}}x^\nu K_\nu (x)F_1(ix)dx,$$
(6.24)
where the prime indicates that the $`m=0`$ term is to be halved. For the particular case $`F_1(z)=1`$, using the relation
$$\underset{k=0}{\overset{\mathrm{}}{}}{}_{}{}^{}(bk)_{}^{\nu }K_\nu (bk)=\frac{\sqrt{\pi }}{b}2^\nu \mathrm{\Gamma }(\nu +1/2)\underset{n=0}{\overset{\mathrm{}}{}}{}_{}{}^{}[\left(\frac{2\pi n}{b}\right)^2+1]_{}^{\nu 1/2}$$
(6.25)
and the known value for the integral on the right, we immediately obtain the result given in . The relation (6.25) can be proved by using the formulae
$$K_\nu (z)=\frac{2^\nu \mathrm{\Gamma }(\nu +1/2)}{\sqrt{\pi }z^\nu }_0^{\mathrm{}}\frac{\mathrm{cos}ztdt}{(t^2+1)^{\nu +1/2}},\underset{k=\mathrm{}}{\overset{+\mathrm{}}{}}e^{ikz}=2\pi \underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}\delta (z2\pi n),$$
(6.26)
(for the integral representation of Macdonald’s function see ), $`\delta (z)`$ is the Dirac delta function.
## 7 Formulae for integrals involving $`J_\nu (z)Y_\mu (\lambda z)Y_\nu (z)J_\mu (\lambda z)`$
In this section we shall consider the applications of GAPF to the integrals involving the function $`J_\nu (z)Y_\mu (\lambda z)Y_\nu (z)J_\mu (\lambda z)`$. In the formula (2.11) we substitute
$$f(z)=\frac{1}{2i}F(z)\underset{l=1}{\overset{2}{}}(1)^l\frac{H_\mu ^{(l)}(\lambda z)}{H_\nu ^{(l)}(z)},g(z)=\frac{1}{2i}F(z)\underset{l=1}{\overset{2}{}}\frac{H_\mu ^{(l)}(\lambda z)}{H_\nu ^{(l)}(z)}.$$
(7.1)
For definiteness we consider the case $`\lambda >1`$ (for $`\lambda <1`$ the expression for $`g(z)`$ have to be choosen with opposite sign). The conditions (2.1) and (2.10) are satisfied if the function $`F(z)`$ is constrained by the one of the following two inequalities
$$|F(z)|<\epsilon (x)e^{c|y|},c<\lambda 1,\epsilon (x)0,x+\mathrm{}$$
(7.2)
or
$$|F(z)|<M|z|^\alpha e^{(\lambda 1)|y|},\alpha >1,|z|\mathrm{},z=x+iy.$$
(7.3)
Then from (2.11) it follows that for the function $`F(z)`$ meromorphic in $`\mathrm{Re}za>0`$ one has
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\mu (\lambda x)J_\mu (\lambda x)Y_\nu (x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}F(x)𝑑x=r_{1\mu \nu }[F(z)]+`$
$`+{\displaystyle \frac{1}{2i}}\left[{\displaystyle _a^{a+i\mathrm{}}}F(z){\displaystyle \frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}}𝑑z{\displaystyle _a^{ai\mathrm{}}}F(z){\displaystyle \frac{H_\mu ^{(2)}(\lambda z)}{H_\nu ^{(2)}(z)}}𝑑z\right],`$ (7.4)
where we have introduced the notation
$$r_{1\mu \nu }[F(z)]=\frac{\pi }{2}\underset{k}{}\mathrm{Res}_{\mathrm{Im}z_k=0}\left[F(z)\underset{l=1}{\overset{2}{}}\frac{H_\mu ^{(l)}(\lambda z)}{H_\nu ^{(l)}(z)}\right]+\pi \underset{k}{}\underset{l=1}{\overset{2}{}}\mathrm{Res}_{(1)^l\mathrm{Im}z_k<0}\left[F(z)\frac{H_\mu ^{(l)}(\lambda z)}{H_\nu ^{(l)}(z)}\right].$$
(7.5)
The most important case for the applications is the limit $`a0`$. The following statements take place :
Theorem 5. Let the function $`F(z)`$ be meromorphic for $`\mathrm{Re}z0`$ (except the possible branch point $`z=0`$) with poles $`z_k,\pm iy_k`$ ($`y_k,\mathrm{Re}z_k>0`$). If this function satisfy conditions (7.2) or (7.3) and
$$F(xe^{\pi i/2})=e^{(\mu \nu )\pi i}F(xe^{\pi i/2}),$$
(7.6)
then for values of $`\nu `$ for which the function $`H_\nu ^{(1)}(z)`$ ($`H_\nu ^{(2)}(z)`$) have no zeros for $`0argz\pi /2`$ ($`\pi /2argz0`$) the following formula is valid
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\mu (\lambda x)Y_\nu (x)J_\mu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}F(x)𝑑x=r_{1\mu \nu }[F(z)]+`$
$`+{\displaystyle \frac{\pi }{2}}{\displaystyle \underset{\eta _k=0,iy_k}{}}\left(2\delta _{0\eta _k}\right)\mathrm{Res}_{z=\eta _k}F(z){\displaystyle \frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}},`$ (7.7)
where it is assumed that the integral on the left exists.
Proof. From the condition (7.6) it follows that for $`\mathrm{arg}z=\pi /2`$
$$\frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}F(z)=\frac{H_\mu ^{(2)}(\lambda z_1)}{H_\nu ^{(2)}(z_1)}F(z_1),z_1=e^{\pi i},$$
(7.8)
and that the possible purely imaginary poles of $`F(z)`$ are conjugate: $`\pm iy_k,y_k>0`$. Hence in rhs of (7.4) in the limit $`a0`$ the term in the square brackets may be presented in the form (it can be seen similarly to (3.36))
$$\left(_{\gamma _\rho ^+}+\underset{k}{}_{C_\rho (iy_k)}\right)\frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}F(z)dz+\left(_{\gamma _\rho ^{}}+\underset{k}{}_{C_\rho (iy_k)}\right)\frac{H_\mu ^{(2)}(\lambda z)}{H_\nu ^{(2)}(z)}F(z)dz$$
(7.9)
with the same notations as in (3.36). By using (7.8) and the condition that the integral converges at the origin we obtain
$$_{\mathrm{\Omega }_\rho ^+(\eta _k)}\frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}F(z)𝑑z+_{\mathrm{\Omega }_\rho ^{}(\eta _k)}\frac{H_\mu ^{(2)}(\lambda z)}{H_\nu ^{(2)}(z)}F(z)𝑑z=\left(2\delta _{0\eta _k}\right)\pi i\mathrm{Res}_{z=\eta _k}\frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}F(z),$$
(7.10)
where $`\mathrm{\Omega }_\rho ^\pm (0)=\gamma _\rho ^\pm `$, $`\mathrm{\Omega }_\rho ^\pm (iy_k)=C_\rho (\pm iy_k)`$. By using this relation from (7.4) we receive the formula (7.7).
Note that one can write the residue at $`z=0`$ in the form
$$\mathrm{Res}_{z=0}\frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}F(z)=\mathrm{Res}_{z=0}\frac{J_\nu (z)J_\mu (\lambda z)+Y_\nu (z)Y_\mu (\lambda z)}{J_\nu ^2(z)+Y_\nu ^2(z)}F(z)$$
(7.11)
as well. Integrals of type (7.7) we have been able to find in literature (see, e.g., ) are special cases of this formula. For example, taking $`F(z)=J_\nu (z)Y_{\nu +1}(\lambda ^{}z)Y_\nu (z)J_{\nu +1}(\lambda ^{}z)`$ for the integral on the left in (7.7) we obtain $`\lambda ^\nu \lambda ^{\nu 1}`$ for $`\lambda ^{}<\lambda `$ and $`\lambda ^\nu \lambda ^{\nu 1}\lambda ^\nu \lambda ^{\nu 1}`$ for $`\lambda ^{}>\lambda `$ . By taking $`z^{2m+1}/(z^2+a^2)`$, $`z^{2m+1}/(z^2c^2)`$ as $`F_1(z)`$ for $`\mu =\nu `$ and integer $`m0`$ one receive
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\nu (\lambda x)Y_\nu (x)J_\nu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}{\displaystyle \frac{x^{2m+1}}{x^2+a^2}}𝑑x`$ $`=`$ $`(1)^ma^{2m}{\displaystyle \frac{\pi }{2}}{\displaystyle \frac{K_\nu (\lambda a)}{K_\nu (a)}},\mathrm{Re}a>0`$ (7.12)
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\nu (\lambda x)Y_\nu (x)J_\nu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}{\displaystyle \frac{x^{2m+1}}{x^2c^2}}𝑑x`$ $`=`$ $`{\displaystyle \frac{\pi }{2}}c^{2m}{\displaystyle \frac{J_\nu (c)J_\nu (\lambda c)+Y_\nu (c)Y_\nu (\lambda c)}{J_\nu ^2(c)+Y_\nu ^2(c)}}`$ (7.13)
where $`c>0,\lambda >1`$. The particular cases of this formula for $`\nu =m=0`$ are given in . In (7.12) taking the limit $`a0`$ and choosing $`m=0`$ we obtain the integral of this type given in . In (7.7) as a function $`F(z)`$ we can choose (3.55), (3.56), (3.58) (the corresponding conditions for paremeters directly follow from (7.2) or (7.3)) with $`\rho =\mu \nu 2m`$ ($`m`$ \- integer), as well as any products between them and with $`_{l=1}^n(z^2c_l^2)^{k_l}`$. For instance,
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\nu (\lambda x)Y_\nu (x)J_\nu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}{\displaystyle \underset{l=1}{\overset{n}{}}}{\displaystyle \frac{J_{\mu _l}(b_l\sqrt{x^2+z_l^2})}{\left(x^2+z_l^2\right)^{\mu _l/2}}}{\displaystyle \frac{dx}{x}}={\displaystyle \frac{\pi }{2\lambda ^\nu }}{\displaystyle \underset{l=1}{\overset{n}{}}}z^{\mu _l}J_{\mu _l}(b_lz_l),`$ (7.14)
$`b_l,\mathrm{Re}\nu >0,\mathrm{Re}z_l0,\lambda >1,{\displaystyle \underset{l=1}{\overset{n}{}}}\mathrm{Re}\mu _l+n/2+1>\delta _{b,\lambda 1},b{\displaystyle \underset{l=1}{\overset{n}{}}}b_l\lambda 1`$
As another consequence of (7.4) one has:
Theorem 6. Let $`F(z)`$ be meromorphic in the right half-plane (with possible exception $`z=0`$) with poles $`z_k,\mathrm{Re}z_k>0`$, and satisfy conditions (7.2) or (7.3) and
$$F(ze^{\pi i})=e^{(\mu \nu )\pi i}F(z)+o\left(z^{|\mathrm{Re}\mu ||\mathrm{Re}\nu |1}\right),z0,$$
(7.15)
then for values of $`\nu `$ for which the function $`H_\nu ^{(1)}(z)`$ ($`H_\nu ^{(2)}(z)`$) have no zeros for $`0argz\pi /2`$ ($`\pi /2argz0`$) the following formula takes place
$`\mathrm{p}.\mathrm{v}.{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\mu (\lambda x)Y_\nu (x)J_\mu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}F(x)𝑑x=r_{1\mu \nu }[F(z)]+{\displaystyle \frac{\pi }{2}}\mathrm{Res}_{z=0}{\displaystyle \frac{H_\mu ^{(1)}(\lambda z)}{H_\nu ^{(1)}(z)}}F(z)+`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{K_\mu (\lambda x)}{K_\nu (x)}}\left[e^{(\nu \mu )\pi i/2}F(xe^{\pi i/2})+e^{(\mu \nu )\pi i/2}F(xe^{\pi i/2})\right]𝑑x,\lambda >1,`$ (7.16)
provided the integral on the left exists.
Proof. This result immediately follows from (7.4) in the limit $`a0`$ and from (7.10) with $`\eta _k=0`$.
For example, by using (7.16) one obtains
$`{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{J_\nu (x)Y_\mu (\lambda x)Y_\nu (x)J_\mu (\lambda x)}{J_\nu ^2(x)+Y_\nu ^2(x)}}{\displaystyle \underset{l=1}{\overset{n}{}}}J_{\mu _l}(b_lx)dx=\mathrm{cos}\mu _s{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{K_\mu (\lambda x)}{K_\nu (x)}}{\displaystyle \underset{l=1}{\overset{n}{}}}I_{\mu _l}(b_lx)dx,`$ (7.17)
$`{\displaystyle \underset{l=1}{\overset{n}{}}}\mathrm{Re}\mu _l+|\mathrm{Re}\nu |>|\mathrm{Re}\mu |1,b={\displaystyle \underset{l=1}{\overset{n}{}}}b_l\lambda 1,b_l>0,n>\delta _{b,\lambda 1},\mu _s\nu \mu +{\displaystyle \underset{l=1}{\overset{n}{}}}\mu _l.`$
Such relations are convenient in numerical calculations of integrals on the left as the subintegrand on the right at infinity goes to zero exponentially fast.
We have considered the formulas containing $`J_\nu (z)Y_\mu (\lambda z)Y_\nu (z)J_\mu (\lambda z)`$. The similar results can be obtained for integrals containing the functions $`J_\nu ^{}(z)Y_\mu (\lambda z)Y_\nu ^{}(z)J_\mu (\lambda z)`$ and $`J_\nu ^{}(z)Y_\mu ^{}(\lambda z)Y_\nu ^{}(z)J_\mu ^{}(\lambda z)`$.
## 8 Applications to the Casimir effect. Vacuum energy density and stress inside a perfectly conducting spherical shell
In this and next sections we shall consider applications of the summations formulae obtained in previous sections to the physical problem, namely the Casimir effect. In what follows on the example of spherical and cylindrical geometries we will show that the using of GAPF allows to obtain the regularized values of physical quantities in cases then the explicit dependence of eigenmodes on quantum numbers is complicated and irregular.
Historically the investigation of the Casimir effect for a perfectly conducting spherical shell was motivated by Casimir semiclassical model of an electron. In this model Casimir suggested that Poincare stress to stabilize the charged particle could arise from vacuum quantum fluctuations and the fine structure constant can be determined by a balance between the Casimir force (assumed attractive) and the Coulomb repulsion. However, as it have been shown by Boyer , the Casimir energy for the sphere is positive, implying a repulsive force. This result has later been reconsidered by a number of authors . More recently new methods have been developed for this problem including a direct mode summation and zeta function approaches. However the main part of studies have focused on global quantities such as total energy. The investigation of the energy distribution inside a perfectly reflecting spherical shell was made in in the case of QED and in for QCD. The distribution of the other components for the electromagnetic EMT inside as well as outside the shell can be obtained from the results of . In these papers the consideration was carried out in terms of Schwinger’s source theory. In the calculations of the regularized values for vev of the EMT components inside and outside the perfectly conducting spherical shell are based on the generalized Abel-Plana summation formula. Our consideration below is based on this approach.
The main quantities we will consider here are vacuum expectation values (vev) of the energy-momentum tensor (EMT) for the electromagnetic field inside a perfectly conducting spherical shell of radius $`a`$. It may be obtained by using the standard formula of mode summation
$$0|T_{ik}|0=\underset{\alpha }{}T_{ik}(x)\{\mathrm{\Psi }_\alpha ^{()}(x),\mathrm{\Psi }_\alpha ^{(+)}(x)\},$$
(8.1)
where bilinear form $`T_{ik}\{f,g\}`$ for a field $`\mathrm{\Psi }`$ is given by the classical EMT. Here $`|0`$ is the amplitude of the vacuum state, $`\left\{\mathrm{\Psi }_\alpha ^{(\pm )}(x)\right\}`$ is a complete set of the positive and negative frequency solutions to the field equations, satisfying the boundary conditions, and subscript $`\alpha `$ may contain discrete and continous components.
In the case of the electromagnetic field inside the perfectly conducting sphere, by using Coulomb gauge for vector potential $`𝐀`$, the corresponding system of solutions, regular at $`r=0`$, can be presented in the form
$$𝐀_\alpha =\omega ^1\beta _l(a,\omega )\{\begin{array}{cc}j_l(\omega r)𝐗_{lm}e^{i\omega t}\hfill & \text{if }\lambda =0\hfill \\ \omega ^1\times \left[j_l(\omega r)𝐗_{lm}\right]e^{i\omega t}\hfill & \text{if }\lambda =1\hfill \end{array},\alpha =(\omega lm\lambda ),$$
(8.2)
where $`\lambda =0`$ and 1 correspond to the spherical waves of electric and magnetic type (TM and TE - modes). They describe photon with definite values of total momentum $`l`$, its projection $`m`$, energy $`\omega `$ and parity $`(1)^{l+\lambda +1}`$ (units $`\mathrm{}=c=1`$ are used). Here the vector spherical harmonics have the form
$$𝐗_{lm}(\theta ,\phi )=i\frac{𝐫\times }{\sqrt{l(l+1)}}Y_{lm}(\theta ,\phi ),l0,$$
(8.3)
with $`Y_{lm}`$ being spherical function, and $`j_l(x)=\sqrt{\pi /2x}J_{l+1/2}(x)`$ is spherical Bessel function. The coefficients $`\beta _l(a,\omega )`$ are determined by the normalization condition
$$𝑑V𝐀_\alpha 𝐀_\alpha ^{}^{}=\frac{2\pi }{\omega }\delta _{\alpha \alpha ^{}},$$
(8.4)
where the integration goes over the region inside the sphere. Using the standard formulae for vector spherical harmonics and spherical Bessel functions (see, for example, ) one finds
$$\beta _l^2(a,\omega )=8\omega ^3T_\nu (\omega a)/a,\nu =l+1/2,$$
(8.5)
where $`T_\nu (z)`$ is defined in (3.12).
Inside the perfectly conducting sphere the photon energy levels are quantized by standard boundary conditions:
$$𝐫\times 𝐄=0,𝐫𝐁=0,r=a,$$
(8.6)
where $`𝐄`$ and $`𝐁`$ are the electric and magnetic fields. They lead to the following eigenvalue equations with respect to $`\omega `$
$`j_l(\omega r)|_{r=a}`$ $`=`$ $`0\text{if }\lambda =0`$ (8.7)
$`{\displaystyle \frac{d}{dr}}\left[rj_l(\omega r)\right]_{r=a}`$ $`=`$ $`0\text{if }\lambda =1`$ (8.8)
It is well known that these equations have infinite number of real simple roots .
By substituting the eigenfunctions into (8.1) with the standard expression of the electromagnetic EMT and after the summation over $`m`$ by using the standard formulae for vector spherical harmonics (see, for example, ) one obtains
$$0|T_{ik}|0=\mathrm{diag}(\epsilon ,p,p_{},p_{})$$
(8.9)
(here index values 1,2,3 correspond to the spherical coordinates $`r,\theta ,\phi `$ with origin at the sphere centre). Energy density, $`\epsilon `$, pressures in transverse, $`p_{}`$, and radial, $`p`$, directions are determined by relations
$`q(a,r)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2a}}{\displaystyle \underset{\omega l\lambda }{}}\omega ^3T_\nu (\omega a)D_l^{(q)}(\omega r),q=\epsilon ,p,p_{},`$ (8.10)
$`p(a,r)`$ $`=`$ $`\epsilon 2p_{},`$ (8.11)
where the following notations are introduced
$$D_l^{(q)}(y)=\{\begin{array}{c}lj_{l+1}^2(y)+(l+1)j_{l1}^2(y)+(2l+1)j_l^2(y),q=\epsilon \hfill \\ l(l+1)(2l+1)j_l^2(y)/y^2,q=p_{}\hfill \end{array}$$
(8.12)
In the sum (8.10) $`\omega `$ takes discrete set of values determined by the equations (8.7) and (8.8). The relation (8.11) corresponds to the zero trace of the EMT.
The vev (8.10) are infinite. The renormalization of $`0|T_{ik}|0`$ in flat spacetime is affected by subtracting from this quantity its singular part $`\overline{0}|T_{ik}|\overline{0}`$, which is precisely the value it would have if the boundary were absent. Here $`|\overline{0}`$ is the amplitude for Minkowski vacuum state. To evaluate the finite difference between these two infinities we will introduce a cutoff function $`\psi _\mu (\omega )`$, which decreases with increasing $`\omega `$ and satisfies the condition $`\psi _\mu (\omega )1,\mu 0`$, and makes the sums finite. After subtracting we will allow $`\mu 0`$ and will show that the result does not depend on the form of cutoff:
$$\mathrm{reg}0|T_{ik}|0=\underset{\mu 0}{lim}\left[0|T_{ik}|0\overline{0}|T_{ik}|\overline{0}\right].$$
(8.13)
Hence we consider the following finite quantities
$$q(\mu ,a,r)=\frac{1}{4\pi ^2a^4}\underset{l=1}{\overset{\mathrm{}}{}}\underset{k=1}{\overset{\mathrm{}}{}}\underset{\lambda =0}{\overset{1}{}}j_{\nu ,k}^{(\lambda )3}T_\nu (j_{\nu ,k}^{(\lambda )})\psi _\mu (j_{\nu ,k}^{(\lambda )}/a)D_l^{(q)}(j_{\nu ,k}^{(\lambda )}x),x=r/a,$$
(8.14)
where $`\omega =j_{\nu ,k}^{(\lambda )}/a`$ are solutions to the eigenvalue equations (8.7) and (8.8) for $`\lambda =0,1`$, respectively. The summations over $`k`$ in (8.14) can be done by using the formula (3.26) and taking $`A=1,B=0`$ for TM-modes ($`\lambda =0`$) and $`A=1,B=2`$ for TE-modes ($`\lambda =1`$) (recall that in (3.22) $`\lambda _{\nu ,k}`$ are zeros of $`\overline{J}_\nu (z)`$ with bared quantities defined as (3.1)). Note that the resulting sums are of type (3.30). Let us substitute in formula (3.26)
$$f(z)=z^3\psi _\mu (z/a)D_l^{(q)}(zx),$$
(8.15)
with $`D_l^{(q)}(y)`$ defined from (8.12). We will assume the class of cutoff functions for which the function (8.15) satisfies conditions for Theorem 2, uniformly with respect to $`\mu `$ (the corresponding restrictions for $`\psi _\mu `$ can be easily found from these conditions using the asymptotic formulae for Bessel functions). Below for simplicity we will consider the functions with no poles. In this case (8.15) is analytic on the right-half plane of the complex variable $`z`$. The discussion on the conditions to cutoff functions under which the difference between divergent sum and integral exists and has a finite value independent any further details of cutoff function see . For TE- and TM-modes by choosing the constants $`A`$ and $`B`$ as mentioned above one obtaines
$$q=\frac{1}{8\pi ^2}\underset{l=1}{\overset{\mathrm{}}{}}\left\{2_0^{\mathrm{}}\omega ^3\psi _\mu (\omega )D_l^{(q)}(\omega r)𝑑\omega \frac{1}{a^4}_0^{\mathrm{}}\chi _\mu (z/a)F_l^{(q)}(z,x)𝑑z\right\},q=\epsilon ,p_{},p,$$
(8.16)
where for $`x<1`$ the functions $`F_l^{(q)}(z,x)`$ are defined as
$`F_l^{(\epsilon )}(z,x)`$ $`=`$ $`{\displaystyle \frac{z}{x^2}}\left[{\displaystyle \frac{e_l(z)}{s_l(z)}}+{\displaystyle \frac{e_l^{}(z)}{s_l^{}(z)}}\right]\left[ls_{l+1}^2(zx)+(l+1)s_{l1}^2(zx)(2l+1)s_l^2(zx)\right],`$ (8.17)
$`F_l^{(p_{})}(z,x)`$ $`=`$ $`(2l+1){\displaystyle \frac{l(l+1)}{zx^4}}\left[{\displaystyle \frac{e_l(z)}{s_l(z)}}+{\displaystyle \frac{e_l^{}(z)}{s_l^{}(z)}}\right]s_l^2(zx),`$ (8.18)
$`F_l^{(p)}(z,x)`$ $`=`$ $`F_l^{(\epsilon )}2F_l^{(p_{})},\chi _\mu (y)=\left[\psi _\mu (iy)+\psi _\mu (iy)\right]/2.`$ (8.19)
In these expressions we have introduced Ricatti-Bessel functions of imaginary argument,
$$s_l(z)=\sqrt{\frac{\pi z}{2}}I_\nu (z),e_l(z)=\sqrt{\frac{2z}{\pi }}K_\nu (z),\nu =l+1/2.$$
(8.20)
As $`\overline{0}|T_{ik}|\overline{0}=lim_a\mathrm{}0|T_{ik}(\mu ,a,r)|0`$ the first integral in (8.16) represents the vacuum EMT for empty Minkowski spacetime:
$$q=\frac{1}{4\pi ^2}\underset{l=1}{\overset{\mathrm{}}{}}_0^{\mathrm{}}\omega ^3\psi _\mu (\omega )D_l^{(q)}(\omega r)𝑑\omega .$$
(8.21)
This expression can be further simplified. For example in the case of the energy density one has
$`\epsilon ^{(0)}`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}{\displaystyle _0^{\mathrm{}}}\omega ^3\psi _\mu (\omega )\left[lj_{l+1}^2(\omega r)+(l+1)j_{l1}^2(\omega r)+(2l+1)j_l^2(\omega r)\right]𝑑\omega =`$ (8.22)
$`={\displaystyle \frac{1}{2\pi ^2}}{\displaystyle _0^{\mathrm{}}}\omega ^3\psi _\mu (\omega ){\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}(2l+1)j_l^2(\omega r)d\omega ={\displaystyle _0^{\mathrm{}}}\omega ^3\psi _\mu (\omega )𝑑\omega .`$
As we see the using of GAPF allows us to extract from infinite quantities the divergent part without specifying the form of cutoff function. Now the regularization of the EMT is equivalent to the omitting the first summand in (8.16), which as we saw corresponds to the contribution of the spacetime without boundaries. For the regularized components one obtains
$$\mathrm{reg}q(a,r)=\frac{1}{8\pi ^2a^4}\underset{l=1}{\overset{\mathrm{}}{}}_0^{\mathrm{}}\chi _\mu (z/a)F_l^{(q)}(z,x)𝑑z,r<a,q=\epsilon ,p,p_{}.$$
(8.23)
By using the recurrence relations for Riccati-Bessel modified functions the expressions for the regularized vacuum energy density and radial pressure may be presented in the form
$$q=\frac{1}{8\pi ^2a^2r^2}\underset{l=1}{\overset{\mathrm{}}{}}(2l+1)_0^{\mathrm{}}𝑑zz\chi _\mu (z/a)\left[\frac{e_l(z)}{s_l(z)}+\frac{e_l^{}(z)}{s_l^{}(z)}\right]\left\{s_l^{}_{}{}^{}2(zx)s_l^2(zx)\left[1\frac{l(l+1)}{(1)^iz^2x^2}\right]\right\},$$
(8.24)
where $`i=0`$ for $`q=\epsilon `$ and $`i=1`$ for $`q=p`$. Here we keep the cutoff factor because it plays an important role in the calculations of the total Casimir energy for the spherical shell (see below). The derivation of the vacuum densities (8.23), (8.24) given above uses GAPF to summarize mode sums and is based on . One can see that this formulae for the case of exponential cutoff function may be obtained also from the results of , where Green function method is used.
We obtained the regularized values (8.23) by introducing a cutoff function and susequent subtracting the contribution due to the unbounded space. The GAPF in the form (3.26) allows to obtain immediately this finite difference. However it should be noted that by using GAPF in the form (3.22) we can derive the expressions for the regularized azimuthal pressure without introducing any special cutoff function. To see this note that for $`x<1`$ the function (8.15) with $`q=p_{}`$ and $`\psi _\mu =1`$ satisfies conditions to Theorem 2. It follows from here that we can apply the formula (3.22) directly to the corresponding sum over $`\omega `$ in (8.10) or over $`k`$ in (8.14) (with $`\psi _\mu =1`$) without introducing the cutoff function. This immediately yields to the formula (8.23) for $`q=p_{}`$ with $`\psi _\mu =1`$.
Let us consider the behaviour of the functions $`F_l^{(q)}(z,x)`$ in various limiting cases. By using the corresponding formulae for Bessel functions one obtains:
(a) When $`l`$ is fixed and $`z0`$
$$F_l^{(\epsilon )}(z,x)\pi (l+1/2)x^{2(l1)},F_l^{(p_{})}(z,x)\pi lx^{2(l1)}/2.$$
(8.25)
(b) When $`l`$ is fixed and $`zx`$ is large
$$F_l^{(\epsilon )}2F_l^{(p_{})}l^2(l+1)^2(2l+1)\frac{\pi }{2(zx)^4}\mathrm{exp}[2z(1x)],$$
(8.26)
and the integral over $`z`$ in (8.23) converges for all $`x1`$;
(c) For large $`l`$ by using the uniform asymptotic expansions for Bessel functions one finds
$$F_l^{(q)}(\nu z,x)\mathrm{\Phi }_l^{(q)}(z,x)\mathrm{exp}\left\{2\nu [\eta (z)\eta (zx)]\right\},\nu =l+1/2$$
(8.27)
with
$`\mathrm{\Phi }_l^{(\epsilon )}(z,x)`$ $`=`$ $`{\displaystyle \frac{\pi \nu }{x^3}}t(zx)t^3(z)\left\{1{\displaystyle \frac{1}{12\nu }}\left[t(z)(t^2(z)+3)+t(zx)(5t^2(zx)9)\right]\right\},`$ (8.28)
$`\mathrm{\Phi }_l^{(p)}(z,x)`$ $`=`$ $`{\displaystyle \frac{\pi }{2x^3}}t^2(zx)t^3(z),`$ (8.29)
where the standard notations are used:
$$t(z)=\frac{1}{\sqrt{1+z^2}},\eta (z)=\sqrt{1+z^2}+\mathrm{ln}\frac{z}{1+\sqrt{1+z^2}}.$$
(8.30)
From these asymptotic formulae it follows that in (8.23) and (8.24) the rhs is finite for $`x<1`$ and the cutoff may be removed by putting $`\chi _\mu =1,\mu 0`$. From here it is obvious the independence of the regularized quantities on the specific form of the cutoff, on class of functions for which (8.15) satisfy conditions for (3.26).
From (8.27), (8.28) and (8.29) it follows that the vev of the EMT diverge at sphere surface, $`x1`$, due to the contribution of large $`l`$ (note that, as it follows from (8.28) and (8.29), in (8.23) with $`\chi _\mu =1`$ the integral over $`z`$ converges at $`x=1`$). The corresponding asymptotic behaviour can be found by using the uniform asymptotic expansions given above and the leading terms have the form
$$\epsilon 2p_{}\frac{1}{30\pi ^2a(ar)^3},p\frac{1}{60\pi ^2a^2(ar)^2}.$$
(8.31)
These surface divergences originate in the unphysical nature of perfect conductor boundary conditions and are well known in quantum field theory with boundaries. They are investigated in detail for various types of fields and general shape of the boundary . Eqs. (8.31) are particular cases of the asymptotic expansions for EMT vev near the smooth boundary given in these papers. In reality the expectation values for the EMT components will attain a limiting value on the conductor surface, which will depend on the molecular details of the conductor. From the asymptotic expansions given above it follows that the main contributions to $`q(r)`$ are due to the frequencies $`\omega <(ar)^1`$. Hence we expect that the formulae (8.23) are valid for real conductors up to distances $`r`$ for which $`(ar)^1\omega _0`$, with $`\omega _0`$ being the characteristic frequency, such that for $`\omega >\omega _0`$ the conditions for perfect conductivity are failed.
At the sphere centre in (8.23) $`l=1`$ multipole contributes only and we obtain
$`\epsilon (0)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi ^2a^4}}{\displaystyle _0^{\mathrm{}}}𝑑zz^3\left[\left({\displaystyle \frac{z1}{z+1}}e^{2z}+1\right)^1\left({\displaystyle \frac{z^2z+1}{z^2+z+1}}e^{2z}1\right)^1\right]=0.0381a^4,`$
$`p(0)`$ $`=`$ $`p_{}(0)=\epsilon (0)/3.`$ (8.32)
At centre the equation of state for the electromagnetic vacuum is the same as that for blackbody radiation. Note that the corresponding results obtained using the uniform asymptotic expansions for Bessel functions are in good agreement with (8.32).
The components of the regularized EMT satisfy continuity equation $`T_{i;k}^k=0`$, which for the spherical geometry takes the form
$$p^{}(r)+\frac{2}{r}(pp_{})=0.$$
(8.33)
From here by using the zero trace condition the following integral relations may be obtained
$$p(r)=\frac{1}{r^3}_0^r\epsilon (t)t^2𝑑t=\frac{2}{r^2}_0^rp_{}(t)t𝑑t,$$
(8.34)
where the integration constant is determined from the relations (8.32) at the sphere centre. It follows from the first relation that the total energy within a sphere with radius $`r`$ is equal to
$$E(r)=4\pi _0^r\epsilon (t)t^2𝑑t=3V(r)p(r),$$
(8.35)
where $`V(r)`$ is the corresponding volume.
The distribution for the vacuum energy density and pressures inside the perfectly conducting sphere can be obtained from the results of the numerical calculations given in . In their calculations Brevik and Kolbenstvedt use the uniform asymptotic expansions of Ricatti-Bessel functions for large values of order. In (see also ) the corresponding quantities are calculated on the base of the exact relations for these functions and the accuracy of the numerical results in is estimated ($`5\%`$). The simple approximataion formulae are presented with the same accuracy as asymptotic expressions. Note that inside the sphere all quantities $`\epsilon ,p,p_{}`$ are negative and corresponding vacuum forces tend to contract sphere.
## 9 Electromagnetic vacuum EMT outside a spherical shell
Now let us consider the electromagnetic vacuum in the region outside of a perfectly conducting sphere. To deal with discrete modes we firstly consider vacuum fields in the region between two cocentric conducting spherical shells with radii $`a`$ and $`b`$, $`a<b`$. Letting $`b\mathrm{}`$ we will obtain from here the result for the region under question.
By using Coulomb gauge the complete set of solutions to the Maxwell equations can be written in the form similar to (8.2)
$$𝐀_{\omega lm\lambda }(𝐫,t)=\frac{e^{i\omega t}}{\sqrt{4\pi }}\beta _{\lambda l}(a,b,\omega )\{\begin{array}{cc}\omega g_{0l}(\omega a,\omega r)𝐗_{lm}\hfill & \text{if }\lambda =0\hfill \\ \times \left[g_{1l}(\omega a,\omega r)𝐗_{lm}\right]\hfill & \text{if }\lambda =1\hfill \end{array},$$
(9.1)
where as above the values $`\lambda =0`$ and $`\lambda =1`$ correspond to the waves of magnetic (TE-modes) and electric (TM-modes) type,
$$g_{\lambda l}(x,y)=\{\begin{array}{cc}j_l(y)n_l(x)j_l(x)n_l(y)\hfill & \text{if }\lambda =0\hfill \\ j_l(y)[xn_l(x)]^{^{}}[xj_l(x)]^{^{}}n_l(y),\hfill & \text{if }\lambda =1\hfill \end{array},$$
(9.2)
with $`n_l(x)`$ being Neumann spherical function. From the standard boundary conditions at surfaces $`r=a`$ and $`r=b`$ one finds that possible energy levels of photon are solutions to the following equations
$$\left(\frac{d}{dr}\right)^\lambda \left[rg_{\lambda l}(\omega a,\omega r)\right]_{r=b}=0,\lambda =0,1.$$
(9.3)
All roots of these equations are real and simple .
The coefficients $`\beta _{\lambda l}`$ in (9.1) are determined from the normalization condition (8.4), where now the integration goes over the region between spherical shells, $`arb`$. By using the standard relations for spherical Bessel functions they can be presented in the form
$`\beta _{0l}^2`$ $`=`$ $`\omega a\left[{\displaystyle \frac{aj_l^2(\omega a)}{bj_l^2(\omega b)}}1\right]^1,\lambda =0`$ (9.4)
$`\beta _{1l}^2`$ $`=`$ $`{\displaystyle \frac{1}{\omega a}}\left\{{\displaystyle \frac{b\left[\omega aj_l(\omega a)\right]^{}_{}{}^{}2}{a\left[\omega bj_l(\omega b)\right]^{}_{}{}^{}2}}\left[1{\displaystyle \frac{l(l+1)}{\omega ^2b^2}}\right]1+{\displaystyle \frac{l(l+1)}{\omega ^2a^2}}\right\}^1,\lambda =1.`$ (9.5)
From (8.1) with the electromagnetic field EMT and functions (9.1) as a complete set of solutions one obtains the vev in the form (8.9) with
$$q(a,b,r)=\frac{1}{8\pi }\underset{\omega l\lambda }{}(2l+1)\omega ^4\beta _{\lambda l}^2f_{\lambda l}^{(q)}(\omega a,\omega r),q=\epsilon ,p,p_{},$$
(9.6)
where the frequencies $`\omega `$ are solutions to the equations (9.3), and
$`f_{\lambda l}^{(\epsilon )}(\omega a,\omega r)`$ $`=`$ $`\left[1+{\displaystyle \frac{l(l+1)}{\omega ^2r^2}}\right]g_{\lambda l}^2(\omega a,\omega r)+{\displaystyle \frac{1}{\omega ^2r^2}}\left[{\displaystyle \frac{d}{d(\omega r)}}\left(\omega rg_{\lambda l}(\omega a,\omega r)\right)\right]^2,`$ (9.7)
$`f_{\lambda l}^{(p_{})}(\omega a,\omega r)`$ $`=`$ $`{\displaystyle \frac{l(l+1)}{\omega ^2r^2}}g_{\lambda l}^2(\omega a,\omega r).`$ (9.8)
It is easy to see that the eigenvalue equations (9.3) can be written in terms of the function $`C_\nu ^{AB}`$, defined by (4.1), as
$$C_\nu ^{AB}(\eta ,\omega a)=0,\nu =l+1/2,\eta =b/a,A=1/(1+\lambda ),B=\lambda ,\lambda =0,1.$$
(9.9)
By this choice of constants $`A`$ and $`B`$ the normalization coefficients (9.4) and (9.5) are related with the function $`T_\nu ^{AB}`$ from (4.6) as:
$$\beta _{\lambda l}^2=T_\nu ^{AB}(\eta ,\omega a).$$
(9.10)
This allows to use the formulae from section 4 for the summation over eigenmodes.
As above to regularize the infinite quantities (9.6) we introduce a cutoff function $`\psi _\mu (\omega )`$ and consider the difference
$$\mathrm{reg}0|T_{ik}|0=\underset{\mu 0}{lim}\left[0|T_{ik}(\mu ,a,b)|0\underset{a0}{lim}\underset{b\mathrm{}}{lim}0|T_{ik}(\mu ,a,b)|0\right].$$
(9.11)
This procedure is equivalent to the subtraction of Minkowskian part without boundaries.
Hence instead of (9.6) we consider the finite quantities
$$q=\frac{1}{8\pi a^4}\underset{l=1}{\overset{\mathrm{}}{}}(2l+1)\underset{k=1}{\overset{\mathrm{}}{}}\underset{\lambda =0}{\overset{1}{}}\gamma _{\nu ,k}^{(\lambda )4}T_\nu ^{AB}(\eta ,\gamma _{\nu ,k}^{(\lambda )})\psi _\mu (\gamma _{\nu ,k}^{(\lambda )}/a)f_{\lambda l}^{(q)}(\gamma _{\nu ,k}^{(\lambda )},\gamma _{\nu ,k}^{(\lambda )}x),q=\epsilon ,p_{},$$
(9.12)
where $`x=r/a`$, and $`\omega a=\gamma _{\nu ,k}^{(\lambda )}`$ are solutions to the equations (9.3) or (9.9). To sum over $`k`$ we will use the formula (4.13) with
$$h(z)=z^4\psi _\mu (z/a)f_{\lambda l}^{(q)}(z,zx),$$
(9.13)
assuming a class of cutoff functions for which (9.13) satisfies to the conditions (4.4) and (4.11) uniformly with respect to $`\mu `$. The corresponding restrictions on $`\psi _\mu `$ can be obtained using the asymptotic formulae for Bessel functions. From (9.12) by applying to the sum over $`k`$ the formula (4.13) for the EMT components one obtains
$`q`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2a^4}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}(2l+1){\displaystyle \underset{\lambda =0}{\overset{1}{}}}\{{\displaystyle _0^{\mathrm{}}}z^3\psi _\mu (z/a){\displaystyle \frac{f_{\lambda l}^{(q)}(z,zx)}{\mathrm{\Omega }_{1\lambda l}(z)}}dz+`$ (9.14)
$`+{\displaystyle \frac{1}{x^2}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{e_l^{(\lambda )}(\eta z)}{e_l^{(\lambda )}(z)}}{\displaystyle \frac{z\chi _\mu (z/a)F_{\lambda l}^{(q)}(z,zx)}{\left[(/y)^\lambda G_{\lambda l}(z,y)\right]_{y=z\eta }}}dz\},`$
where we use the notations
$$e_l^{(\lambda )}(y)\left(\frac{d}{dy}\right)^\lambda e_l(y),s_l^{(\lambda )}(y)\left(\frac{d}{dy}\right)^\lambda s_l(y)$$
(9.15)
for the Riccati-Bessel functions derivatives,
$$\mathrm{\Omega }_{1\lambda l}(z)=\{\begin{array}{cc}j_l^2(z)+n_l^2(z),\hfill & \lambda =0\hfill \\ \left[zj_l(z)\right]^{}_{}{}^{}2+[zn_l(z)]^{}_{}{}^{}2,\hfill & \lambda =1\hfill \end{array},$$
(9.16)
and
$`G_{\lambda l}(x,y)`$ $`=`$ $`e_l^{(\lambda )}(x)s_l(y)e_l(y)s_l^{(\lambda )}(x),\lambda =0,1,`$ (9.17)
$`F_{\lambda l}^{(\epsilon )}(z,y)`$ $`=`$ $`\left[{\displaystyle \frac{}{y}}G_{\lambda l}(z,y)\right]^2+\left[{\displaystyle \frac{l(l+1)}{y^2}}1\right]G_{\lambda l}^2(z,y),`$ (9.18)
$`F_{\lambda l}^{(p_{})}(z,y)`$ $`=`$ $`l(l+1)G_{\lambda l}^2(z,y)/y^2.`$ (9.19)
The function $`\chi _\mu `$ is determined by (8.18). To obtain the vev for the EMT components outside of a single conducting spherical shell with radius $`a`$ let us consider the limit $`b\mathrm{}`$. In this limit the second integral on the right of formula (9.14) tends to zero (for large $`\eta =b/a`$ the subintegrand is proportional to $`e^{2\eta z}`$), whereas the first one does not depend on $`b`$. Hence one obtains
$$q=\frac{1}{8\pi ^2a^4}\underset{l=1}{\overset{\mathrm{}}{}}(2l+1)\underset{\lambda =0,1}{}_0^{\mathrm{}}z^3\psi _\mu (z/a)\frac{f_{\lambda l}^{(q)}(z,zx)}{\mathrm{\Omega }_{1\lambda l}(z)}𝑑z,q=\epsilon ,p_{}.$$
(9.20)
To regularize the expressions (9.20) we have to subtract the Minkowskian part, namely the expression (8.21). It can be easily seen that
$$\frac{f_{\lambda l}^{(q)}(z,zx)}{\mathrm{\Omega }_{1\lambda l}(z)}D_l^{(q)}(zx)=\frac{1}{2}\underset{m=1,2}{}\mathrm{\Omega }_{\lambda l}^{(m)}(z)D_l^{(mq)}(zx).$$
(9.21)
Here the functions $`D_l^{(mq)}(y)`$ are obtained from the relations (8.12) by replacing $`j_lh_l^{(m)}`$, with $`h_l^{(m)},m=1,2`$ being spherical Hankel functions, and
$$\mathrm{\Omega }_{\lambda l}^{(m)}(z)=\{\begin{array}{c}j_l(z)/h_l^{(m)}(z),\lambda =0\hfill \\ \left[zj_l(z)\right]^{^{}}/[zh_l^{(m)}(z)]^{^{}},\lambda =1\hfill \end{array}$$
(9.22)
The function $`h_l^{(1)}(z)k`$ ($`h_l^{(2)}(z)`$) has no zeros for $`0\mathrm{arg}z\pi /2`$ ($`\pi /2\mathrm{arg}z0`$) and from this it follows that
$$\underset{m=1,2}{}_0^{\mathrm{}}z^3\psi _\mu (z/a)\mathrm{\Omega }_{\lambda l}^{(m)}(z)D_l^{(mq)}(zx)𝑑z=2_0^{\mathrm{}}z^3\mathrm{Re}\left[\psi _\mu (iz/a)\mathrm{\Omega }_{\lambda l}^{(1)}(iz)D_l^{(1q)}(izx)\right]𝑑z.$$
(9.23)
By introducing Ricatti-Bessel functions (8.20), for the regularized components of the vacuum EMT outside the sphere we find
$$q(a,r)=\frac{1}{8\pi ^2a^4}\underset{l=1}{\overset{\mathrm{}}{}}_0^{\mathrm{}}\chi _\mu (z/a)F_l^{(q)}(z,x)𝑑z,r>a,q=\epsilon ,p_{},p,$$
(9.24)
where for $`x>1`$ the functions $`F_l^{(q)}(z,x)`$ are defined as
$`F_l^{(\epsilon )}(z,x)`$ $`=`$ $`{\displaystyle \frac{z}{x^2}}\left[{\displaystyle \frac{s_l(z)}{e_l(z)}}+{\displaystyle \frac{s_l^{}(z)}{e_l^{}(z)}}\right]\left[le_{l+1}^2(zx)+(l+1)e_{l1}^2(zx)(2l+1)e_l^2(zx)\right]`$ (9.25)
$`F_l^{(p_{})}(z,x)`$ $`=`$ $`(2l+1){\displaystyle \frac{l(l+1)}{zx^4}}\left[{\displaystyle \frac{s_l(z)}{e_l(z)}}+{\displaystyle \frac{s_l^{}(z)}{e_l^{}(z)}}\right]e_l^2(zx),F_l^{(p)}=F_l^{(\epsilon )}2F_l^{(p_{})}.`$ (9.26)
The exterior mode sum consideration given in this section follows . For the case of exponential cutoff function the formulae (9.24) and (9.26) can be obtained also from the results , where Green’s function formalism was used. Note that the expressions for the exterior components are obtained from the interior ones replacing $`s_li_l`$, $`i_ls_l`$. In particular, the exterior components can be presented in the form analog to (8.24).
Let us consider the behaviour of the functions $`F_l^{(q)}(z,x)`$ in various limiting cases. By using the corresponding formulae for Bessel functions one obtains:
(a) When $`l`$ is fixed and $`z0`$
$$F_l^{(\epsilon )}(z,x)\pi (l+1/2)x^{2(l+2)},F_l^{(p_{})}(z,x)\pi (l+1)x^{2(l+1)}/2.$$
(9.27)
(b) When $`l`$ is fixed and $`z`$ is large
$$F_l^{(\epsilon )}2F_l^{(p_{})}l^2(l+1)^2(2l+1)\frac{\pi }{2(zx)^4}\mathrm{exp}[2z(x1)];$$
(9.28)
(c) For large $`l`$ by using the uniform asymptotic expansions for Bessel functions one finds
$$F_l^{(q)}(\nu z,x)\mathrm{\Phi }_l^{(q)}(z,x)\mathrm{exp}\left\{2\nu [\eta (zx)\eta (z)]\right\},\nu =l+1/2$$
(9.29)
with
$`\mathrm{\Phi }_l^{(\epsilon )}(z,x)`$ $`=`$ $`{\displaystyle \frac{\pi \nu }{x^3}}t(zx)t^3(z)\left\{1+{\displaystyle \frac{1}{12\nu }}\left[t(z)(t^2(z)+3)+t(zx)(5t^2(zx)9)\right]\right\},`$ (9.30)
$`\mathrm{\Phi }_l^{(p)}(z,x)`$ $`=`$ $`{\displaystyle \frac{\pi }{2x^3}}t^2(zx)t^3(z),`$ (9.31)
whith notations (8.30).
It follows from here that for the values $`x>1`$ the expressions (9.24) are finite and hence cutoff may be removed. In this case the independence of the result on specific form of cutting function is obvious.
The expressions (9.24) with $`\psi _\mu =1`$ diverge at sphere surface. The leading terms of these divergences may be found using (9.29) and are as following
$$\epsilon 2p_{}\frac{1}{30\pi ^2a(ra)^3},p(r)\frac{1}{60\pi ^2a^2(ra)^2},$$
(9.32)
and
$$\underset{ra}{lim}(\epsilon /p_{})^{}=\underset{ra}{lim}(p/p_{})^{}=1.$$
(9.33)
Comparing (9.32) with (8.31) we see that the cancellation of interior and exterior leading divergent terms occurs in calculating the total energy and force acting on sphere. The same cancellations take place for the next subleading divergent terms as well (see below). Formulas (9.32) are particular cases of general asymptotic expansions of the vacuum EMT components for conformally invariant fields near an arbitrary smooth boundary given in .
For distances far from the sphere one finds
$$p_{}\frac{1}{4\pi ^2a^4x^7}_0^{\mathrm{}}z^2e_1^2(z)𝑑z=\frac{5a^3}{16\pi ^2r^7},\epsilon 4p\frac{a^3}{2\pi ^2r^7},ra.$$
(9.34)
The results of numerical calculations of the vacuum EMT components outside the sphere are given in . In calculations are carried out by using the uniform asymptotic expansions for Riccati-Bessel functions. The accuracy of this approximation is estimated in , where exact relations are used in numerical calculations. The simple approximating formulas with the same accuracy as those for the asymptotic calculations are presented as well. The energy density and azimuthal pressure are positive, and radial pressure is negative. The latter means that the exterior vacuum forces tend to expand sphere. As we will see below this dominates the interior contraction force.
Note that the continuity equation (8.33) now may be written in the following integral form
$$p(r)=\frac{1}{r^3}_{\mathrm{}}^r\epsilon (t)t^2𝑑t=\frac{2}{r^2}_{\mathrm{}}^rp_{}(t)t𝑑t,$$
(9.35)
where the integration constant is determined from the asymptotic relations (9.32). From (8.34) and (9.35) it follows that
$$E(a)=\epsilon (r)𝑑V=4\pi a^3[p(a)p(a+)],$$
(9.36)
where $`E(a)`$ is the total vacuum energy for a spherical shell with radius $`a`$, $`p(a\pm )=lim_{r0}p(a\pm r)`$. By using the expressions for $`p(r)`$ given above one can obtain the following formula for the total energy (the same result can be obtained also by integrating the energy density)
$`E(a)`$ $`=`$ $`{\displaystyle \frac{1}{2\pi a}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}(2l+1){\displaystyle _0^{\mathrm{}}}𝑑z\chi _\mu (z/a)z\left(\mathrm{ln}|s_l(z)e_l(z)|\right)^{}\left[1+\left({\displaystyle \frac{l(l+1)}{z^2}}+1\right){\displaystyle \frac{s_l(z)e_l(z)}{s_l^{}(z)e_l^{}(z)}}\right]=`$ (9.37)
$`=`$ $`{\displaystyle \frac{1}{2\pi a}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}(2l+1){\displaystyle _0^{\mathrm{}}}𝑑z\chi _\mu (z/a)z{\displaystyle \frac{d}{dz}}\mathrm{ln}\left\{1\left[s_l(z)e_l(z)\right]^{}_{}{}^{}2\right\}.`$
By taking the cutting function $`\psi _\mu (\omega )=e^{\mu \omega }`$ one obtains the expression for the Casimir energy of the sphere derived in by Green function method. Note that in this method the factor $`\psi _\mu (iz/a)=e^{i\omega \mu }`$ appears automatically as a result of the point splitting procedure. The evaluation of (9.37) leads to the result $`E=0.092353/2a`$ for the Casimir energy of a spherical conducting shell . This corresponds to the repulsive vacuum force on the sphere. Here the cancellation of interior and exterior divergent terms in the energy density occurs. The discussion on cancellations of divergences between interior and exterior modes see .
## 10 Electromagnetic vacuum in spherical layer between perfectly conducting surfaces
Electromagnetic vev of the EMT in the region between two cocentric perfectly conducting surfaces with radii $`a`$ and $`b`$, $`a<b`$, may be obtained from the results of previous section. The corresponding nonrenormalized components are given by (9.14). Using this formula they can be presented in the form
$$q(a,b,r)=q(a,r)+q^{(ab)}(r),a<r<b,q=\epsilon ,p_{},p,p=\epsilon 2p_{},$$
(10.1)
where $`q(a,r)`$ is given by (9.20), and
$$q^{(ab)}(r)=\frac{1}{8\pi ^2a^2r^2}\underset{l=1}{\overset{\mathrm{}}{}}(2l+1)\underset{\lambda =0,1}{}_0^{\mathrm{}}z\psi _\mu (z/a)\mathrm{\Omega }_{\lambda l}(z,\eta )F_{\lambda l}^{(q)}(z,zx)𝑑z,x=r/a.$$
(10.2)
Here the functions $`F_{\lambda l}^{(q)}`$ are defined by relations (9.18), (9.19), $`\eta =b/a`$, and
$$\mathrm{\Omega }_{\lambda l}(z,\eta )=\frac{e_l^{(\lambda )}(z\eta )/e_l^{(\lambda )}(z)}{e_l^{(\lambda )}(z)s_l^{(\lambda )}(z\eta )e_l^{(\lambda )}(z\eta )s_l^{(\lambda )}(z)}$$
(10.3)
(see notation (9.15)). In (10.1) the dependence on $`b`$ is contained in the summand $`q^{(ab)}`$ only. This quantity is finite for $`ar<b`$ and the regularization of $`q(a,b,r)`$ is equivalent to the renormalization of the first summand. This procedure have been done in previous section, where we have seen that $`q(a,r)`$ (see expressions (9.24) and (9.25)) coincides with the corresponding quantity for the exterior region of a single shell with radius $`a`$. The expressions (10.2) for $`ar<b`$ are finite when $`\mu 0`$ and hence for these values the cuttoff function may be removed putting $`\chi _\mu =1`$.
It can be seen that the quantities (10.1) may be written also in the form
$$q(a,b,r)=q(b,r)+\stackrel{~}{q}^{(ab)}(r),q=\epsilon ,p_{},p,$$
(10.4)
where
$$\stackrel{~}{q}^{(ab)}(r)=\frac{1}{8\pi ^2b^2r^2}\underset{l=1}{\overset{\mathrm{}}{}}(2l+1)\underset{\lambda =0,1}{}_0^{\mathrm{}}z\stackrel{~}{\mathrm{\Omega }}_{\lambda l}(z,\sigma )F_{\lambda l}^{(q)}(z,zy)𝑑z,$$
(10.5)
with $`y=r/b`$, $`\sigma =a/b`$, and
$$\stackrel{~}{\mathrm{\Omega }}_{\lambda l}(z,\sigma )\frac{s_l^{(\lambda )}(z\sigma )/s_l^{(\lambda )}(z)}{e_l^{(\lambda )}(z\sigma )s_l^{(\lambda )}(z)e_l^{(\lambda )}(z)s_l^{(\lambda )}(z\sigma )}.$$
(10.6)
In (10.4) $`\stackrel{~}{q}^{(ab)}(r)0`$ when $`a0`$ and $`q(b,r)`$ coincides with the corresponding quantities inside a single conducting shell with radius $`b`$ (the latter can be seen also by direct evaluation of $`q(b,r)`$). Note that in (10.5) the sum and integral are convergent for $`a<rb`$.
As we said above from the expressions $`q(a,b,r)`$ in limiting cases $`a0`$ or $`b\mathrm{}`$ may be obtained the vacuum stress inside and outside of a single shell. Consider now the another limiting case: $`h=ba=const`$, $`b\mathrm{}`$. For $`a/b1`$ the main contribution in (10.2) is due to large $`l`$. This allows us to use asymptotic formulae for Bessel functions. For instance, in the case of the energy density one has
$$\epsilon \epsilon ^{(ab)}\frac{1}{\pi ^2b^4}_0^{\mathrm{}}z^2\frac{\mathrm{\Lambda }(z,a/b)}{\sqrt{1+z^2}}𝑑z,$$
(10.7)
where
$$\mathrm{\Lambda }=\nu ^3\left\{e^{2\nu [\eta (z)\eta (za/b)]}1\right\}^1\frac{b^4}{16h^4(1+z^2)^2}_0^{\mathrm{}}\frac{s^3ds}{e^s1}=\frac{\pi ^4b^4}{240h^4(1+z^2)^2}.$$
(10.8)
By substituting this into (10.7) we receive the standard result for the Casimir parallel plate configuration: $`\epsilon =\pi ^2/720h^4`$.
Let us present the quantites $`q=\epsilon ,p,p_{}`$ in the form
$$q=q(a,r)+q(b,r)+\mathrm{\Delta }q(a,b,r),a<r<b,$$
(10.9)
where ”interference” term may be written in two ways
$`\mathrm{\Delta }q(a,b,r)`$ $`=`$ $`q^{(ab)}(a,b,r)q(b,r)`$ (10.10)
$`\mathrm{\Delta }q(a,b,r)`$ $`=`$ $`\stackrel{~}{q}^{(ab)}(a,b,r)q(a,r).`$ (10.11)
Here $`q^{(ab)}`$ and $`\stackrel{~}{q}^{(ab)}`$ are defined by relations (10.2) and (10.5). It can be seen that $`\mathrm{\Delta }q(a,b,r)`$ is finite for all $`arb`$, $`a<b`$. Near the surface $`r=a`$ it is convenient to use (10.10), as for $`ra`$ both summands in this formula are finite. For the same reason the formula (10.11) is convenient for calculations near the surface $`r=b`$.
So far we have considered the electromagnetic vacuum in the region between two perfectly conducting spherical surfaces. Consider now a system consisting two cocentric thin spherical shells with radii $`a`$ and $`b`$, $`a<b`$. In this case the vev for the EMT components may be written in the form
$$q(a,b,r)=q(a,r)\theta (ar)+q(b,r)\theta (rb)+\left[q(a,r)+q^{(ab)}(r)\right]\theta (ra)\theta (br),$$
(10.12)
where $`\theta (x)`$ is the unit step function. By using the continuity equation (8.33) it is easy to see that the total Casimir energy for the system under consideration can be presentaed in the form
$$E^{(ab)}=E(a)+E(b)+4\pi \left[b^3\stackrel{~}{p}^{(ab)}(b)a^3p^{(ab)}(a)\right],$$
(10.13)
where $`E(i)`$ is the Casimir energy for a single sphere with radius $`i,i=a,b`$. As it follows from (10.2) and (10.5) the additional vacuum pressures on the spheres are equal to
$`p^{(ab)}(a)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2a^4}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}(2l+1){\displaystyle _0^{\mathrm{}}}𝑑zz\left\{\left[{\displaystyle \frac{l(l+1)}{z^2}}+1\right]\mathrm{\Omega }_{1l}(z,\eta )\mathrm{\Omega }_{0l}(z,\eta )\right\}`$ (10.14)
$`\stackrel{~}{p}^{(ab)}(b)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2b^4}}{\displaystyle \underset{l=1}{\overset{\mathrm{}}{}}}(2l+1){\displaystyle _0^{\mathrm{}}}𝑑zz\left\{\left[{\displaystyle \frac{l(l+1)}{z^2}}+1\right]\stackrel{~}{\mathrm{\Omega }}_{1l}(z,\sigma )\stackrel{~}{\mathrm{\Omega }}_{0l}(z,\sigma )\right\},`$ (10.15)
where $`\mathrm{\Omega }_{\lambda l}`$ and $`\stackrel{~}{\mathrm{\Omega }}_{\lambda l}`$ are defined by relations (10.3) and (10.6).
The vacuum force per unit area of the inner sphere is equal to
$$F^{(a)}=F_1^{(a)}+\mathrm{\Delta }F^{(a)},\mathrm{\Delta }F^{(a)}=p^{(ab)}(a)$$
(10.16)
where $`F_1^{(a)}`$ is the force per unit area of a single sphere with radius $`a`$, and $`\mathrm{\Delta }F^{(a)}`$ is due to the existence of the second sphere (”interaction” force). By similar way vacuum force acting on per unit area of outer sphere is
$$F^{(b)}=F_1^{(b)}+\mathrm{\Delta }F^{(b)},\mathrm{\Delta }F^{(b)}=\stackrel{~}{p}^{(ab)}(b).$$
(10.17)
The results of numerical calculations of quantities $`\mathrm{\Delta }q(a,b,r)`$, $`q=\epsilon ,p,p_{}`$, as well as those for $`\mathrm{\Delta }F^{(a,b)}`$ are presented in . Note that as it follows from the results of these calculations the quantities (10.14) and (10.15) are always negative, and therefore the interaction forces between two spheres are always attractive (as in the parallel plate configuration). The total Casimir energy is positive for small values of $`a/b`$ and is negative for values close to 1. At $`a/b0.7`$ this energy is zero.
## 11 EMT vev inside a perfectly conducting cylindrical shell
In this and next sections we will consider the case of perfectly conducting cylindrically symmetric boundaries. The Casimir effect for a perfectly conducting cylindrical shell was considered in (see also ) and for a dielectric cylinder in by using the Green function formalism. Recently the problem is reconsidered in using the mode summation technique and in , within the framework of the zeta-function regularization scheme. In these papers global quantities, such as the total enehgy and stress on a shell, are investigated. Local characteristics of the electromagnetic vacuum are considered in for the interior and exterior regions of a conducting cylindrical shell, and in for two coaxial shells. In this papers the mode summation method is used combined with generalized Abel-Plana formula. Our consideration below is based on these works (see also ). The vev of the EMT for electromagnetic field inside a perfectly conducting cylindrical surface with radius $`a`$ can be found by the way similar to the spherical case. As an eigenfunctions we use the vector potentials corresponding to the cylindrical waves of magnetic ($`\lambda =0`$) and electric ($`\lambda =1`$) type:
$$𝐀_\alpha =\beta _{\lambda m}\{\begin{array}{cc}𝐞_3\times _t\left\{J_m(\gamma r)\mathrm{exp}\left[i(m\phi +kz\omega t)\right]\right\},\hfill & \lambda =0\hfill \\ (1/i\omega )\left[𝐞_3+(ik/\gamma ^2)_t\right]J_m(\gamma r)\mathrm{exp}\left[i(m\phi +kz\omega t)\right],\hfill & \lambda =1\hfill \end{array},$$
(11.1)
where the cylindrical coordinates $`(r,\phi ,z)`$ are used with unit vectors $`𝐞_i`$, $`\gamma ^2=\omega ^2k^2`$, $`m`$ is an integer, $`_t`$ is the transverse to the $`z`$ axis part of the nabla operator. From the standard boundary conditions we obtain the following equations for the possible values of the quantum number $`\gamma `$:
$`J_m^{}(\gamma a)`$ $`=`$ $`0,\lambda =0,`$
$`J_m(\gamma a)`$ $`=`$ $`0,\lambda =1.`$ (11.2)
The constants $`\beta _{\lambda m}`$ are determined from the normalization condition and are equal to
$$\beta _{\lambda m}^2=\frac{\gamma ^2}{\pi \omega a^2}\left[J_m^{}_{}{}^{}2(\gamma a)+(1m^2/\gamma ^2a^2)J_m^2(\gamma a)\right]^1=\frac{\gamma ^3}{\pi \omega a}T_m(\gamma a),$$
(11.3)
where $`T_m(z)`$ is defined by (3.12). As independent quantum numbers we will choose the set $`\alpha =(mk\gamma \lambda )`$. In this case $`\omega ^2=\gamma ^2+k^2`$ and $`\gamma `$ takes discrete values being solutions to (11.2). By using the standard formula (8.1) with (11.1) one obtains
$$0|T_k^i|0=\mathrm{diag}(\epsilon ,p_1,p_2,p_3),$$
(11.4)
where the energy density $`\epsilon `$, the pressure $`p_i`$ in direction $`𝐞_i`$ can be presented in the form (below the index $`c`$ will specify quantities for the cylindrical geometry)
$$q_c(a,r)=\frac{1}{8\pi }\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k\underset{\lambda ,\gamma }{}\beta _{\lambda m}^2f_m^{(q)}(\gamma r),q_c=\epsilon ,p_1,p_2$$
(11.5)
with
$`f_m^{(\epsilon )}(y)`$ $`=`$ $`\left({\displaystyle \frac{2k^2}{\gamma ^2}}+1\right)\left[J_m^{}_{}{}^{}2(y)+{\displaystyle \frac{m^2}{y^2}}J_m^2(y)\right]+J_m^2(y)`$ (11.6)
$`f_m^{(p_i)}(y)`$ $`=`$ $`(1)^i\left[J_m^{}_{}{}^{}2(y)\left({\displaystyle \frac{m^2}{y^2}}+(1)^i\right)J_m^2(y)\right],i=1,2,`$ (11.7)
and $`p_3=\epsilon p_1p_2`$. The latter corresponds to the zero trace of the vacuum EMT. The quantities (11.5) are divergent. To make them finite we introduce the cutoff function $`\psi _\mu (\gamma )`$ and consider the finite quantities
$$q_c=\frac{1}{8\pi ^2a^4}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k\underset{\lambda =0}{\overset{1}{}}\underset{n=1}{\overset{\mathrm{}}{}}\frac{j_{m,n}^{(\lambda )3}\psi _\mu (j_{m,n}^{(\lambda )}/a)}{\sqrt{k^2a^2+j_{m,n}^{(\lambda )2}}}T_m(j_{m,n}^{(\lambda )})f_m^{(q)}(j_{m,n}^{(\lambda )}x),q=\epsilon ,p_1,p_2$$
(11.8)
with $`x=r/a`$, $`\gamma a=j_{m,n}^{(\lambda )}`$ are the roots of the equations (11.2) for $`\lambda =0`$ and $`\lambda =1`$, correspondingly. To calculate the sums over zeros of the functions (11.2) here we use the summation formula obtained in section 2, namely the formula (3.25). Let us choose as a function $`f(z)`$ in GAPF
$$f(z)=\frac{z^3}{\sqrt{z^2+k^2a^2}}\psi _\mu (z/a)f_m^{(q)}(zx).$$
(11.9)
This function has branch point on the imaginary axis and we have to use the version (3.25) with lower sign. Here we will assume a class of cutoff functions for which (11.9) satisfies conditions (3.4) and (3.15) uniformly with respect to $`\mu `$. By using the asymptotic formulae for Bessel functions these conditions can be easily translated in terms of $`\psi _\mu `$. We choose $`A=0,B=1`$ in the case $`\lambda =0`$ and $`A=1,B=0`$ in the case $`\lambda =1`$ (see (3.1)), and $`\nu =m`$. Using the relation
$$f_m^{(q)}(ye^{\pi i/2})=e^{2m\pi i}f_m^{(q)}(ye^{\pi i/2})$$
(11.10)
we see that the subintegrand of the first integral on rhs of the formula (3.25) is proportianal to $`\psi _\mu (iz/a)\psi _\mu (iz/a)`$. Consequently after removing the cutoff ($`\psi _\mu 1`$) the contribution of the first integral will be zero. For this reason we shall write only the second integral on the right of (3.25). For the simplicity we will assume also that the cutoff function has no poles in the right-half plane. In this case the residue terms are zero. It can be seen that the residue term on the right vanishes as well. Hence by applying GAPF to the sums over zeros of Bessel functions in (11.8) and omitting the term which will vanish after the cutoff removing we obtain
$`q_c`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^+\mathrm{}}dk\{{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{z^3\psi _\mu (z)}{\sqrt{z^2+k^2}}}f_m^{(q)}(zr)dz+`$ (11.11)
$`+`$ $`{\displaystyle \frac{e^{m\pi i}}{\pi a^3}}{\displaystyle _{|ak|}^{\mathrm{}}}[{\displaystyle \frac{K_m(z)}{I_m(z)}}+{\displaystyle \frac{K_m^{}(z)}{I_m^{}(z)}}]f_m^{(q)}(zxe^{\pi i/2}){\displaystyle \frac{z^3\chi _\mu (z/a)dz}{\sqrt{z^2a^2k^2}}}\},`$
where the function $`\chi _\mu (y)`$ is defined in (8.18). The second integral on the right of this formula vanishes in the limit $`a\mathrm{}`$, whereas the first one does not depend on $`a`$. It follows from here that the latter corresponds to the Minkowskian part without boundaries. This can be seen also directly by explicit summation over $`m`$ using the formula $`_{m=\mathrm{}}^+\mathrm{}J_{n\pm m}^2(z)=1`$. For instance, in the case of the energy density one has
$`\epsilon ^{(0)}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}{\displaystyle \underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑k{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑z{\displaystyle \frac{z^3\psi _\mu (z)}{\sqrt{z^2+k^2}}}f_m^{(\epsilon )}(zr)=`$ (11.12)
$`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle _{\mathrm{}}^+\mathrm{}}𝑑k{\displaystyle _0^+\mathrm{}}𝑑zz\sqrt{z^2+k^2}\psi _\mu (z)={\displaystyle \frac{1}{2\pi }}{\displaystyle _0^+\mathrm{}}\omega ^3\stackrel{~}{\psi }_\mu (\omega )𝑑\omega ,`$
with $`\omega ^2=z^2+k^2`$. Hence GAPF allows us to extract the contribution of unbounded space without specifying the cutoff function. The remained part is finite for $`x<1`$ and $`\mu 0`$, and can be written in the form
$$q_c=\frac{1}{2\pi ^3a^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}e_{}^{m\pi i}_0^{\mathrm{}}𝑑t_0^{\mathrm{}}𝑑yz^2\left[\frac{K_m(z)}{I_m(z)}+\frac{K_m^{}(z)}{I_m^{}(z)}\right]\chi _\mu (z/a)f_m^{(q)}(zxe^{\pi i/2}),$$
(11.13)
with $`z^2=t^2+y^2`$ and $`t=ka`$. Here we have introduced a new integration variable $`y`$ and the prime on the summation sign indicates that the $`m=0`$ term is to be halved. For $`q=\epsilon `$ from (11.6) one has
$$e^{m\pi i}z^2f_m^{(q)}(zxe^{\pi i/2})=z^2I_m^2(zx)+(t^2y^2)\left[I_m^{}_{}{}^{}2(zx)+\frac{m^2}{z^2x^2}I_m^2(zx)\right]$$
(11.14)
and it can be easily seen that the contribution of the summand containing $`t^2y^2`$ in (11.13) is zero. Introducing the polar coordinates $`(z,\theta )`$ on the plane $`(t,y)`$ for the EMT components inside a perfectly conducting cylindrical surface from (11.13) one finds
$$q_c(a,r)=\frac{1}{4\pi ^2a^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}𝑑z\chi _\mu (z/a)F_m^{(q)}(z,x),r<a,q=\epsilon ,p_i,$$
(11.15)
where the following notations are introduced
$`F_{cm}^{(q)}(z,x)`$ $`=`$ $`z^3\left[{\displaystyle \frac{K_m(z)}{I_m(z)}}+{\displaystyle \frac{K_m^{}(z)}{I_m^{}(z)}}\right]\{\begin{array}{cc}I_m^2(zx),\hfill & q=\epsilon \hfill \\ \left(1+m^2/z^2x^2\right)I_m^2(zx)I_m^{}_{}{}^{}2(zx),\hfill & q=p_1\hfill \end{array}`$ (11.18)
$`F_{cm}^{(p_3)}(z,x)`$ $`=`$ $`F_{cm}^{(\epsilon )},F_{cm}^{(p_2)}=2F_{cm}^{(\epsilon )}F_{cm}^{(p_1)}.`$ (11.19)
In particular we see that inside the cylinder $`\epsilon =p_3`$. This relation is the same as in the case of the Minkowski vacuum. This is natural, as we have no constraint on $`z`$ direction. On cylinder axis ($`x=0`$) the $`m=0`$ term contributes only and we have
$$\epsilon (0)=p_1(0)=p_2(0)=\frac{1}{8\pi ^2a^4}_0^{\mathrm{}}𝑑zz^3\left[\frac{K_0(z)}{I_0(z)}+\frac{K_0^{}(z)}{I_0^{}(z)}\right]=0.0168a^4$$
(11.20)
with $`q^{}(0)=0`$. The vacuum EMT satisfy continuity equation which can be written now as
$$\frac{dp_1}{dr}+\frac{2}{r}(p_1\epsilon )=0,$$
(11.21)
or in the integral form
$$E_c(r)=2\pi _0^r\epsilon (t)t𝑑t=\pi r^2p_1(r),r<a.$$
(11.22)
To determine the integration constant here we have used the relations (11.20) between the EMT components on the cylinder axis. As we see the total energy per unit length inside the cylinder with radius $`r`$ is equal to the radial pressure on the surface of this cylinder multiplied by the corresponding volume.
Let us consider the behavior of the functions (11.19) in two limiting cases:
1) For fixed $`m`$ and large $`zx`$ from the asymptotic expansions of Bessel functions we find
$$F_{cm}^{(\epsilon )}\frac{1}{2}F_{cm}^{(p_2)}\frac{z}{2x}e^{2z(1x)},F_{cm}^{(p_1)}\frac{1}{2x^2}e^{2z(1x)}.$$
(11.23)
2) For large $`m`$ by using the uniform asymptotic expansions of Bessel functions one obtains
$$F_{cm}^{(q)}(mz,x)\mathrm{\Phi }_{cm}^{(q)}(z,x)\mathrm{exp}\left\{2m[\eta (z)\eta (zx)]\right\},$$
(11.24)
with
$`\mathrm{\Phi }_{cm}^{(\epsilon )}(z,x)`$ $`=`$ $`{\displaystyle \frac{m}{2}}z^5t(zx)t^3(z)\left\{1{\displaystyle \frac{1}{12m}}\left[t(z)(t^2(z)3)+t(zx)(5t^2(zx)3)\right]\right\}`$ (11.25)
$`\mathrm{\Phi }_{cm}^{(p_1)}(z,x)`$ $`=`$ $`{\displaystyle \frac{1}{2}}z^5t^2(zx)t^3(z),`$ (11.26)
where the standard notations are used.
It follows from here that at cylinder surface, $`ra`$, the expressions for the EMT components are divergent and near the surface the corresponding quantities are dominated by large $`m`$. Hence to obtain the asymptotic behaviour we can use the corresponding asymptotic formulae for modified Bessel functions. Then after the elementary summation over $`m`$ we find the following asymptotic behaviour
$$\epsilon \frac{1}{2}p_2\frac{1}{60\pi ^2a(ar)^3},p_1\frac{1}{60\pi ^2a^2(ar)^2}.$$
(11.27)
This formulae are special cases of the general expansions for the EMT near a smooth boundary of arbitrary shape . Note that, as it follows from (11.23) now, unlike the spherical case, in (11.15) with $`\chi _\mu =1`$ the integral over $`z`$ diverges for $`x=1`$.
The results of the numerical calculations for vacuum EMT components (11.15) are presented in . Note that $`\epsilon ,p_i<0`$, $`i=1,2`$ everywhere inside the cylinder. The ratio of the energy density to the azimuthal pressure is a decreasing function on $`r`$ and $`0.5\epsilon /p_21`$.
## 12 Vacuum EMT outside a perfectly conducting cylinder
First we consider the vev of the electromagnetic EMT in the region between two coaxial cylindrical surfaces with radii $`a`$ and $`b`$, $`a<b`$. The corresponding eigenfunctions have the form (11.1) with replacement $`J_m(\gamma r)P_{\lambda m}(\gamma a,\gamma r)`$, where
$$P_{\lambda m}(x,y)=\{\begin{array}{cc}J_m(y)Y_m(x)Y_m(y)J_m(x),\hfill & \lambda =1\hfill \\ J_m(y)Y_m^{}(x)Y_m(y)J_m^{}(x),\hfill & \lambda =0\hfill \end{array}$$
(12.1)
From the boundary conditions on $`r=a,b`$ one obtains that the eigennumbers $`\gamma `$ have to be solutions to the following equations
$`P_{1m}(\gamma a,\gamma r)|_{r=b}`$ $`=`$ $`0,\lambda =1`$ (12.2)
$`\left[{\displaystyle \frac{}{r}}P_{0m}(\gamma a,\gamma r)\right]_{r=b}`$ $`=`$ $`0,\lambda =0`$ (12.3)
These equations have infinite number of simple real solutions. Now the normalization coefficients $`\beta _{\lambda m}`$ are in form
$$\beta _{\lambda m}^2=\frac{\pi z^4}{4a^4\omega }\{\begin{array}{cc}\left[J_m^2(z)/J_m^2(z\eta )1\right]^1,\hfill & \lambda =1\hfill \\ \left[\left(1m^2/z^2\eta ^2\right)J_m^{}_{}{}^{}2(z)/J_m^{}_{}{}^{}2(z\eta )1+m^2/z^2\right]^1,\hfill & \lambda =0\hfill \end{array}$$
(12.4)
where $`z=\gamma a`$, $`\eta =b/a`$.
From Eq.(8.1) it follows that the vacuum EMT has diagonal form (11.4) with components
$$q_c(a,b,r)=\frac{1}{8\pi }\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k\underset{\gamma ,\lambda }{}\beta _{\lambda m}^2f_{\lambda m}^{(q)}(\gamma a,\gamma r),q_c=\epsilon ,p_i,$$
(12.5)
where the expressions for the functions $`f_{\lambda m}^{(q)}(\gamma a,y)`$ are obtained from (11.6) and (11.7) replacing $`J_m(y)P_{\lambda m}(\gamma a,y)`$.
The eigenvalue equations (12.2) and (12.3) can be written in terms of the function (4.1) as
$$C_m^{AB}(\eta ,\gamma b)=0,A=\lambda ,B=1\lambda ,\lambda =0,1$$
(12.6)
(see the notation (3.1)). Note that the normalization coefficients can be expressed in terms of the function (4.6):
$$\beta _{\lambda m}^2=\frac{\pi z^{52\lambda }}{4a^4\omega }T_m^{AB}(\eta ,z).$$
(12.7)
Using these relations and introducing a cutoff function $`\psi _\mu `$ the divergent quantities (12.5) can be written in the form of the following finite integrosums
$$q_c=\frac{1}{32a^3}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k\underset{\lambda =0}{\overset{1}{}}\underset{n=1}{\overset{\mathrm{}}{}}\frac{(\gamma _{m,n}^{(\lambda )})^{52\lambda }\psi _\mu (\gamma _{m,n}^{(\lambda )}/a)}{\sqrt{k^2a^2+\gamma _{m,n}^{(\lambda )2}}}T_m^{AB}(\eta ,\gamma _{m,n}^{(\lambda )})f_{\lambda m}^{(q)}(\gamma _{m,n}^{(\lambda )},\gamma _{m,n}^{(\lambda )}x),$$
(12.8)
where $`q_c=\epsilon ,p_i`$ and $`\gamma a=\gamma _{m,n}^{(\lambda )}`$, $`n=1,2,\mathrm{}`$ are the solutions to the eigenvalue equations (12.2), (12.3) or (12.6). By choosing in the formula (4.13)
$$h(z)=\frac{z^{52\lambda }}{\sqrt{z^2+k^2a^2}}\psi _\mu (z/a)f_{\lambda m}^{(q)}(z,zx).$$
(12.9)
(as noted above this formula is valid in the case when the corresponding function has branch point on the imaginary axis (see also Remark to the Theorem 2)) one obtains
$`{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}h(\gamma _{m,n}^{(\lambda )})T_m^{AB}(\eta ,\gamma _{m,n}^{(\lambda )})`$ $`=`$ $`{\displaystyle \frac{2}{\pi ^2}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{h(x)dx}{\overline{J}_m^2(x)+\overline{Y}_m^2(x)}}`$ (12.10)
$`{\displaystyle \frac{1}{2\pi }}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{\overline{K}_m(\eta x)}{\overline{K}_m(x)}}{\displaystyle \frac{\left[h(xe^{\pi i/2})+h(xe^{\pi i/2})\right]dx}{\overline{K}_m(x)\overline{I}_m(\eta x)\overline{K}_m(\eta x)\overline{I}_m(x)}}.`$
Here in accordance with (3.1) and (12.6)
$`\overline{J}_m(z)=J_m(z),\lambda =1`$ (12.11)
$`\overline{J}_m(z)=zJ_m^{}(z),\lambda =0,`$ (12.12)
and in similar way for other Bessel functions in (12.10). To obtain the EMT components for the outside region of a perfectly conducting cylindrical shell we consider the limit $`b\mathrm{}`$. It can be seen that the second sum on the right of (12.10) is zero in this limit and the first one does not depend on $`b`$. Hence for the outside region of a single cylinder we obtain
$$q_c(a,r)=\frac{1}{16\pi ^2a^4}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k_0^{\mathrm{}}𝑑z\underset{\lambda =0,1}{}\frac{z^3\psi _\mu (z/a)}{\sqrt{k^2+z^2/a^2}}\frac{f_{\lambda m}^{(q)}(z,zx)}{\overline{J}_m^2+\overline{Y}_m^2}.$$
(12.13)
To regularize we subtract from these quantities the contribution of unbounded Minkowski spacetime which can be presented in the form (see (11.12)):
$$q^{(0)}=\frac{1}{8\pi ^2a^4}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k_0^{\mathrm{}}𝑑z\frac{z^3\psi _\mu (z/a)}{\sqrt{k^2+z^2/a^2}}f_m^{(q)}(zx)$$
(12.14)
with the function $`f_m^{(q)}`$ defined as (11.6), (11.7). By using the definitions of $`f_{\lambda m}^{(q)}`$ and $`f_m^{(q)}`$ it is easy to see that
$$\frac{f_{\lambda m}^{(q)}(z,zx)}{\overline{J}_m^2+\overline{Y}_m^2}f_m^{(q)}(zx)=\frac{1}{2}\underset{n=1,2}{}\mathrm{\Omega }_{\lambda m}^{(n)}(z)f_m^{(nq)}(zx),$$
(12.15)
where by definition the expression for $`f_m^{(nq)}(zx)`$ is obtained from that for $`f_m^{(q)}(zx)`$ replacing $`J_m(zx)H_m^{(n)}(zx)`$, $`n=1,2`$, and
$$\mathrm{\Omega }_{\lambda m}^{(n)}(z)=\{\begin{array}{cc}J_m(z)/H_m^{(n)}(z),\hfill & \lambda =1\hfill \\ J_m^{}(z)/H_m^{(n)^{}}(z),\hfill & \lambda =0\hfill \end{array}$$
(12.16)
Hence
$$\mathrm{reg}q_c(a,r)=\frac{1}{16\pi ^2a^4}\underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}_{\mathrm{}}^+\mathrm{}𝑑k_0^{\mathrm{}}𝑑z\underset{\lambda ,n}{}\mathrm{\Omega }_{\lambda m}^{(n)}(z)f_m^{(nq)}(zx).$$
(12.17)
By rotating the integration contour for $`z`$ by angle $`\pi /2`$ for $`n=1`$ and by angle $`\pi /2`$ for $`n=2`$ (note that the function $`H_m^{(1)}(z)`$ ($`H_m^{(2)}(z)`$) has no zeros for $`0\mathrm{arg}z\pi /2`$ ($`\pi /2\mathrm{arg}z0`$)) and introducing Bessel modified functions for the regularized components we obtain (the $`\mathrm{reg}`$ sign is suppressed)
$`q_c`$ $`=`$ $`{\displaystyle \frac{1}{16\pi ^3a^4}}{\displaystyle \underset{m=\mathrm{}}{\overset{+\mathrm{}}{}}}{\displaystyle _{\mathrm{}}^+\mathrm{}}dk\{i{\displaystyle _0^{|ak|}}dz[\psi _\mu \left({\displaystyle \frac{iz}{a}}\right)\psi _\mu ({\displaystyle \frac{iz}{a}})]{\displaystyle \frac{F_{cm}^{(q)}(z,x)}{\sqrt{k^2z^2/a^2}}}+`$ (12.18)
$`+2{\displaystyle _{|ak|}^{\mathrm{}}}dz\chi _\mu \left({\displaystyle \frac{z}{a}}\right){\displaystyle \frac{F_{cm}^{(q)}(z,x)}{\sqrt{z^2/a^2k^2}}}\}`$
(the definition of the functions $`F_{cm}^{(q)}`$ see below). In (12.18) the integrals are convergent for $`x>1`$ and $`\mu =0`$ and hence the cutoff can be removed. In this limit the first integral is zero and for regularized components of EMT after transformations, similar to the interior case, one obtains
$$q_c(a,r)=\frac{1}{4\pi ^2a^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}𝑑z\chi _\mu (z/a)F_{cm}^{(q)}(z,x),r>a,q=\epsilon ,p_i,$$
(12.19)
where for $`x>1`$ the functions $`F_{cm}^{(q)}(z,x)`$ are defined as
$`F_{cm}^{(q)}(z,x)`$ $`=`$ $`z^3\left[{\displaystyle \frac{I_m(z)}{K_m(z)}}+{\displaystyle \frac{I_m^{}(z)}{K_m^{}(z)}}\right]\{\begin{array}{cc}K_m^2(zx),\hfill & q=\epsilon \hfill \\ \left(1+m^2/z^2x^2\right)K_m^2(zx)K_m^{}_{}{}^{}2(zx),\hfill & q=p_1\hfill \end{array}`$ (12.22)
$`F_{cm}^{(p_3)}(z,x)`$ $`=`$ $`F_{cm}^{(\epsilon )},F_{cm}^{(p_2)}=2F_{cm}^{(\epsilon )}F_{cm}^{(p_1)}.`$ (12.23)
It follows from here that $`\epsilon =p_3`$.
The asymptotic expressions for the functions (12.23) are as follows:
1) For fixed $`m`$ and large $`z`$:
$$F_{cm}^{(\epsilon )}\frac{1}{2}F_{cm}^{(p_2)}\frac{z}{2x}e^{2z(1x)},F_{cm}^{(p_1)}\frac{1}{2x^2}e^{2z(1x)}.$$
(12.24)
2) For large $`m`$ from the uniform asymptotic expansions of Bessel functions one obtains
$`F_{cm}^{(\epsilon )}(mz,x)`$ $``$ $`{\displaystyle \frac{m}{2}}z^5t(zx)t^3(z)\{1+{\displaystyle \frac{1}{12m}}[t(z)(t^2(z)3)+t(zx)(5t^2(zx)3)]\}\times `$ (12.25)
$`\times \mathrm{exp}\left\{2m[\eta (z)\eta (zx)]\right\},`$
$`F_{cm}^{(p_1)}(mz,x)`$ $``$ $`{\displaystyle \frac{1}{2}}z^5t^2(zx)t^3(z)\mathrm{exp}\left\{2m[\eta (z)\eta (zx)]\right\},`$ (12.26)
As in the case of the interior components of the vacuum EMT is divergent when $`ra`$ with asymptotic behaviour
$$\epsilon \frac{1}{2}p_2\frac{1}{60\pi ^2a(ra)^3},p_1\frac{1}{60\pi ^2a^2(ra)^2}$$
(12.27)
Comparing with (11.27) we see that in calculating the total energy for the infinitely thin cylindrical shell the leading divergences cancel.
The asymptotic expressions for the vev at large distances from the cylinder axis, $`ra`$, can be found from (12.19) introducing new integration variable $`y=zx`$ and expanding the integrands over $`1/x`$. In this limit the main contribution comes from the lowest order mode with $`m=0`$ and one obtaines
$$q_c(a,r)\frac{c^{(q)}}{8\pi ^2r^4\mathrm{ln}(r/a)},c^{(\epsilon )}=c^{(p_1)}=\frac{1}{3},c^{(p_2)}=1,ra$$
(12.28)
Here compared to the spherical case the corresponding quantities tend to zero more slowly.
From the continuity equation for the vacuum EMT one has the following integral relation
$$p_1(r)=\frac{2}{r^2}_{\mathrm{}}^r\epsilon (t)t𝑑t=\frac{E_c^{out}(r)}{\pi r^2},$$
(12.29)
where $`E_c^{out}(r)`$ is the total energy (per unit length) outside cylinder with radius $`r`$. Combining this relation with (11.22) for the total vacuum energy of the cylindrical shell per unit length we obtain
$$E_c=E_c^{in}(a)+E_c^{out}(a)=\pi a^2\left[p_1(a)p_1(a+)\right].$$
(12.30)
By taking into account the corresponding expressions for the radial pressure this yields
$`E_c`$ $`=`$ $`{\displaystyle \frac{1}{4\pi a^2}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle {}_{}{}^{}_{0}^{\mathrm{}}}dz\chi _\mu (z/a)\left(\mathrm{ln}[I_m(z)K_m(z)]\right){}_{}{}^{}[z^2+(z^2+m^2){\displaystyle \frac{I_m(z)K_m(z)}{I_m^{}(z)K_m^{}(z)}}]=`$ (12.31)
$`=`$ $`{\displaystyle \frac{1}{4\pi a^2}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle {}_{}{}^{}_{0}^{\mathrm{}}}𝑑z\chi _\mu (z/a)z^2{\displaystyle \frac{d}{dz}}\mathrm{ln}\left[1z^2\left(I_m(z)K_m(z)\right)^2\right].`$
In the last expression integrating by part and omitting the boundary term we obtain the Casimir energy in the form used in numerical calculations. The corresponding results are presented in . Note that in the evaluation of the Casimir energy for a perfeclty conducting cylindrical shell by Green function method to perform the complex frequency rotation procedure an additional cutoff function have to be introduced (see ). This is related to the abovmentioned divergency of the integrals over $`z`$ for $`x=1`$.
The results of the numerical evaluations for the energy density and pressures distributions (formula (12.19)) are presented in . The energy density and azimuthal pressure in the exterior region are always positive, and radial pressure is negative. The ratio of the energy density to the azimuthal pressure is decreasing function on $`r`$, and $`1/3\epsilon /p_20.5`$. Note that this ratio is continous function for all $`r`$ and monotonically decreases from 1 at the cylinder axis to 1/3 at infinity.
## 13 Vacuum EMT between two coaxial cylindrical shells
By using the results from previous section the vev of the electromagnetic EMT in the region between two coaxial conducting cylindrical surfaces may be presented in the form (11.4) with components
$$q_c(a,b,r)=q_c(a,r)+q_c^{(ab)}(r),a<r<b,q_c=\epsilon ,p_i,$$
(13.1)
where $`q_c(a,r)`$ is given by (12.13), and
$$q_c^{(ab)}(r)=\frac{1}{16\pi a^3}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}𝑑k\underset{\lambda =0}{\overset{1}{}}_0^{\mathrm{}}\frac{\overline{K}_m(z\eta )}{\overline{K}_m(z)}\frac{\left[h(ze^{\pi i/2})+h(ze^{\pi i/2})\right]dz}{\overline{K}_m(z)\overline{I}_m(z\eta )\overline{K}_m(z\eta )\overline{I}_m(z)}.$$
(13.2)
Here the function $`h(z)`$ is defined according to (12.9). As we have shown the first summand on the right of (13.1) presents the corresponding quantity for the vacuum outside a single perfectly conducting cylindrical shell with radius $`a`$. As we shall see later $`q^{(ab)}(r)`$ is finite for $`ar<b`$ at $`\mu =0`$, and hence regularization is necessary for $`q_c(a,r)`$ only. This have been done in previous section. We have shown that result does not depend on specific form of the cutoff function and can be presented in the form (12.19) and (12.23).
In (13.2) the integral over $`z`$ can be presented as a sum of two integrals along segments $`(0,|ak|)`$ and $`(|ak|,\mathrm{})`$. By using the relation (3.24) and the explicit form of $`h(z)`$ it is easy to see that the first integral will contain the cutoff function in the form $`\psi _\mu (iz/a)\psi _\mu (iz/a)`$ and hence vanishes after the cutoff removing. For this reason below we will consider the second integral only. After the transformations similar to those we used to obtain (11.15), the quantities $`q^{(ab)}`$ can be written in the form
$$q_c^{(ab)}(r)=\frac{1}{4\pi ^2a^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}𝑑z\underset{\lambda =0}{\overset{1}{}}z^3\mathrm{\Omega }_{\lambda m}^c(\eta ,z)F_{\lambda m}^{(q)}(z,zx),x=r/a,$$
(13.3)
where
$`\mathrm{\Omega }_{1m}^c(\eta ,z)`$ $`=`$ $`{\displaystyle \frac{K_m(z\eta )/K_m(z)}{K_m(z)I_m(z\eta )K_m(z\eta )I_m(z)}},`$ (13.4)
$`\mathrm{\Omega }_{0m}^c(\eta ,z)`$ $`=`$ $`{\displaystyle \frac{K_m^{}(z\eta )/K_m^{}(z)}{K_m^{}(z)I_m^{}(z\eta )K_m^{}(z\eta )I_m^{}(z)}},`$ (13.5)
and
$`F_{c\lambda m}^{(\epsilon )}(z,y)`$ $`=`$ $`Q_{\lambda m}^2(z,y),F_{c\lambda m}^{(p_3)}=F_{c\lambda m}^{(\epsilon )}F_{c\lambda m}^{(p_1)}F_{c\lambda m}^{(p_2)}`$ (13.6)
$`F_{c\lambda m}^{(p_i)}(z,y)`$ $`=`$ $`\left(1(1)^i{\displaystyle \frac{m^2}{y^2}}\right)Q_{\lambda m}^2(z,y)+(1)^i\left[{\displaystyle \frac{}{y}}Q_{\lambda m}(z,y)\right]^2,i=1,2`$
Here we have introduced the notation
$`Q_{1m}(z,y)`$ $`=`$ $`K_m(z)I_m(y)I_m(z)K_m(y)`$ (13.7)
$`Q_{0m}(z,y)`$ $`=`$ $`K_m^{}(z)I_m(y)I_m^{}(z)K_m(y).`$
The quantities (13.1) with (12.19) and (13.3) present the regularized vev of the EMT components in the region between two coaxial conducting cylindrical surfaces. Let us consider the limiting cases of the term (13.3). First let $`a/r,a/b1`$. After replacing $`zz\eta `$ and expanding the subintegrand over $`a/r`$ and $`a/b`$ it can be seen that
$$q_c^{(ab)}(r)q_c(b,r),a/r,a/b1,r<b,$$
(13.8)
where $`q_c(b,r)`$ are the components for the vacuum EMT inside a single cylindric shell with radius $`b`$ (see (11.15), (11.18)).
When $`ab`$ the sum over $`m`$ in (13.3) diverges. Consequently for $`bab`$ the main contribution to $`q_c^{(ab)}`$ is due to large $`m`$. By using the uniform asymptotic expansions for Bessel functions in this limit one obtaines
$$\epsilon \frac{1}{2\pi ^2a^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}\frac{m^3z^3dz}{e^{2m[\eta (zb/a)\eta (z)]}1}\frac{\pi ^2}{720(ba)^4},$$
(13.9)
which coincides with the corresponding quantity for the Casimir parallel plate configuration.
From (13.1) and (13.3) it can be seen that the vev of EMT components can be written also in the form
$$q_c(a,b,r)=q_c(b,r)+\stackrel{~}{q}_c^{(ab)}(r),a<r<b,q_c=\epsilon ,p_i.$$
(13.10)
Here $`q_c(b,r)`$ are vev inside a single cylindrical surface with radius $`b`$ (see (11.15), (11.18) with replacement $`ab`$), and
$$\stackrel{~}{q}_c^{(ab)}(r)=\frac{1}{4\pi ^2b^4}\underset{m=0}{\overset{\mathrm{}}{}}{}_{}{}^{}_{0}^{\mathrm{}}𝑑z\underset{\lambda =0}{\overset{1}{}}z^3\stackrel{~}{\mathrm{\Omega }}_{\lambda m}^c(\sigma ,z)F_{\lambda m}^{(q)}(z,zy)$$
(13.11)
where $`y=r/b`$, $`\sigma =a/b`$, and
$`\stackrel{~}{\mathrm{\Omega }}_{1m}^c(\sigma ,z)`$ $`=`$ $`{\displaystyle \frac{I_m(z\sigma )/I_m(z)}{I_m(z)K_m(z\sigma )I_m(z\sigma )K_m(z)}},`$ (13.12)
$`\stackrel{~}{\mathrm{\Omega }}_{0m}^c(\sigma ,z)`$ $`=`$ $`{\displaystyle \frac{I_m^{}(z\sigma )/I_m^{}(z)}{I_m^{}(z)K_m^{}(z\sigma )I_m^{}(z\sigma )K_m^{}(z)}}.`$ (13.13)
The quantites (13.11)are finite for all $`a<rb`$ and diverge on surface $`r=a`$.
From the above it follows that if we present the vacuum EMT components between cylindrical surfaces in the form
$$q_c=q_c(a,r)+q_c(b,r)+\mathrm{\Delta }q_c(a,b,r)$$
(13.14)
then the quantities
$$\mathrm{\Delta }q_c(a,b,r)=q_c^{(ab)}(r)q_c(b,r)=\stackrel{~}{q}_c^{(ab)}(r)q_c(a,r)$$
(13.15)
are finite for all $`r`$ from $`arb`$. In (13.15) the first presentation is convenient near the surface $`r=a`$, as in this case both summands are finite. Similarly the second presentation is convenient near $`r=b`$.
Let us consider a system of two coaxial thin cylindrical shells with radii $`a`$ and $`b`$, $`a<b`$. The vacuum EMT components may be written in the form
$$q_c=q_c(a,r)\theta (ar)+q_c(b,r)\theta (rb)+\left[q_c(a,r)+q_c^{ab}(r)\right]\theta (ra)\theta (br).$$
(13.16)
Similar to the spherical case using the continuity equation (11.21) the total Casimir energy for this system may be written as
$$E_c^{(ab)}=E_c(a)+E_c(b)+\pi b^2\stackrel{~}{p}_{c1}^{(ab)}(b)\pi a^2p_{c1}^{(ab)}(a),$$
(13.17)
where $`E_c(i)`$ is the Casimir energy for a single cylindrical shell with radius $`i,i=a,b`$. For the additional vacuum pressures on the cylindrical surfaces from (13.3) and (13.11) one has:
$`p_{c1}^{(ab)}(a)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2a^4}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle {}_{}{}^{}_{0}^{\mathrm{}}}z𝑑z\left[\left({\displaystyle \frac{m^2}{z^2}}+1\right)\mathrm{\Omega }_{0m}^c(\eta ,z)\mathrm{\Omega }_{1m}^c(\eta ,z)\right],`$ (13.18)
$`\stackrel{~}{p}_{c1}^{(ab)}(b)`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2b^4}}{\displaystyle \underset{m=0}{\overset{\mathrm{}}{}}}{\displaystyle {}_{}{}^{}_{0}^{\mathrm{}}}z𝑑z\left[\left({\displaystyle \frac{m^2}{z^2}}+1\right)\stackrel{~}{\mathrm{\Omega }}_{0m}^c(\sigma ,z)\stackrel{~}{\mathrm{\Omega }}_{1m}^c(\sigma ,z)\right],`$ (13.19)
where $`\mathrm{\Omega }_{\lambda m}^c(\eta ,z)`$ and $`\stackrel{~}{\mathrm{\Omega }}_{1m}^c(\sigma ,z)`$ are defined in (13.4), (13.5), (13.12) and (13.13). In (13.18) and (13.19) the first summands in braces come from the magnetic waves contribution, and second ones from the electric type waves.
Let us now consider the interaction forces between cylindrical surfaces. The force acting per unit area of the inner surface can be presented in the form
$$F_c^{(a)}=F_{c1}^{(a)}+\mathrm{\Delta }F_c^{(a)},\mathrm{\Delta }F_c^{(a)}=p_c^{(ab)}(a),$$
(13.20)
where $`F_{c1}^{(a)}`$ is the force acting on a single cylindrical surface with radius $`a`$, and $`\mathrm{\Delta }F_c^{(a)}`$ is additional force due to the existence of the outer surface and is determined from (13.18). The latter is finite without additional subtractions.
By similar way the force acting per unit area of the outer cylinder
$$F_c^{(b)}=F_{c1}^{(b)}+\mathrm{\Delta }F_c^{(b)},\mathrm{\Delta }F_c^{(b)}=\stackrel{~}{p}_c^{(ab)}(b)$$
(13.21)
where additional term $`\mathrm{\Delta }F_c^{(b)}`$ is due to the existence of the inner cylinder and is defined by (13.19).
The results of the numerical calculations for quantities $`\mathrm{\Delta }q(a,b,r)`$ are given in . Note that the sign of $`\mathrm{\Delta }\epsilon `$ and $`\mathrm{\Delta }p_{c1}`$ is the same as in the case of interior of the parallel plate configuration. In particular the additional forces $`\mathrm{\Delta }F_c^{(a,b)}`$ always have attractive nature.
## 14 Summary
In the present paper we considered a possible way for generalization of Abel-Plana summation formula, proposed in . The generalized version contains two meromorphic functions $`f(z)`$ and $`g(z)`$ and is formulated in the form of Theorem 1. The special choice $`g(z)=if(z)\mathrm{cot}\pi z`$ with $`f(z)`$ being an analytic function in the right half-plane gives APF with additional residue terms coming from the poles of $`f(z)`$. Another consequence from GAPF is the summation formula (2.17) over the points with integer values of an analytic function. An application of this formula to the Casimir effect is given in .
Further we consider the applications to the series and integrals involving Bessel functions. First of all, in section 3 choosing the function $`g(z)`$ in the form (3.2) we derive two types of summation formulae for the series $`_kT_\nu (\lambda _{\nu ,k})f(\lambda _{\nu ,k})`$ (the definition $`T_\nu (z)`$ see (3.12)) with $`\lambda _{\nu ,k}`$ being the zeros of the function $`\overline{J}_\nu (z)=AJ_\nu (z)+BzJ_\nu ^{}(z)`$. Such a type of series arises in a number of problems of mathematical physics with spherical and cylindrical symmetry. As a special case they include Fourier-Bessel and Dini series (see ). Using the formula (3.22) the difference between the sum over zeros of $`\overline{J}_\nu (z)`$ and corresponding integral can be presented in terms of an integral involving Bessel modified functions plus residue terms. For a large class of functions the last integral converges exponentially fast and is useful for numerical calculations. The mode summation method for calculating the vev of the EMT inside perfectly conducting spherical and cylindrical shells used in is based on this formula. In this method the independence of the regularized EMT components on specific form of the cutoff function becomes obvious. APF is a special case of (3.22) with $`\nu =1/2`$, $`A=1`$, $`B=0`$ and an analytic function $`f(z)`$. Choosing $`\nu =1/2`$, $`A=1`$, $`B=2`$ we obtain APF in the form (2.15) useful for fermionic field calculations. Note that the formula (3.22) may be used also for some functions having poles and branch points on the imagianary axis.
The second type of summation formulae, formula (3.37), considered in subsection 3.2 (Theorem 3), is valid for functions satisfying condition (3.31) and presents the difference between the sum over zeros of $`\overline{J}_\nu (z)`$ and corresponding integral in terms of residues over poles for $`f(z)`$ in the right half-plane (including purely imaginary ones). It may be used to summarize a large class of series of this type in finite terms. In particular, the examples we found in literature, when the corresponding sum may be presented in closed form, are special cases of this formula. A number of new series summable by this formula and some classes of functions to which it can be applied is presented.
In Section 4 we consider applications to the series of type $`_kT_\nu ^{AB}(\lambda ,\gamma _{\nu ,k})h(\gamma _{\nu ,k})`$ (with $`T_\nu ^{AB}(\lambda ,z)`$ defined as (4.6)), where $`\gamma _{\nu ,k}`$ are zeros of the function $`\overline{J}_\nu (z)\overline{Y}_\nu (\lambda z)\overline{J}_\nu (\lambda z)\overline{Y}_\nu (z)`$. The corresponding results are formulated in the form of Corollary 2 and Corollary 3. Using the formula (4.13) the difference between the sum and corresponding integral can be expressed as an integral containing Bessel modified functions plus residue terms. For the large class of functions $`h(z)`$ this integral converges exponentially fast. The formula of the second type, (4.16), allows to find in closed form the sums of some types of the series over $`\gamma _{\nu ,k}`$. To evaluate the corresponding integral the formula can be used derived in section 7. This yields to the another summation formula, (4.19), containing residue terms only. The similar formulae can be obtained for the series over zeros of the function $`J_\nu ^{}(z)Y_\nu (\lambda z)J_\nu (\lambda z)Y_\nu ^{}(z)`$ as well. Note that the several examples we found in literature when the corresponding sum was evaluated in closed form are special cases of the formulae considered here. We present new examples and some classes of functions satisfying the corresponding conditions. The possibilities are endless.
The results from GAPF for the integrals of type $`\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)\overline{J}_\nu (x)𝑑x`$ (see notation (3.1)) and $`\mathrm{p}.\mathrm{v}._0^{\mathrm{}}F(x)[J_\nu (x)\mathrm{cos}\delta +Y_\nu (x)\mathrm{sin}\delta ]𝑑x`$ are considered in section 5. The corresponding formulae have the form (5.4), (5.7) and (5.18). In particaular the formula (5.4) is useful to express the integrals containing Bessel functions with oscillating subintegrand through the integrals of modified Bessel functions with exponentially fast convergence. The results obtained in are special cases of these formulae. The illustrating examples of applications of the formulae for integrals are given in section 6 (see (6.3)-(6.7) and (6.13)- (6.17)). Looking the standard books (see, e.g., , -) one will find many particular cases which follow from these formulae. Many new integrals can be evaluated as well. We consider also two examples of functions having purely imaginary poles, (6.19) and (6.23), with corresponding formulae (6.20) and (6.24) (two special cases of these formulae see ).
By choice of the functions $`f(z)`$ and $`g(z)`$ in accord with (7.1) formulae (7.7) and (7.16) for integrals of type
$$\mathrm{p}.\mathrm{v}._0^{\mathrm{}}\frac{J_\nu (x)Y_\mu (\lambda x)J_\mu (\lambda x)Y_\nu (x)}{J_\nu ^2(x)+Y_\nu ^2(x)}F(x)𝑑x$$
can be derived from GAPF. The corresponding results are formulated in the form of Theorem 5 and Theorem 6 in section 7. The several examples for the integrals of this type we have been able to find in literature are particular cases of the formula (7.7). New examples when the integral is evaluated in finite terms are presented. Some classes of functions are distinguished to which the corresponding formulae may be applied.
In the following sections, based on -, the physical applications of the summation formulae obtained from GAPF are reviewed. We consider the vacuum expectation values of the energy-momentum tensor for the electromagnetic field inside and outside the perfectly conducting spherical and cylindrical shells, as well as between two conducting cocentric spherical and coaxial cylindrical surfaces.The corresponding mode sums contain the series over zeros of Bessel functions and their combinations. The application of the summation formulae from sections 3 and 4 allows (i) to extract from corresponding divergent quantities the contribution of the unbounded space in explicitly cutoff independent way, and (ii) to obtain for the regularized values strongly convergent integrals. To compare note that in the Green function method after the subtraction of the Minkowskian part the additional complex frequency rotation is used. In the regularization scheme based on the summation formulae of APF type the complex frequency rotation is made automatically. The corresponding global quantities such as total Casimir energy or forces acting on the surfaces can be obtained from the EMT components. It is shown that in the geometries with two surfaces the additional vacuum forces due to the existence of the second surface always have attractive nature. In the limiting case of the large radii the corresponding results for the Casimir parallel plate configuration are obtained.
Of course the applications of the summation formulae obtained from GAPF are not restricted by the Casimir effect only. Similar types of series will arise in considerations of various physical phenomenon near the boundaries with spherical and cylindrical symmetries, for example in calculations of the electron self-energy and the electron anomalous magnetic moment (for the similar problems in the plane boundary case see, e.g., and references therein). The dependence of these quantities on boundaries originates from the modification of the photon propagator due to the boundary conditions imposed by the walls of the cavity.
## Acknowledgements
I am indebted to Prof. G. Sahakyan, Prof. E. Chubaryan and Prof. A. Mkrtchyan for general encouragement, valuable comments and suggestions. I am grateful to Levon Grigoryan for fruitful colaboration and stimulating discussions. A part of this paper was written during my stay in Tehran, when I given lectures on the Casimir effect at Sharif University of Technology. I acknowledge the Physics Department and Prof. Reza Mansouri for the hospiality. I would also like acknowledge the hospitality of the Abdus Salam International Centre for Theoretical Physics, Trieste, Italy.
|
warning/0002/cond-mat0002413.html
|
ar5iv
|
text
|
# Phonon Scattering and Internal Friction in Dielectric and Metallic Films at Low Temperatures
## I Introduction
Structure and perfection of thin films on substrates are still poorly understood. The purpose of the present investigation is to show that thermal phonons with wavelengths on the order of 100 nm can be used as very sensitive probes of their disorder. It will be shown that strong phonon scattering occurs in a large number of films, including silicon films produced on silicon substrates by molecular beam epitaxy (MBE). The nature of this disorder is, however, not understood.
It has recently been shown that the internal friction of crystalline metal films below 10 K resembles that of amorphous solids both in magnitude and temperature independence, the so-called internal friction plateau . A possible explanation was that crystalline metal films have the same density of tunneling states as amorphous solids. In disordered crystals, such states are called glass-like excitations. According to the Tunneling Model (TM), both the low temperature thermal conductivity below 1 K and the internal friction plateau below 10 K are determined by the same quantity, the tunneling strength $`C`$:
$$C=\frac{\overline{P}\gamma ^2}{\rho v_t^2},$$
(1)
where $`\overline{P}`$ is the uniform spectral density of the tunneling states, $`\gamma `$ is their coupling energy to phonons, $`\rho `$ is the mass density, and $`v_t`$ is the transverse sound velocity. $`C`$ determines the internal friction plateau through relaxational scattering of the elastic wave, and the phonon thermal conductivity through resonant scattering of the thermal phonons. This quantitative connection between internal friction and thermal conductivity has been proven in many cases for bulk amorphous solids and also for disordered crystals, and constitutes a major proof of the validity of the TM, as reviewed in Refs. and .
We have recently described a technique by which we can measure thermal phonon scattering in thin films on substrates . Using this technique, we have verified the quantitative connection between internal friction and thermal conductivity successfully for amorphous silica films and for crystalline silicon layers which had been disordered by ion implantation to the point of amorphization. We will use this technique here for a comparison with the internal friction on a variety of dielectric and metallic films. If the lattice vibration of the films are
indeed glass-like, the phonon scattering should be determined by the same tunneling strength as determined from the internal friction.
In addition to the phonon scattering in the metal films studied previously in internal friction , we will also present measurements of both internal friction and phonon scattering in amorphous Si (a-Si) films, in dielectric crystalline CaF<sub>2</sub> films, and in a crystalline Si film produced by MBE, which is expected to contain fewer defects than any of the other films.
## II Experimental Matters
### A Thin Films
The thin films for heat conduction measurements were deposited either on Czochralski-grown, $`111`$ oriented silicon substrate surfaces, or on float-zone refined, $`100`$ oriented ones. For internal friction measurements, the thin films were deposited onto the double-paddle oscillators, which were float-zone refined, $`100`$ oriented. All substrates were of high purity, and were double-side polished. Film thickness was determined by calibrated vibrations of a 6 MHz plano-convex quartz crystal. When possible, it was double-checked with a step surface profiler. Details on the films for heat conduction measurements are contained in Table I.
In order to eliminate surface contaminations to thermal conduction measurements, samples were, when appropriate, either put through an RCA clean or cleaned in a hot sulfuric acid solution . For the internal friction measurements, all films were deposited directly onto double-paddle oscillators after cleaning of the substrate by diluted HF solution, except for the MBE Si film. Because of the stringent cleaning requirements prior to the MBE deposition in the UHV chamber, which were not suitable for double-paddle oscillators, the MBE Si film was deposited onto a wafer from which an oscillator was subsequently fabricated. The fabrication process involves heating the wafer to 850C for 20 minutes. Thus, the MBE Si film for the internal friction measurement was considered as annealed. All the other annealing processes were done in the MOS area of the Cornell Nanofabrication Facility and were preceded by a stringent RCA cleaning, as described in Ref., in order to avoid any contamination of the silicon which is known to occur during annealing under regular clean laboratory conditions. This annealing process will be referred to in the following as “MOS-cleaned-and-annealed.”
Since we are primarily interested in the thermal phonon mean free path in the film from the heat conduction measurement, we would like to minimize phonon scattering at the film-substrate interface and at the free surface of the film. As the film-substrate interface is expected to be much smoother than the free surface of the film, we first consider the free surface roughness of the films studied in this work. Table II presents the root-mean-square (RMS) roughness of the free surfaces of the films studied as determined by atomic force microscopy (AFM). Table II also presents the dominant thermal phonon wavelength at 1 K based on the Debye speed of sound of the materials listed. The wave length of the thermal phonons is the relevant length. As can be seen, the RMS roughness is small compared to the length scale of the thermal phonons; and so, the free surfaces of the
TABLE II. Comparison of the free surface RMS roughness as determined by AFM measurements with the dominant thermal phonon wavelength $`\lambda _{\mathrm{dom}}`$ at 1 K based on the Debye speed of sound, $`v_D`$, for the thin film samples studied. film $`v_{D}^{}{}_{}{}^{\mathrm{a}}`$ $`\lambda _{\mathrm{dom}}`$ RMS thickness at 1 K Roughness ($`\mu `$m) (10<sup>5</sup> cm/s) (Å) (Å) a-Si 0.5 4.62 510 5 MBE Si 0.4 5.93 650 $`<2`$ CaF<sub>2</sub> 0.1 4.10 450 25 MBE CaF<sub>2</sub> 0.4 4.10 450 7 Al 0.2 3.42 370 40 0.4 3.42 370 20 0.6 3.42 370 40 Al 5056 0.5 3.42 370 40 Ti 0.1 3.48 380 10 Cu 0.1 2.78 300 10 <sup>a</sup> Taken from Ref. or estimated using Eq. 6 with $`v_t`$ listed in Table III. films should have little effect on the overall scattering of thermal phonons and thus on the overall conclusions of this work. In section III C, we will show experimental evidence that phonon scattering at the surfaces/interfaces is indeed negligible relative to that in the films, by studying films thickness dependence.
### B Methods
In contrast to the conventional thermal conductivity measurements on bulk metals, our thermal method, as applied to metal films, has the advantage of solely determining the phonon heat transport in the form of a phonon mean free path in the film. In our investigation the metal films act primarily as phonon scatterers rather than as heat conductors because of their relatively small thickness to that of the substrate. At low temperatures, most of the heat in normal metals is carried by electrons. This thermal conductivity can be calculated using the Wiedemann-Franz-Lorenz law (reviewed in Ref.) and an appropriate electrical resistivity. Because heat transport is parallel to the film-substrate interface, the amount of heat carried by the film or the substrate depends on their relative thermal resistance, which is the inverse of thermal conductivity multiplied by length and divided by cross-sectional area. Since the film and substrate have the same length and width and differ only in thickness, a comparison of the products of thermal conductivity and thickness is enough to determine which carries most of the heat, which is shown in Fig. 1 using a 0.2 $`\mu `$m thick Cu film as an example. The low temperature thermal conductivity of the Cu film was determined with the FIG. 1.: Measured thermal conductivity of a high purity Si substrate (large faces polished and thin faces sandblasted, see Appendix) multiplied by its thickness, 300 $`\mu `$m: solid triangles; and the calculated electronic thermal conductivity of a Cu film (see text) multiplied by its thickness, 0.2 $`\mu `$m: solid line. At 50 mK, only about 5% of the heat is carried in the film (by electrons); at higher temperatures, the percentage drops as $`T^2`$. Wiedemann-Franz-Lorenz law using a room temperature electrical resistivity of $`1.6\times 10^6`$ $`\mathrm{\Omega }`$cm and a residual resistivity ratio of 2, measured in our laboratory on similar films . Fig. 1 demonstrates that the substrate phonons are the dominant heat carriers, primarily because the thickness of the substrate is so much greater than that of the film. The same is true for the other metal films studied in this work, even more so when they become superconducting in the temperature range investigated here.
Thermal phonon mean free paths were determined, using a Monte Carlo (MC) simulation, from heat conduction measurements between 0.05 and 1.0 K by the technique described in Ref., denoted as $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in the following. We mention briefly that additional scattering mechanisms such as scattering from free surface roughness can be included in the simulations, should that become necessary. For those who wish to determine $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ from heat conduction measurements below 1 K without having to resort to performing their own simulations, we provide information in the Appendix, using the results of our MC simulations on a film-substrate sample with dimensions as typically used in our work.
We can also predict the phonon mean free path from internal friction measurements, denoted as $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$, if we assume that the film has the low energy excitations that are common in amorphous solids (and no other scattering centers) within the TM model. In the present work, the low-temperature internal friction of thin films is measured with double-paddle oscillators vibrating in their antisymmetric mode at $`5.5`$ kHz, which have exceptionally small background damping as described previously. Thin films increase the internal friction of the paddle oscillator, $`Q_{\mathrm{paddle}}^1`$. From this, the internal friction of the film, $`Q_{\mathrm{film}}^1`$, is determined by
$$Q_{\mathrm{film}}^1=\frac{G_{\mathrm{sub}}t_{\mathrm{sub}}}{3G_{\mathrm{film}}t_{\mathrm{film}}}(Q_{\mathrm{paddle}}^1Q_{\mathrm{sub}}^1),$$
(2)
where $`t`$ and $`G`$ are thicknesses and shear moduli of substrate and film, respectively, and $`Q_{\mathrm{sub}}^1`$ is the internal friction of the bare paddle (including the mounting losses). $`G_{\mathrm{film}}`$ is assumed to be equal to that of the bulk material . The specific model used to obtain $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ from the internal friction of a film is the TM, originally proposed by Anderson, et al., and independently by Phillips, expanded for elastic measurements by Jäckle. The TM connects the thermal phonon mean free path, $`\mathrm{}`$, with the internal friction plateau, $`Q_0^1`$, as follows. From Ref., the expression for the thermal conductivity $`\mathrm{\Lambda }`$ is
$$\mathrm{\Lambda }=\frac{1}{3}C_vv_D\mathrm{}=\frac{2.66k_B^3}{6\pi \mathrm{}^2}\frac{\pi }{2Q_0^1v_t}T^2$$
(3)
where, in the gas-kinetic picture, $`C_v`$ is the low temperature specific heat per unit volume, $`v_D`$ is the Debye speed of sound, $`k_B`$ is Boltzmann’s constant, $`\mathrm{}`$ is Planck’s constant, and $`T`$ is the temperature. Note that
$$Q_0^1=\frac{\pi }{2}C,$$
(4)
where $`C`$ is defined in Eq. 1. Substituting for $`C_v`$ within the Debye model of the phonon spectrum, Eq. 3 becomes
$$\mathrm{}=(1.59\times 10^{12}[\mathrm{s}\mathrm{K}])\frac{v_t}{Q_0^1}T^1,$$
(5)
assuming the empirical relation:
$$v_t0.9v_D,$$
(6)
where \[s K\] are units of second and Kelvin. These equations provide the means to predict (within the TM) what $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ should be if the internal friction plateau of the film, $`Q_{0}^{1}{}_{\mathrm{film}}{}^{}`$, is known. To repeat, Eq. 5 assumes that the internal friction plateau is due to the presence of glassy states in the film, and that no defects other than the glass-like excitations scatter the thermal phonons. The validity of these assumptions will be tested for the films investigated here by comparing $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ with $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$.
## III Results and Discussion
### A Silicon Films
Since we are searching for tunneling states in thin films, we start with e-beam a-Si, a highly disordered film known FIG. 2.: The internal friction of a bare double-paddle oscillator (solid curve “background”) and of such oscillators carrying e-beam a-Si and MBE Si films. Note the negligible effect of the MBE film. The annealing of the e-beam a-Si film was done at 700C for 1 hr under the MOS-cleaned-and-annealed condition (see text). The MBE Si film had been annealed at 850C for 20 min (see text). FIG. 3.: Internal friction of an e-beam a-Si film, before and after annealing, compared to that of bulk a-SiO<sub>2</sub> (solid curve). The bulk a-SiO<sub>2</sub> data, measured at 4.5 kHz, are taken from J.E. Van Cleve, Ph.D. thesis, Cornell, published in Ref. . The double-headed vertical arrow indicates the range of the temperature-independent internal friction plateau, measured on a wide range of bulk amorphous solids as reviewed in Ref. . to have such states, and will compare it with a crystalline silicon film produced by MBE which is expected to be a simple extension of the silicon lattice. Fig. 2 shows the internal friction of a bare double paddle oscillator, called “background,” and of the same kind of oscillator carrying the films. As expected, the MBE film has negligible internal friction, while the e-beam a-Si films, both as-deposited and annealed, lead to a considerable increase of the internal friction. The MBE Si was measured only in the annealed state as explained in Section II A.
From the change of the internal friction of the paddle carrying the films, the internal friction of the films, $`Q_{\mathrm{film}}^1`$, can be determined using Eq. 2, and is compared to that of bulk a-SiO<sub>2</sub> in Fig. 3. The annealing of the e-beam film causes almost an order of magnitude reduction in $`Q_{\mathrm{film}}^1`$, while the $`Q_{\mathrm{film}}^1`$ of the annealed MBE film is too small to be determined. Annealing at 700C for 1 hr leads to almost complete crystallization of a $`0.5`$ $`\mu `$m thick e-beam a-Si film . The internal friction confirmed that $``$ 90% of the low energy excitations had been removed. Below 1.0 K, however, the internal friction increased, indicative of a contamination in c-Si , although the most stringent MOS-cleaned-and-annealed process, as described in Section II A, was strictly followed. Most probably, an impurity was trapped on the substrate surface during mounting in the e-beam evaporator which is located outside the MOS area. This impurity, trapped underneath the a-Si film, survived the MOS cleaning and led to the contamination during the annealing. In contrast, there is no such problem for the MBE Si film, and hence no contamination is observed.
Table III summarizes $`Q_{0}^{1}{}_{\mathrm{film}}{}^{}`$, $`v_t`$, $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}(T)`$ given by Eq. 5, for the films presented here and below. This en-
TABLE III. The internal friction plateau $`Q_{0}^{1}{}_{\mathrm{film}}{}^{}`$, the transverse speed of sound $`v_t`$, and the thermal phonon mean free path $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$($`T`$) predicted by Eq. 5, where $`T`$ is measured in Kelvin. $`Q_{0}^{1}{}_{\mathrm{film}}{}^{}`$ $`v_t`$ $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$($`T`$) (10<sup>5</sup> cm/s) $`(\mu `$m) a-Si $`1.3\times 10^4`$ 4.16<sup>b</sup> $`50.9/T`$ annealed a-Si $`2.4\times 10^5`$ 5.33<sup>c</sup> $`353/T`$ MBE Si negligible 5.33<sup>c</sup> <sup>e</sup> CaF<sub>2</sub> $`6.0\times 10^5`$ 3.69<sup>c</sup> $`97.8/T`$ Al $`1.0\times 10^4`$<sup>a</sup> 3.04<sup>c</sup> $`48.3/T`$ Al 5056 $`1.0\times 10^5`$<sup>a</sup> 3.04<sup>c</sup> $`483.4/T`$ Ti $`2.0\times 10^4`$<sup>a</sup> 3.13<sup>d</sup> $`24.9/T`$ Cu $`5.3\times 10^4`$<sup>a</sup> 2.50<sup>c</sup> $`7.5/T`$ <sup>a</sup> Taken from Ref. <sup>b</sup> Taken from Ref. <sup>c</sup> Taken from Ref. <sup>d</sup> Taken from Ref. <sup>e</sup> There is no $`Q_{0}^{1}{}_{\mathrm{film}}{}^{}`$ value for the MBE Si film because the internal friction of this film was not detectable (see Fig. 2), and hence $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ is expected to be very long. ables us to compare the internal friction and the phonon mean free path directly within the TM.
Fig. 4 shows $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in the MBE Si and e-beam a-Si films both in as-deposited and annealed states obtained from the heat conduction measurements. The dotted line represents $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ of the e-beam a-Si based on the internal friction measurement (see Table III). For e-beam a-Si, $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ is significantly smaller than $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ which assumes that the scattering occurs by tunneling states alone. Evidently, $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ cannot be used to test for the existence of glassy excitations in e-beam a-Si film. In addition to the tunneling states, other scattering centers must be present. The situation may be similar to e-beam a-SiO<sub>2</sub> films in which the additional thermal phonon scattering was explained by cracks or voids. It is well known that a-Si films produced by e-beam evaporation contain similar defects. After annealing, only a small increase of $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ above 0.3 K and even a decrease below that temperature can be seen in Fig. 4. This change of $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ upon annealing, which cannot be explained within the TM, is interpreted as resulting from a combination of a decreased scattering by the tenfold smaller number of tunneling states, an increased scattering by the contaminants, and possibly a change in scattering from the defects of unknown nature which had been noticed already in the film prior to annealing. Obvi- FIG. 4.: Phonon mean free path $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ of as-deposited and annealed silicon films. As-deposited e-beam a-Si: solid circles; annealed (700C, 1 hr) e-beam a-Si: Open circles; as-deposited MBE Si: solid squares; annealed MBE Si (500C, 1 hr followed by 700C, 1.5 hrs): Open squares. The dotted line is $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}(T)`$, the TM prediction based on the internal friction plateau of the as-deposited e-beam a-Si. Dashed lines are guides for the eye. ously, these results cannot be used to extract any knowledge about the annealing of these unknown defects. They do, however, provide further evidence for the presence of the contaminants introduced into crystalline silicon during annealing under all but the most stringent conditions, thus further emphasizing the need for their identification and control .
Very surprisingly, Fig. 4 shows that the phonon scattering in MBE Si film both before and after annealing is very large as well. Since the contamination that plagued the e-beam a-Si films is not an issue here, secondary ion mass spectroscopy (SIMS) was performed by R. Reedy at the National Renewable Energy Laboratory (NREL) in order to identify other possible chemical impurities. The following chemical elements were detected: boron, $`<10^{16}`$ cm<sup>-3</sup>; nitrogen, $`3\times 10^{16}`$ cm<sup>-3</sup>; carbon, $`3\times 10^{17}`$ cm<sup>-3</sup>; and hydrogen, $`2\times 10^{18}`$ cm<sup>-3</sup>. These concentrations were similar in the MBE film and the silicon substrate. Only the oxygen contents differed between substrate ($`5\times 10^{17}`$ cm<sup>-3</sup>) and film ($`5\times 10^{18}`$ cm<sup>-3</sup>). For all these detected impurities, an anneal (500C, 1 hr followed by 700C, 1.5 hr) caused no measurable change of their concentrations. Since no evidence for such scattering was observed on bare silicon samples from the same batch as used in these experiments, the only possible impurity scatterer is the oxygen. But the tenfold increase of oxygen in the MBE film should not lead to a noticeable decrease of the experimental mean free path $`\mathrm{}`$ (defined in the Appendix, Eq. 7), and thus to a decrease of $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ given the relatively small thickness of the film, unless the oxygen in the film somehow acts as a much stronger scattering center. Phonon scattering in oxygen-doped silicon has been found to depend on heat treatment at frequencies in excess of $`300`$ GHz. But, phonons in this frequency range carry heat predominantly above 3 K, and no evidence exists for phonon scattering by oxygen at lower frequencies (corresponds to $`T<1`$ K).
In an attempt to detect any evidence for structural disorder in this MBE film, an x-ray diffraction (XRD) analysis was performed by M. Sardela and D. Cahill at the University of Illinois (Champaign-Urbana). High resolution open-detector scans around the Si(004) peak showed no difference between measurements conducted on the film side and on the substrate side of the MBE Si film-substrate sample. Full width at half maximum values were found to be almost identical on both sides, and no diffuse scattering or any disorder feature on the film side was seen. In addition, a triple axis reciprocal space map around the Si(004) peak on the film side also could not detect any diffuse distribution nor asymmetry of the Si peak. These observations speak against the existence of grains with different orientations which might cause the phonon scattering.
The near $`T^1`$ temperature dependence may be suggestive of scattering by sessile dislocations as reported, for example, by Wasserbäch in plastically deformed bulk copper, niobium, and tantalum. If we assume that the coupling between dislocations and phonons is similar, the dislocation density in the MBE film would have to range between $`10^{10}`$ and $`10^{13}`$ cm<sup>-2</sup>, which seems rather high for MBE silicon. At this point, no search for dislocations in this MBE film has been undertaken.
MOS-cleaning-and-annealing of the MBE film as described above leads to an increase of $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ shown in Fig. 4, although it remains below the mean free path predicted even for e-beam a-Si, except at the lowest temperature. Thus, noticeable disorder, other than the tunneling states, remains in the MBE Si film even after annealing, although the internal friction of the annealed MBE Si film shown in Fig. 2 give no evidence for any low energy excitations. Thus, the only firm conclusion we can draw at this point is that the defects in the MBE film are not glass-like.
### B CaF<sub>2</sub> Films
As was just shown, crystalline films produced by crystallizing an a-Si film or by MBE deposition show little or no evidence for tunneling states in low temperature internal friction. It is therefore surprising that a 0.6 $`\mu `$m thick film of crystalline e-beam CaF<sub>2</sub> increases the damping of the double paddle oscillator by more than one order of magnitude, see Fig. 5. The internal friction of the film itself, as compared with that of a-SiO<sub>2</sub> shown in Fig. 6, is nearly temperature independent and close to the range found for all amorphous solids studied to date (with the exception of certain hydrogenated a-Si films, as discussed in Ref.). Thus, the large internal friction observed previously in polycrystalline metal films apparently also occurs in some crystalline dielectric films. Assuming that its cause is glass-like excitations, we can again predict an $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ for the CaF<sub>2</sub> film, shown as the dotted line in Fig. 7. The measured $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ for an identical CaF<sub>2</sub> film, also shown in Fig. 7, is more than two orders of magnitude smaller than predicted by the TM. We also measured the phonon mean free path of another crystalline CaF<sub>2</sub> film which was prepared by MBE technique at 750C substrate temperature to improve its structure. However, the internal friction of the MBE CaF<sub>2</sub> film cannot be measured because of the difficulty of preparing such a film on silicon paddle oscillators and meeting the requirements of special cleaning and fabrication at the same time. The $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ of the MBE CaF<sub>2</sub> film, though larger than that of the e-beam one, is still smaller than that predicted by the TM for the e-beam CaF<sub>2</sub> film. Furthermore, it is smaller than that of an a-SiO<sub>2</sub> film prepared by wet-thermal oxidation, in which the structure is much improved in comparison with e-beam a-SiO<sub>2</sub>, and the phonon scattering is determined solely by the glassy excitations. The FIG. 5.: The internal friction of a bare high purity silicon double-paddle oscillator (solid curve “background”) and of such a paddle carrying the e-beam CaF<sub>2</sub> film on the polished silicon surface. FIG. 6.: Internal friction of the e-beam CaF<sub>2</sub> film compared to that of bulk a-SiO<sub>2</sub> (solid curve). The bulk a-SiO<sub>2</sub> data, measured at 4.5 kHz, is taken from J.E. Van Cleve, Ph.D. thesis, Cornell, published in Ref.. The double-headed vertical arrow indicates the range of the temperature-independent internal friction plateau, measured on a wide range of bulk amorphous solids as reviewed in Ref.. $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ of the thermal a-SiO<sub>2</sub> film agrees perfectly with that of the TM’s prediction, just as one would expect for bulk a-SiO<sub>2</sub>, as shown in Fig. 7 (see Ref. for details). The $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ of the MBE CaF<sub>2</sub> film locates between those of the thermal a-SiO<sub>2</sub> film and a macroscopically as well as microscopically disordered e-beam a-SiO<sub>2</sub> film, in which the thermal phonon scattering is not dominated by the tunneling states, also shown in Fig. 7. We suggest that this MBE CaF<sub>2</sub> film, al- FIG. 7.: Phonon mean free path $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ of two different CaF<sub>2</sub> films. MBE CaF<sub>2</sub>: solid circles; e-beam CaF<sub>2</sub>: solid squares. The dotted lines are the TM prediction based on internal friction measurements for e-beam CaF<sub>2</sub> and for thermal a-SiO<sub>2</sub>, respectively. Data for the thermal (open squares) and e-beam (solid stars) a-SiO<sub>2</sub> films are taken from Ref.. For thermal a-SiO<sub>2</sub>, good agreement is shown between $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ and $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$, which had also been found in ion-implanted silicon , as mentioned in Section I. Dashed lines are guides for the eye. though probably more highly ordered than an e-beam one, contains disorder because the pseudomorphic epitaxial growth is known to break down at film thickness exceeding 10 nm, leading to structural relaxation. Internal stresses are also expected to result from differential thermal contraction as the sample is cooled from the deposition temperature. There is, however, no convincing evidence for glass-like excitations in this CaF<sub>2</sub> film. As in the e-beam a-Si film, some unknown scattering process masks the effect of the glass-like excitations, if they exist at all.
### C Metal Films
Fig. 8 shows $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ for three e-beam Al films, 0.2, 0.4, and 0.6 $`\mu `$m thick. The absence of any significant dependence on the film thickness validates the assumption used in our analysis that the scattering occurs predominantly within the films and not at the interfaces, an assumption which so far had been based only on the smoothness observed on the free surfaces as listed in Table II. The dotted line is the prediction for $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ based on the internal friction of the e-beam Al film reported FIG. 8.: Phonon mean free path $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ for e-beam Al films that are 0.2 $`\mu `$m (open triangles), 0.4 $`\mu `$m (open squares), and 0.6 $`\mu `$m (open diamonds) thick. The dotted line is the TM prediction based on internal friction (see Table III). Dashed lines are guides for the eye. earlier (see also Table III). As for the two previous examples, the observed phonon scattering far exceeds the scattering expected on the basis of the TM.
In Ref., it had been shown that the low temperature internal friction of an Al film on a Si substrate was very similar to that of heavily deformed bulk Al. It was therefore interesting to compare the phonon mean free path in the film with that observed in deformed bulk Al. For that purpose, a 99.999% pure polycrystalline Al rod (2.5 mm in diameter and 25.7 mm long) was first annealed at 560C and subsequently stretched by 5%. Its thermal conductivity, measured by the standard steady-state technique, is shown in Fig. 9 along with that of bulk a-SiO<sub>2</sub>. The steep rise of the thermal conductivity of the bulk Al above 0.1 K is caused by the onset of heat transport by normal state electrons. However, below that temperature, heat is expected to be carried predominantly by the lattice, and a temperature dependence similar to that of the bulk glass (a-SiO<sub>2</sub>) is observed. Although the magnitude is three times smaller, it still falls within the glassy range in thermal conductivity, see Fig. 1 in ref. . The phonon thermal conductivity of the 0.2 $`\mu `$m e-beam Al film, as calculated from $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in Fig. 8, is also shown in Fig. 9. Above 0.1 K, the thermal conductivity of the deformed bulk Al and the phonon thermal conductivity of the e-beam Al film show the difference between the heat transport by electrons and phonons, and by the phonons alone, separated here experimentally for the first time. Below 0.1 K, the phonon thermal conduc- FIG. 9.: The thermal conductivity of 5% deformed bulk Al. The thermal conductivity of a bulk a-SiO<sub>2</sub>, taken from Ref., along with the phonon thermal conductivity of the 0.2 $`\mu `$m thick e-beam Al film converted from their phonon mean free path shown in Fig. 8, is shown for comparison. tivity of the e-beam Al film is very close to that of the deformed bulk sample. This suggests that the defects which scatter the phonons in the film are very similar to those in the heavily deformed bulk sample. The same conclusion had been reached previously in internal friction measurements as stated above. The defects causing the internal friction had been tentatively identified as dislocations or dislocation kinks. It is tempting to suggest that the thermal phonons in the films are scattered by the same defects. Since we see in Fig. 8 that $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ and $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ of the e-beam Al films are not connected by the TM, we can conclude that the same holds for the deformed bulk Al because of the similarity in the internal friction and phonon mean free path between the thin films and the bulk samples. Thus, the non-glasslike phonon scattering phenomena observed in this work are not limited to thin films alone. In addition, the defects or the mechanisms causing the resonant scattering of thermal phonons in heat conduction and those leading to the relaxational process in internal friction may not even be related, as shown by the following observation.
The alloy Al 5056 in bulk form has an exceptionally small low temperature internal friction, even as a sputtered film (Table III), which has been explained by dislocation pinning. The $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in this film, however, is still close to that of all other metal films, see Fig. 10. It follows that pinning of dislocations has no influence on the thermal FIG. 10.: Phonon mean free path $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ for a 0.4 $`\mu `$m thick e-beam Al film: open squares; for a 0.5 $`\mu `$m thick sputtered alloy Al 5056 film: solid circles; for a 0.1 $`\mu `$m thick e-beam Ti film: solid triangles; and for a 0.1 $`\mu `$m thick e-beam Cu film: solid squares. The thermal conductivity of for the 5% deformed bulk Al is converted to its mean free path: open circles. The labelled dotted lines are TM predictions of $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ based on internal friction measurements (see Table III). The solid curve is the phonon mean free path in bulk a-SiO<sub>2</sub> taken from Ref.. Dashed lines are guides for the eye. phonon scattering. We conclude that the mechanisms causing the internal friction and the thermal phonon scattering are not understood.
A comparison between $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ and $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ is shown in Fig. 10 for four metal films: Al, alloy Al 5056, Ti, and Cu, with the phonon mean free path of a-SiO<sub>2</sub> for comparison. The apparent lack of correlation between $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ and $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ enable us to generalize the same conclusion from the Al films to other metallic films, which is that if glass-like lattice vibrations exist in them, their effect is masked by the unknown defects.
As observed in internal friction, $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ is unaffected by superconductivity ($`T_c`$ is 0.4 K for Ti, 0.92 for alloy Al 5056, and 1.2 K for Al, ). It is concluded that phonon scattering by electrons is unimportant. Klemens has derived an expression for the phonon-electron scattering coefficient $`P`$ in terms of the electron-phonon scattering coefficient $`E`$. Using the value for $`E`$ measured by Berman and MacDonald for pure copper, we calculate $`\mathrm{}_{\mathrm{film}}`$ (of phonons being scattered by electrons) at 1 K to be $`15`$ $`\mu `$m. This phonon scattering rate (due to electrons) is more than an order of magnitude less than the phonon scattering rate observed in Fig. 10 for the Cu film. Note that the calculation of 15 $`\mu `$m should not be taken too seriously as its assumptions of the adiabatic principle, of a phonon Debye spectrum, and of a free electron gas may not be adequate at these temperatures for a thin polycrystalline Cu film with a residual resistivity ratio of 2. Nevertheless, this estimate agrees with our observation that electron-phonon interaction is not significant in our experiment.
## IV Conclusions
Measurements of the thermal phonon mean free path on films of amorphous and MBE Si, of polycrystalline and MBE CaF<sub>2</sub>, of pure metallic Al, Cu, and Ti, and of the metallic alloy Al 5056 below 1.0 K have revealed, in all cases, similar strong phonon scattering. Scattering by surface and interface roughness can be excluded, since nearly the same $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ has been observed in Al films of different thicknesses. In searching for the origin of this phonon scattering, we have also measured the low temperature internal friction of the Si and CaF<sub>2</sub> films (that of the metal films had been measured previously, Ref.) and also the thermal conductivity of a bulk Al rod after a 5% plastic elongation. In all cases, $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ was found to be much smaller than $`\mathrm{}_{\mathrm{film}(\mathrm{TM})}`$ based on the internal friction and assuming that the lattice vibrations are glass-like. The discrepancy is particularly striking for the MBE Si film in which no internal friction was observed, yet $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ was similar to that found in all other films. In this case, the phonon scattering is particularly puzzling since the film is expected to be structurally more perfect. In all other films, macroscopic defects like grain boundaries, voids, cracks, or dislocations may be the cause for the phonon scattering. In the deformed bulk Al, the phonon mean free path was found to be equal to that in thin Al films. Since in the bulk sample, individual dislocations or aggregates thereof are likely phonon scatterers, they may also be the cause for the scattering in the films. However, dislocation motion, presumably tunneling, which has been invoked to explain the internal friction of deformed Al and of Al films (see Ref.) is an unlikely cause for the thermal phonon scattering since the same $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in Al was also found in the alloy Al 5056, in which dislocation motion appears to be suppressed, resulting in a greatly reduced internal friction. In conclusion, both internal friction and phonon scattering have been shown to be sensitive probes for thin film disorder, including that in MBE Si. The nature of such disorder and the mechanisms by which it affects the elastic and thermal properties are completely unknown. No evidence for the existence of glass-like lattice vibrations has been detected.
Acknowledgements
We gratefully acknowledge the help of Aaron Judy with the AFM measurements, Glen Wilk in preparing the MBE Si film at Texas Instruments (Dallas), and Ken Krebs in fabricating the MBE CaF<sub>2</sub> film at the University of Georgia (Athens) and in providing very useful information on that film’s defects. We thank Mauro Sardela and David Cahill for XRD analysis at the University of Illinois (Champaign-Urbana) and Bob Reedy for the SIMS investigation of the MBE Si film at the National Renewable Energy Laboratory. We also thank R.S. Crandall for fruitful discussions. This work was supported by the National Science Foundation, Grant No. DMR–9701972, the National Renewable Energy Laboratory, Grant No. RAD-8-18668, and the Naval Research Laboratory. Additional support was received from the Cornell Nanofabrication Facility, NSF Grant No. ECS–9319005, and the Cornell Center for Materials Research, Award No. DMR-9121564.
Appendix
The technique used in this investigation for the measurement of the thermal phonon mean free path in thin films has been described before. Although the experimental schematic resembles that of a thermal conductivity measurement, see Fig. 11, it should be emphasized that our experiment leads directly to a thermal phonon mean free path, rather than to a thermal conductivity $`\mathrm{\Lambda }`$, from which the mean free path $`\mathrm{}`$ has to be calculated using the gas-kinetic expression
$$\mathrm{\Lambda }=\frac{1}{3}C_v\overline{\upsilon }\mathrm{},$$
(7)
which requires knowledge of the specific heat $`C_v`$ of the heat carrying excitations or phonons traveling with an average velocity $`\overline{\upsilon }`$. In amorphous solids, for example, this $`C_v`$ cannot be measured. It can only be calculated from $`\overline{\upsilon }`$ through the use of the Debye model.
The analysis of the heat conduction measurements on the silicon substrate carrying the film requires a Monte Carlo simulation which, though straightforward, is nonetheless time-consuming. By strictly adhering to the specifics as listed below (including sample and clamp geometry), one can extract the phonon mean free path in a film, $`\mathrm{}_{\mathrm{film}}`$ (called $`\mathrm{}_{\mathrm{film}(\mathrm{HC})}`$ in this paper), from $`\mathrm{}_{\mathrm{exp}}`$ (the experimentally measured phonon mean free path of the film-substrate sample) without having to repeat any MC simulations, as will be shown in this Appendix.
Fig. 12 is a plot of the results of MC simulations on a film-substrate sample with dimensions typically used in this investigation. For any particular simulation, $`\mathrm{}_{\mathrm{film}}/x`$ is an input parameter where $`x`$ is the film thickness. The output parameter is $`\mathrm{}_{\mathrm{MC}}`$, a simulated phonon mean free path of the film-substrate sample. To determine $`\mathrm{}_{\mathrm{film}}`$, one sets an $`\mathrm{}_{\mathrm{exp}}`$ equal to an $`\mathrm{}_{\mathrm{MC}}`$ in Fig. 12 to find the
FIG. 11.: Schematic of 1-D heat conduction experiment; the sample is cleaved from a high purity commercial silicon wafer (orientation $`111`$ or $`100`$) with both large faces polished; the thin faces are sandblasted as described in Ref.. Also shown is a ballistic path of a thermal phonon from the silicon substrate through a thin film as modeled in the MC simulations. FIG. 12.: Typical plot of $`\mathrm{}_{\mathrm{film}}/x`$ versus $`\mathrm{}_{\mathrm{MC}}`$ where $`x`$ is the film thickness. Normalizing $`\mathrm{}_{\mathrm{film}}`$ with respect to $`x`$ allows the same plot to be used for different film thicknesses as long as the sample and clamp geometry has not changed (see text for details). The actual code of the Monte Carlo programs may be found in Ref.. corresponding $`\mathrm{}_{\mathrm{film}}/x`$; multiplying by $`x`$ then yields the desired value, $`\mathrm{}_{\mathrm{film}}`$, that directly corresponds to the measured $`\mathrm{}_{\mathrm{exp}}`$.
In order to use Fig. 12 as above, the film-substrate sample must be mounted as shown in Fig. 11. Furthermore, the four thin sides of the high purity silicon substrate must be completely roughened (by sandblasting, for example) while the two wide faces must be smooth from Syton polishing, for example), as explained in Ref.. Any film of thickness $`x`$ must cover both wide faces entirely and uniformly. The height of the sample above the top of the base clamp should be 44.5 mm, the width of the sample 7 mm, and the thickness of the substrate 0.279 mm. The height of the heater, cold thermometer, and hot thermometer clamp above the top of the base clamp should be 41 mm, 4 mm, and 25 mm, respectively. The two interfaces between the heater clamp and the sample (areas of contact) should each be of dimensions $`0.279\times 2`$ mm<sup>2</sup> while the four interfaces between the thermometer clamps and the sample should each be $`0.279\times 1.5`$ mm<sup>2</sup>. The effect of varying these details are discussed in Ref..
|
warning/0002/quant-ph0002033.html
|
ar5iv
|
text
|
# Multi-Valued Logic Gates for Quantum Computation
## I Introduction
Binary logic gates and Boolean algebra play an important role in classical and quantum theories of computation. The unit of memory for binary quantum computation is the qu-bit, a quantum system existing in a linear superposition of two basis states, labeled $`|0`$ and $`|1`$. Any computation, however large, can be performed using universal logic gates that operate on a small, fixed number of bits or qubits. In the quantum case, a unitary transformation of any number of entangled qubits can be constructed from logic gates that operate on only two qubits at a time , a result that has no analog in classical reversible logic where three-bit gates are needed to simulate all reversible Boolean functions . We consider the extension of universal quantum logic to the multi-valued domain, where the unit of memory is the qu-dit , a $`d`$-dimensional quantum system with the basis states, $`|0`$, $`|1`$, …, $`|d1`$. This offers greater flexibility in the storage and processing of quantum information, and more importantly, provides an alternate route to the scaling up of quantum computation.
As in the binary case, a tensor product of many such qudits is essential for the efficient storage of information, since the number of dimensions in the Hilbert space scales exponentially with the number of qudits in the system. Allowing $`d`$ to be arbitrary enables a trade-off between the number of qudits making up the quantum computer and the number of levels in each qudit. For example, the linear ion trap quantum computer uses only two levels in each ion for computing, although additional levels can be accessed, and are typically needed, for processing and reading out the state of the ion . By using $`d`$ computational levels in each ion, we reduce the number of ions needed for a computation in this scheme by a factor of $`\mathrm{log}_2d`$, since the Hilbert space of $`n`$ qudits has the same dimensionality as $`n(\mathrm{log}_2d)`$ qubits, namely $`d^n=2^{n\mathrm{log}_2d}`$. Given the difficulty of trapping and coherently manipulating a large number of ions in their vibrational ground state in this scheme, a reduction in the number of ions offered by a multi-valued memory is an advantage.
A tensor product of qudits is also essential for the efficient processing of quantum information. As in the binary case, we build unitary transforms on the whole system from logic gates that operate within and between qudits, creating entangled superpositions, rather than by transforming subsets of a non-entangled, unary Hilbert space. These elementary multi-valued gates are necessarily more complex than their binary counterparts, involving a controlled transformation of all the levels of each qudit. However, a logical network of these gates becomes simpler at this expense, invoking a trade-off between the complexity of each gate and the number of gates needed for a computation . We implement a multi-level gate in the linear ion trap scheme by using multiple lasers to address the different transitions in each ion simultaneously. In this approach, each multi-level gate takes less time to implement than the equivalent binary gate sequence on two-level systems, enabling larger computations within the decoherence time.
Quantum computing in multi-level systems is ideally described using a multi-valued basis for logic. The information stored in a $`d`$-level quantum system is fundamentally non-binary in character, since a measurement collapses the system to one of these $`d`$ levels, specifying a single value for the qudit, rather than the $`\mathrm{log}_2d`$ values characteristic of a binary representation of the same Hilbert space. Moreover, as the $`d`$ levels in a single qudit need not contain any entanglement, two-bit conditional logic among these levels is not well-defined, and cannot simulate multi-level unitary transforms in practice. By contrast, the entanglement between two different $`d`$-level systems enables conditional two-qudit logic gates, which we show to be the elementary operations of multi-valued quantum computing.
Quantum error-correction codes have recently been extended to the multi-valued domain, for correcting errors in a single qudit , and multiple qudits . Fault-tolerant procedures for implementing two-qudit and three-qudit analogs of universal binary gates have also been developed . A proposal for using a correlated photon pair to represent the ternary analog of a qubit has been investigated , but no general scheme for implementing multi-valued quantum logic has been proposed. In the following two sections, we derive a set of one- and two-qudit gates that are sufficient for universal multi-valued computing, and show that these can be implemented using multi-level ions in the linear ion trap model.
## II Multi-valued Logic Gates
We review the gates that are universal for binary quantum logic in a way that facilitates their multi-valued generalization. The universal binary gates belong to a family of unitary transforms described by three parameters . This derives from the fact that up to an overall phase factor, any two-dimensional unitary matrix can be written as
$$Y_2(\lambda ,\nu ,\varphi )=\left[\begin{array}{cc}\mathrm{cos}\lambda & e^{i\nu }\mathrm{sin}\lambda \\ e^{i(\varphi \nu )}\mathrm{sin}\lambda & e^{i\varphi }\mathrm{cos}\lambda \end{array}\right],$$
(1)
expressed in the basis states of a qubit, $`|0`$ and $`|1`$. The parameters $`\lambda `$, $`\nu `$, and $`\varphi `$ are usually taken to be irrational multiples of $`\pi `$ and of each other , since this allows even a single gate in Eq. (1) to generate all single-qubit transforms asymptotically by repeated application. However, we find it more useful to consider these three parameters as arbitrary variables in a simulation, with $`Y_2`$ representing a family of gates that can be implemented by appropriate choice of three physical controls. One of the properties of $`Y_2`$ is that it can transform any known state of a qubit to $`|1`$. Such a transformation, labeled $`Z_2`$, depends on the coefficients of the state being transformed,
$$\begin{array}{c}Z_2(c_0,c_1)=Y_2(\mathrm{cos}^1|c_1|,\mathrm{arg}[c_0c_1^{}],\mathrm{arg}[c_1^{}]):\\ c_0|0+c_1|1|1,\end{array}$$
(2)
where $`|c_0|^2+|c_1|^2=1`$. $`Y_2`$ also contains the phase gate, $`X_2`$, that advances the phase of $`|1`$ without affecting $`|0`$,
$$X_2(\varphi )=Y_2(0,0,\varphi ):\{\begin{array}{c}|1e^{i\varphi }|1;\hfill \\ |0|0.\hfill \end{array}$$
(3)
Using these two transformation properties of $`Y_2`$, we can show that the two-qubit gates that are universal for quantum logic take the form,
$$𝚪_\mathrm{𝟐}[Y_2]=\left[\begin{array}{cc}& \\ \widehat{1}_2\hfill & \widehat{0}\\ & \\ & \\ & \\ \widehat{0}\hfill & Y_2\end{array}\right],$$
(4)
acting in the four-dimensional basis of the two qubits. The two-dimensional identity $`\widehat{1}_2`$ acts in the basis of $`|0,0`$ and $`|0,1`$, and $`Y_2`$ acts in the basis of $`|1,0`$ and $`|1,1`$. Taken together, this transforms the second qubit by $`Y_2`$ conditional on the first qubit being in $`|1`$. The family of gates, $`𝚪_\mathrm{𝟐}[Y_2]`$, is universal for binary quantum logic in the sense that a unitary transform on any number of qubits can be simulated by repeated application of these gates on no more than two qubits at a time.
We generalize Eqs. (2-4) to the multi-valued case. We define $`Z_d`$ as a family of $`d`$-dimensional transforms that maps a known single-qudit state to $`|d1`$,
$$\begin{array}{c}Z_d(c_0,c_1,\mathrm{},c_{d1}):\\ c_0|0+c_1|1+\mathrm{}+c_{d1}|d1|d1,\end{array}$$
(5)
where the $`d`$ complex coefficients, $`c_0,\mathrm{},c_{d1}`$, are normalized to unity, yielding $`2d1`$ real quantities that parametrize $`Z_d`$. As in the binary case, Eq. (5) does not determine $`Z_d`$ uniquely, since it gives the transformation of only one of the $`d`$ states in the basis, namely the superposition state with the coefficients, $`c_0,\mathrm{},c_{d1}`$. Since it reduces this superposition to a single specified state, $`|d1`$, we may regard $`Z_d`$ as an instance of the quantum search algorithm , and relate it asymptotically to the Walsh-Hadamard and phase transforms used in this algorithm. Generalizing Eq. (3), we define the $`d`$-dimensional phase gate $`X_d`$ as a function of a single parameter,
$$X_d(\varphi ):\{\begin{array}{c}|d1e^{i\varphi }|d1;\hfill \\ |p|p\text{ for }pd1,\hfill \end{array}$$
(6)
which advances the phase of $`|d1`$ by $`\varphi `$ without affecting any other state in the qudit. It turns out that the gates, $`Z_d`$ and $`X_d`$, are sufficient to simulate all single-qudit unitary transforms. We can implement these gates by controlling only $`2d`$ real parameters, rather than the $`d^21`$ that correspond to generalizing Eq. (1), thus greatly simplifying the physical realization of single-qudit gates. If $`Y_d`$ represents either $`Z_d`$ or $`X_d`$, then the multi-valued analog of $`𝚪_\mathrm{𝟐}`$ becomes
$$𝚪_\mathrm{𝟐}[Y_d]=\left[\begin{array}{cc}& \\ \widehat{1}_{d^2d}\hfill & \widehat{0}\\ & \\ & \\ & \\ \widehat{0}\hfill & Y_d\end{array}\right],$$
(7)
acting in the $`d^2`$-dimensional basis of two qudits. The identity $`\widehat{1}_{d^2d}`$ acts on the states, $`|0,0`$, …, $`|d2,d1`$, and $`Y_d`$ acts on the remaining $`d`$ states, $`|d1,0`$, …, $`|d1,d1`$. This transforms the second qudit by $`Y_d`$ conditional on the first qudit being in $`|d1`$. We now show that such gates are sufficient for constructing arbitrary unitary transforms on any number of qudits.
Consider an $`N`$-dimensional unitary transform $`𝐔`$ acting on $`n=\mathrm{log}_dN`$ qudits. Each state in the computational Hilbert space can be written as a tensor product of these $`n`$ qudits,
$`|k=|k_1|k_2\mathrm{}|k_n,`$ (8)
$`k=0,1,\mathrm{},N1;k_i=0,1,\mathrm{},d1\text{ for all }i,`$ (9)
where $`k_1k_2\mathrm{}k_n`$ is the base-$`d`$ representation of $`k`$, with $`|k_i`$ denoting the state of the $`i^{\mathrm{th}}`$ qudit. We will use the abbreviation, $`|k_1,k_2,\mathrm{},k_n`$, for $`|k_1`$$`|k_2`$$`|k_n`$. Let the eigenstates of $`𝐔`$ be $`|\mathrm{\Psi }_m`$, for $`m=1,2,\mathrm{},N`$, with corresponding eigenvalues $`e^{i\mathrm{\Psi }_m}`$. Each such eigenstate can be expanded in the computational basis,
$`|\mathrm{\Psi }_m`$ $`=`$ $`c_0|0+\mathrm{}+c_{N1}|N1`$ (10)
$`=`$ $`c_0|0,\mathrm{},0+\mathrm{}+c_{N1}|d1,\mathrm{},d1,`$ (11)
where the coefficients are determined by $`𝐔`$. Following an argument given by Deutsch , we write $`𝐔`$ as a product of $`N`$ unitary transforms, each $`N`$-dimensional, that has the same eigenstates and eigenvalues as that of $`𝐔`$,
$$𝐔=\underset{m=1}{\overset{N}{}}e^{i\mathrm{\Psi }_m}|\mathrm{\Psi }_m\mathrm{\Psi }_m|=𝐖_\mathrm{𝟏}𝐖_\mathrm{𝟐}\mathrm{}𝐖_𝐍,$$
(12)
$$𝐖_𝐦:\{\begin{array}{ccc}|\mathrm{\Psi }_m\hfill & \hfill & e^{i\mathrm{\Psi }_m}|\mathrm{\Psi }_m;\hfill \\ |\mathrm{\Psi }_m^{}\hfill & \hfill & |\mathrm{\Psi }_m^{}\text{ for }m^{}m.\hfill \end{array}$$
(13)
The problem then reduces to simulating $`𝐖_𝐦`$ for an arbitrary $`m`$. We decompose $`𝐖_𝐦`$ into two transforms that are easier to simulate using elementary gates,
$$𝐖_𝐦=𝐙_𝐦^{}𝐗_𝐦𝐙_𝐦=𝐙_𝐦^\mathrm{𝟏}𝐗_𝐦𝐙_𝐦,$$
(14)
where $`𝐙_𝐦`$ and $`𝐗_𝐦`$ are the $`N`$-dimensional analogs of $`Z_d`$ and $`X_d`$. We require only that $`𝐙_𝐦`$ transform the $`m^{\mathrm{th}}`$ eigenstate to $`|N1`$,
$$𝐙_𝐦(c_0,c_1,\mathrm{},c_{N1}):|\mathrm{\Psi }_m|N1,$$
(15)
which does not determine the transform uniquely, as in the case of $`Z_d`$. We define $`𝐗_𝐦`$ as the transform that advances the phase of $`|N1`$ by the $`m^{\mathrm{th}}`$ eigenphase, leaving all other computational states unchanged,
$$𝐗_𝐦(\mathrm{\Psi }_m):\{\begin{array}{c}|N1e^{i\mathrm{\Psi }_m}|N1,\hfill \\ |m^{}|m^{}\text{ for }m^{}N1.\hfill \end{array}$$
(16)
We need to show that $`𝐖_𝐦=𝐙_𝐦^\mathrm{𝟏}𝐗_𝐦𝐙_𝐦`$ satisfies Eq. (13). First note that
$$𝐙_𝐦^\mathrm{𝟏}𝐗_𝐦𝐙_𝐦|\mathrm{\Psi }_m=𝐙_𝐦^\mathrm{𝟏}e^{i\mathrm{\Psi }_m}|N1=e^{i\mathrm{\Psi }_m}|\mathrm{\Psi }_m.$$
(17)
For $`m^{}m`$, the state $`𝐙_𝐦|\mathrm{\Psi }_m^{}`$ has no projection along $`|N1`$,
$$N1|𝐙_𝐦|\mathrm{\Psi }_m^{}=\mathrm{\Psi }_m|𝐙_𝐦^{}𝐙_𝐦|\mathrm{\Psi }_m^{}=\mathrm{\Psi }_m|\mathrm{\Psi }_m^{}=0,$$
which implies that $`𝐗_𝐦`$ has no effect on $`𝐙_𝐦|\mathrm{\Psi }_m^{}`$. Hence,
$$𝐙_𝐦^{}𝐗_𝐦𝐙_𝐦|\mathrm{\Psi }_m^{}=𝐙_𝐦^{}𝐙_𝐦|\mathrm{\Psi }_m^{}=|\mathrm{\Psi }_m^{}.$$
(18)
Combining Eqs. (12-14), we see that $`𝐙_𝐦`$ and $`𝐗_𝐦`$ are sufficient to simulate $`𝐔`$. For $`N=d`$, this implies that $`Z_d`$ and $`X_d`$ contain all single-qudit unitary transforms. In the multi-qudit case, we show that $`𝐙_𝐦`$ and $`𝐗_𝐦`$ can be built from the elementary two-qudit gates, $`𝚪_\mathrm{𝟐}[Z_d]`$ and $`𝚪_\mathrm{𝟐}[X_d]`$, and their one-qudit counterparts, $`Z_d`$ and $`X_d`$. We first show this for the $`n`$-qudit analog of $`𝚪_\mathrm{𝟐}`$,
$$𝚪_𝐧[Y_d]\begin{array}{c}\text{Apply }Y_d\text{ to the }n^{\mathrm{th}}\text{ qudit if and only}\hfill \\ \text{if the first }n1\text{ qudits are in }|d1\text{,}\hfill \end{array}$$
(19)
where $`Y_d=Z_d`$ or $`X_d`$. Eq. (19) has a matrix representation analogous to Eq. (7), with $`\widehat{1}_{d^2d}`$ replaced by $`\widehat{1}_{d^nd}`$. It is easy to see that $`𝐗_𝐦=𝚪_𝐧[X_d(\mathrm{\Psi }_m)]`$, since $`𝐗_𝐦`$ affects only the last computational state, $`|N1=|d1,\mathrm{},d1`$. It is less apparent that $`𝐙_𝐦`$ is also contained in Eq. (19). $`𝐙_𝐦`$ and $`Z_d`$ are similar in their transformation properties in that both take a superposition to a single state. However, $`Z_d`$ acts within the state space of a single qudit, while $`𝐙_𝐦`$ transforms the Hilbert space of all $`n`$ qudits. This suggests that $`𝐙_𝐦`$ can be achieved by using $`𝚪_𝐧[Z_d]`$ to target the last qudit repeatedly, while successively permuting these states with the rest of the states in the computational basis. First, applying $`𝚪_𝐧[Z_d(c_{Nd},\mathrm{},c_{N1})]`$ to $`|\mathrm{\Psi }_m`$ reduces the superposition of the last $`d`$ states in Eq. (10) to $`|N1`$, shown symbolically as
$`\{|d1,\mathrm{},d1,0,\mathrm{},|d1,\mathrm{},d1,d1\}`$ (20)
$`=\{|Nd,\mathrm{},|N1\}|N1.`$ (21)
The superposition of the next $`d1`$ states in $`|\mathrm{\Psi }_m`$ is reduced to $`|N1`$ by permuting these with the last $`d`$ states but $`|N1`$,
$$\begin{array}{c}\{|N2d+1,\mathrm{},|Nd1\}\\ \{|Nd,\mathrm{},|N2\},\end{array}$$
(22)
and using Eq. (21) again. Continuing in this manner, successive blocks of $`d1`$ states in $`|\mathrm{\Psi }_m`$ are permuted with the last $`d`$ states but $`|N1`$, and reduced to $`|N1`$ using $`𝚪_𝐧[Z_d]`$, until the entire $`N`$-dimensional state, $`|\mathrm{\Psi }_m`$, has been so reduced, completing the simulation of $`𝐙_𝐦`$. The permutation of states in Eq. (22) can also be done using $`𝚪_𝐧[Z_d]`$ and $`𝚪_𝐧[X_d]`$. To see this, note that a single-qudit permutation is already contained in $`Z_d`$ and $`X_d`$. In particular, if $`P_d(p,q)`$ denotes the permutation of $`|p`$ and $`|q`$ for $`p,q=0,1,\mathrm{},d1`$, then
$`P_d(p,q)=Z_d^{}(c_0,\mathrm{},c_{d1})X_d(\pi )Z_d(c_0,\mathrm{},c_{d1}),`$ (23)
$`c_p=c_q={\displaystyle \frac{1}{\sqrt{2}}};c_{rp,q}=0.`$ (24)
Permuting two states in the $`n`$-qudit computational basis,
$`|j_1,j_2,\mathrm{},j_n|k_1,k_2,\mathrm{},k_n;`$ (25)
$`j_i,k_i=0,1,\mathrm{},d1\text{ for all }i,`$ (26)
can be done one qudit at a time, starting with the first qudit. The last $`n1`$ qudits in $`|j_1,j_2,\mathrm{},j_n`$ are first converted to $`|d1`$ by applying a single-qudit permutation, $`P_d(j_i,d1)`$, to each qudit $`i`$. Then, conditional
on all but the first qudit being in $`|d1`$, an analog of $`𝚪_𝐧[P_d(j_1,k_1)]`$ is applied to permute the first qudit from $`|j_1`$ to $`|k_1`$. The remaining qudits are then restored to their original states by the same single-qudit permutations, $`P_d(j_i,d1)`$, used earlier. This procedure,
$$\begin{array}{c}_{i=2}^nP_d(j_i,d1):\hfill \\ |j_1,j_2,\mathrm{},j_n|j_1,d1,\mathrm{},d1;\hfill \\ 𝚪_𝐧[P_d(j_1,k_1)]:\hfill \\ |j_1,d1,\mathrm{},d1|k_1,d1,\mathrm{},d1;\hfill \\ _{i=2}^nP_d(j_i,d1):\hfill \\ |k_1,d1,\mathrm{},d1|k_1,j_2,\mathrm{},j_n,\hfill \end{array}$$
is repeated for each of the $`n`$ qudits, permuting $`|j_1,j_2,\mathrm{},j_n`$ and $`|k_1,k_2,\mathrm{},k_n`$ without affecting any other computational state.
Thus, any $`n`$-qudit unitary operator $`𝐔`$ can be written in terms of the logic gates, $`𝚪_𝐧[Y_d]`$, for $`Y_d=Z_d`$ or $`X_d`$. We now show that $`𝚪_𝐧[Y_d]`$ can be built from the two-qudit gates, $`𝚪_\mathrm{𝟐}[Y_d]`$, of Eq. (7). One way of doing this is illustrated in Fig. 1 for $`d>2`$. The horizontal lines denote the qudits, with solid lines denoting the $`n`$ computational qudits and dashed lines denoting additional auxiliary qudits that have been initialized to $`|0`$. This simulation uses $`r=(n2)/(d2)`$ auxiliary qudits ($`x`$ means the smallest integer greater than $`x`$), where $`(n2)/(d2)`$ has been assumed for simplicity to be an integer in the figure. The vertical lines represent the two-qudit conditional gates, originating from the control qudit (which is required to be in $`|d1`$ for the gate to apply) and terminating in a box on the target qudit. The boxes with two rows p and q represent $`𝚪_\mathrm{𝟐}[P_d(p,q)]`$, the conditional permutation of $`|p`$ and $`|q`$. The box containing $`\mathrm{Y}_\mathrm{d}`$ represents $`𝚪_\mathrm{𝟐}[Y_d]`$, for $`Y_d=Z_d`$ or $`X_d`$. We want the combination of all these gates to implement $`𝚪_𝐧[Y_d]`$, applying $`Y_d`$ to qudit n if and only if the first $`n1`$ qudits are in $`|d1`$.
Reading from left to right in the figure, the first permutation, $`𝚪_\mathrm{𝟐}[P_d(0,1)]`$, increments auxiliary qudit n+1 from $`|0`$ to $`|1`$ if and only if qudit 1 is in $`|d1`$. The second permutation, $`𝚪_\mathrm{𝟐}[P_d(1,2)]`$, increments qudit n+1 from $`|1`$ to $`|2`$ if and only if qudit 2 is in $`|d1`$, and so on. Continuing this way, we see that qudit n+1 reaches $`|d1`$ if and only if all of the first $`d1`$ computational qudits are in $`|d1`$. This information is then transferred to the second auxiliary qudit, n+2, by the gate $`𝚪_\mathrm{𝟐}[P_d(0,1)]`$, which increments qudit n+2 from $`|0`$ to $`|1`$ provided qudit n+1 is in $`|d1`$. This procedure is carried out sequentially through all of the computational states, until finally we have the auxiliary qudit n+r reaching the state $`|d1`$ (in the case where $`(n2)/(d2)`$ is an integer) if and only if all of the first $`n1`$ computational qudits are in $`|d1`$. Controlled by this last qudit, $`𝚪_\mathrm{𝟐}[Y_d]`$ then acts on qudit n, completing the simulation of $`𝚪_𝐧[Y_d]`$. Although not shown in the figure, the two-qudit permutation gates $`𝚪_\mathrm{𝟐}[P_d(p,q)]`$ are re-applied to the auxiliary qudits at the end to disentangle them from the computational basis and restore them to $`|0`$ for re-use.
This completes the proof that two-qudit gates of the form, $`𝚪_\mathrm{𝟐}[Z_d]`$ and $`𝚪_\mathrm{𝟐}[X_d]`$, together with the one-qudit gates, $`Z_d`$ and $`X_d`$, are universal for quantum computing.
## III Ion Trap Implementation
In this section, we discuss one method of implementing the gates, $`𝚪_\mathrm{𝟐}[Z_d]`$ and $`𝚪_\mathrm{𝟐}[X_d]`$, in which each qudit is represented by a $`d`$-level atom. We use the linear ion trap scheme for quantum computing, proposed by Cirac and Zoller , to model the two-qudit interaction.
The transform $`X_d`$ does not affect the populations of the $`d`$ states in the qudit, but only changes the phase of $`|d1`$, relative to the other states. Since only one state is affected in the process, $`X_d`$ is effectively the same as its binary counterpart, $`X_2`$, from a physical standpoint. We can implement this transform in the atom by coupling $`|d1`$ to an auxiliary state in the atom using a $`2\pi `$-pulse, which does not leave any population in this state at the end of the pulse. In the interaction picture, the phase of $`|d1`$ after the pulse will be different if the detuning is made time-dependent. One way to realize a time-dependent detuning is by using a Stark field to shift the energies of the two levels over time. The other computational states in the atom are not affected in the process if they are far off-resonance and the fields are sufficiently weak.
Unlike $`X_d`$, the transform $`Z_d`$ involves all of the states in the qudit, acting on a $`d`$-state superposition with the coefficients, $`c_0,c_1,\mathrm{},c_{d1}`$, as shown in Eq. (5). The implementation of such a transform can be posed as a problem in quantum optimal control . If there are
$`2d1`$ physical controls available for manipulating the qudit, an optimization on these controls can be done with the fidelity governed by Eq. (5).
A time-domain approach to this problem in a multi-level atom was studied by Noel and Stroud , where the control parameters were the amplitudes and time delays of a sequence of $`d`$ laser pulses, and the goal was to excite a $`d`$-state wave packet in the atom, starting from the ground state. However, when the ground state population becomes significantly depleted, non-iterative methods may not be sufficient for creating arbitrary wave packets in the atom . We consider an alternate frequency-domain approach to implementing $`Z_d`$, where the control parameters are the amplitudes and phases of different laser fields that are tuned near resonance to $`d1`$ atomic transitions, and that are adiabatically turned on and off over a controlled time period. To realize the two-qudit gates $`𝚪_\mathrm{𝟐}`$, we consider a multi-valued extension of the linear ion trap scheme.
Consider $`q`$ identical $`d`$-level ions confined in a linear harmonic trap with frequency $`\nu _x`$, each of which can interact with $`d1`$ lasers at a given time (see Fig. 2). If each laser is detuned from the associated atomic resonance by $`\nu _x`$, only the center-of-mass (CM) normal mode of the trap is excited in the absence of power broadening. By laser cooling, the ions are initially assumed to be in the vibrational ground state of this mode, where each ion vibrates about its equilibrium position with an amplitude that is small compared to an optical wavelength. The trap is then characterized by a Lamb-Dicke parameter, $`\eta =k_x(\mathrm{}/2m\nu _x)^{1/2}`$, that is small compared to unity, where $`m`$ is the mass of each ion and $`k_x`$ is the laser wave vector along the trap axis. If $`\widehat{a}^{}`$ and $`\widehat{a}`$ are the creation and annihilation operators for the CM mode, and $`\widehat{\sigma }_{jj}=|jj|`$ are the internal projection operators for a given $`d`$-level ion in the trap, the Hamiltonian for this ion in the absence of interaction fields is
$$\widehat{H}_0=\mathrm{}\nu _x(\widehat{a}^{}\widehat{a}+\frac{1}{2})+\underset{j=0}{\overset{d1}{}}\mathrm{}\omega _j\widehat{\sigma }_{jj}.$$
(27)
The computational level scheme considered is shown in Fig. 3, where the transition frequencies, $`\omega _{j,j+1}=`$ $`|\omega _{j+1}\omega _j|`$, are distinct compared to the linewidths of the levels. For the purpose of implementing $`Z_d`$, it is sufficient to have only the neighboring levels coupled. This can be reinforced by using appropriate selection rules to suppress the other transitions. The $`d1`$ neighboring transitions are driven by near-resonant laser fields that have a standing-wave configuration along the trap axis,
$`𝐄(\widehat{x},t)`$ (32)
$`={\displaystyle \underset{j=0}{\overset{d2}{}}}\mathit{ϵ}_{j,j+1}[E_{j,j+1}e^{i\alpha _{j,j+1}t}+\text{c.c.}]\mathrm{cos}(k_{j,j+1}\widehat{x}+\phi )`$
$`={\displaystyle \underset{j=0}{\overset{d2}{}}}\mathit{ϵ}_{j,j+1}[E_{j,j+1}e^{i\alpha _{j,j+1}t}+\text{c.c.}]`$
$`\times \text{[}\mathrm{cos}(\phi ){\displaystyle \frac{\eta _{j,j+1}}{\sqrt{q}}}(\widehat{a}^{}+\widehat{a})\mathrm{sin}(\phi )+𝒪(\eta _{j,j+1}^2)\text{]},`$
where $`E_{j,j+1}`$ and $`\alpha _{j,j+1}`$ are the (complex) field amplitudes and field frequencies corresponding to the atomic transitions, and $`\mathit{ϵ}_{j,j+1}`$ and $`k_{j,j+1}`$ are the associated polarizations and wave vector components. The field dependence on $`\widehat{y}`$ and $`\widehat{z}`$ has been suppressed due to the strong trap confinement along these directions. When $`\phi =\pi /2`$ or $`0`$, the standing waves make a node or antinode at the ion’s equilibrium position, $`\widehat{x}=0`$. We have used
$$\widehat{x}=(\mathrm{}/2qm\nu _x)^{1/2}(\widehat{a}^{}+\widehat{a})$$
(33)
for the displacement of the ion from equilibrium, where $`qm`$ is the effective mass of the CM mode. Each cosine in Eq. (LABEL:field) has been expanded in powers of the corresponding Lamb-Dicke parameter, $`\eta _{j,j+1}=k_{j,j+1}(\mathrm{}/2m\nu _x)^{1/2}`$, and only terms up to first order in $`\eta _{j,j+1}`$ are kept in the limit, $`\eta _{j,j+1}1`$, for each $`j`$. For the level scheme under consideration, the internal dipole moment of the ion is effectively
$$\widehat{𝐝}=\underset{j=0}{\overset{d2}{}}[𝐝_{j,j+1}\widehat{\sigma }_{j,j+1}^{}+𝐝_{j,j+1}^{}\widehat{\sigma }_{j,j+1}],$$
(34)
where $`\widehat{\sigma }_{j,j+1}`$ and $`𝐝_{j,j+1}^{}`$ are the transition operator and matrix element corresponding to the downward transition between levels $`|j`$ and $`|j+1`$. The ion-field interaction is described in the dipole approximation, using the Hamiltonian,
$$\widehat{H}_{\mathrm{dip}}=\widehat{𝐝}𝐄(\widehat{x},t),$$
(35)
where the field depends on the center-of-mass position, $`\widehat{x}`$, of the ion. This interaction couples the electronic ($`\widehat{\sigma },\widehat{\sigma }^{}`$) and vibrational ($`\widehat{a},\widehat{a}^{}`$) degrees of freedom of the ion.
We study the time evolution in the interaction picture, where the operators $`\widehat{a}(t)`$ and $`\widehat{\sigma }_{j,j+1}(t)`$ evolve according to $`H_0`$, as $`\widehat{a}e^{i\nu _xt}`$ and $`\widehat{\sigma }_{j,j+1}e^{i\omega _{j,j+1}t}`$. When $`\phi =0`$, the field expansion in Eq. (32) does not contain $`\widehat{a}^{}`$ and $`\widehat{a}`$ to first approximation, and $`H_{\mathrm{dip}}`$ only affects the internal states of the ion. Tuning each laser to resonance in this case, $`\alpha _{j,j+1}=\omega _{j,j+1}`$, we find that $`H_{\mathrm{dip}}`$ becomes time-independent in the interaction picture under the rotating-wave approximation,
$$\widehat{H}_{\mathrm{dip},\mathrm{V}}=\mathrm{}\underset{j=0}{\overset{d2}{}}[\mathrm{\Omega }_{j,j+1}\widehat{\sigma }_{j,j+1}^{}+\mathrm{\Omega }_{j,j+1}^{}\widehat{\sigma }_{j,j+1}],$$
(36)
where $`\mathrm{\Omega }_{j,j+1}=(𝐝_{j,j+1}\mathit{ϵ}_{j,j+1}E_{j,j+1})/\mathrm{}`$ is the Rabi frequency for the transition between levels $`|j`$ and $`|j+1`$. Alternately, when $`\phi =\pi /2`$, the field is linear in $`\widehat{a}^{}`$ and $`\widehat{a}`$ to first approximation, and $`H_{\mathrm{dip}}`$ affects both the internal and external states of the ion. In this case, detuning each laser above or below resonance by the trap frequency, $`\alpha _{j,j+1}=\omega _{j,j+1}\pm \nu _x`$, we find
$$\widehat{H}_{\mathrm{dip},\mathrm{U}_+}=\mathrm{}\underset{j=0}{\overset{d2}{}}\frac{\eta _{j,j+1}}{\sqrt{q}}[\mathrm{\Omega }_{j,j+1}\widehat{\sigma }_{j,j+1}^{}\widehat{a}^{}+\mathrm{\Omega }_{j,j+1}^{}\widehat{\sigma }_{j,j+1}\widehat{a}];$$
(37)
$$\widehat{H}_{\mathrm{dip},\mathrm{U}_{}}=\mathrm{}\underset{j=0}{\overset{d2}{}}\frac{\eta _{j,j+1}}{\sqrt{q}}[\mathrm{\Omega }_{j,j+1}\widehat{\sigma }_{j,j+1}^{}\widehat{a}+\mathrm{\Omega }_{j,j+1}^{}\widehat{\sigma }_{j,j+1}\widehat{a}^{}].$$
(38)
The unitary time evolution operator corresponding to Eq. (36),
$$\widehat{V}=\mathrm{exp}[i(t/\mathrm{})\widehat{H}_{\mathrm{dip},\mathrm{V}}],$$
(39)
mixes the $`d`$ internal states of the ion without affecting the trap state, and turns out to be sufficient for generating the single-qudit gates $`Z_d`$, up to an overall phase factor. The time evolution operators corresponding to Eqs. (37,38),
$$\widehat{U}_\pm =\mathrm{exp}[i(t/\mathrm{})\widehat{H}_{\mathrm{dip},\mathrm{U}_\pm }],$$
(40)
conditionally couple the internal and external co-ordinates of the ion. Whenever the internal energy of the ion is raised ($`\widehat{\sigma }^{}`$), $`\widehat{U}_+`$ raises the trap energy ($`\widehat{a}^{}`$), while $`\widehat{U}_{}`$ lowers the trap energy ($`\widehat{a}`$), in tandem. This conditionality arises from the rotating-wave approximation, which retains only the energy-conserving terms in the Hamiltonian. Using $`\widehat{U}_\pm `$ and $`\widehat{V}`$ in stages, we will show that the two-qudit gates, $`𝚪_\mathrm{𝟐}[Y_d]`$, can be implemented between two ions in the trap.
First we show how to construct $`Z_d`$ from $`\widehat{V}`$. In the binary case, $`d=2`$, we set all the Rabi frequencies except $`\mathrm{\Omega }_{0,1}`$ equal to zero. The levels $`|0`$ and $`|1`$ then undergo two-level Rabi oscillations,
$$\widehat{V}\mathrm{\Omega }^1\left[\begin{array}{cc}\mathrm{\Omega }C& i\mathrm{\Omega }_{0,1}^{}S\\ i\mathrm{\Omega }_{0,1}S& \mathrm{\Omega }C\end{array}\right]\begin{array}{c}|0\hfill \\ |1\hfill \end{array},$$
(41)
where $`C=\mathrm{cos}\mathrm{\Omega }t,S=\mathrm{sin}\mathrm{\Omega }t`$, and $`\mathrm{\Omega }=|\mathrm{\Omega }_{0,1}|`$. Given a state $`c_0|0+c_1|1`$, we can choose $`\mathrm{\Omega }_{0,1}`$ and $`t`$ such that
$$\frac{\mathrm{\Omega }_{0,1}}{\mathrm{\Omega }}=\frac{c_0^{}c_1}{i|c_0c_1|};\mathrm{cos}\mathrm{\Omega }t=|c_1|,$$
(42)
which makes $`\widehat{V}`$ implement the transform of Eq. (2), up to an overall phase, $`i\mathrm{arg}c_1`$, simulating the binary gate $`Z_2`$. In the ternary case, we set all the Rabi frequencies except $`\mathrm{\Omega }_{0,1}`$ and $`\mathrm{\Omega }_{1,2}`$ equal to zero, leaving a three-level $`\mathrm{\Lambda }`$-system (see Fig. 3). In this case, the levels $`|0`$, $`|1`$ and $`|2`$ evolve according to
$$\widehat{V}\mathrm{\Omega }^2\times $$
(43)
$$\left[\begin{array}{ccc}|\mathrm{\Omega }_{1,2}|^2+|\mathrm{\Omega }_{0,1}|^2C& i\mathrm{\Omega }_{0,1}^{}\mathrm{\Omega }S& \mathrm{\Omega }_{0,1}^{}\mathrm{\Omega }_{1,2}(C1)\\ i\mathrm{\Omega }_{0,1}\mathrm{\Omega }S& \mathrm{\Omega }^2C& i\mathrm{\Omega }_{1,2}\mathrm{\Omega }S\\ \mathrm{\Omega }_{0,1}\mathrm{\Omega }_{1,2}^{}(C1)& i\mathrm{\Omega }_{1,2}^{}\mathrm{\Omega }S& |\mathrm{\Omega }_{0,1}|^2+|\mathrm{\Omega }_{1,2}|^2C\end{array}\right]\begin{array}{c}|0\hfill \\ |1,\hfill \\ |2\hfill \end{array}$$
where $`C=\mathrm{cos}\mathrm{\Omega }t`$, $`S=\mathrm{sin}\mathrm{\Omega }t`$, and $`\mathrm{\Omega }^2=|\mathrm{\Omega }_{0,1}|^2+|\mathrm{\Omega }_{1,2}|^2`$. Given a state $`c_0|0+c_1|1+c_2|2`$, we can choose $`\mathrm{\Omega }_{0,1}`$, $`\mathrm{\Omega }_{1,2}`$ and $`t`$ such that
$$\frac{\mathrm{\Omega }_{0,1}}{\mathrm{\Omega }}=\frac{c_0^{}c_1}{i|c_1|^2}\frac{S}{1C};\frac{\mathrm{\Omega }_{1,2}}{\mathrm{\Omega }}=\frac{ic_1c_2^{}}{S|c_2|};$$
(44)
$$\mathrm{cos}\mathrm{\Omega }t=\frac{|c_1|^2}{1|c_2|}1,$$
which makes $`\widehat{V}`$ implement the transform of Eq. (5) for $`d=3`$, up to an overall phase, $`i\mathrm{arg}c_2`$, simulating the ternary gate $`Z_3`$. In the $`d`$-valued case, we require $`\widehat{V}`$ to implement $`Z_d`$ in Eq. (5) for an arbitrary $`d`$,
$$\begin{array}{c}\widehat{V}(\mathrm{\Omega }_{0,1},\mathrm{\Omega }_{1,2},\mathrm{},\mathrm{\Omega }_{d2,d1};t):\\ c_0|0+c_1|1+\mathrm{}+c_{d1}|d1e^{i\varphi }|d1,\end{array}$$
(45)
where $`\varphi `$ has been introduced to allow for an overall phase offset. The controls in $`\widehat{V}`$ are the $`d1`$ complex Rabi frequencies, and the interaction time. The adjoint of the transform in Eq. (45) can be written as
$$\widehat{V}^{}|d1=e^{i\varphi }[c_0|0+c_1|1+\mathrm{}+c_{d1}|d1].$$
(46)
Projecting this equation onto $`p|`$ and taking the complex conjugate of both sides, we get
$$d1|\widehat{V}|p=c_p^{}e^{i\varphi },p=0,1,\mathrm{},d1,$$
(47)
which relates $`d`$ of the matrix elements of $`\widehat{V}`$ to the coefficients, $`c_0,\mathrm{},c_{d1}`$. These $`d`$ equations have to be inverted to find the controls, $`\mathrm{\Omega }_{0,1},\mathrm{},\mathrm{\Omega }_{d2,d1}`$, and $`t`$. Analytical solutions are given for the binary and ternary cases in Eqs. (42) and (44). This allows $`\widehat{V}`$ to implement $`Z_d`$ up to a phase gate, $`X_d(\varphi )`$.
Two-qudit gates of the form $`𝚪_\mathrm{𝟐}[Y_d]`$ can be implemented using both $`U_\pm `$ and $`V`$ interactions, and an auxiliary manifold of $`d`$ additional levels in each ion. To see this, write the original state of the two-ion system in the form,
$$|\mathrm{\Psi }_\mathrm{C}|\mathrm{\Phi }_\mathrm{T}|\underset{¯}{0},$$
(48)
where $`|\mathrm{\Psi }_\mathrm{C}`$ is the original control ion state, $`|\mathrm{\Phi }_\mathrm{T}`$ is the original target ion state, and $`|\underset{¯}{0}`$ is the trap ground state. Applying a $`\pi `$-pulse of the $`U_\pm `$ interaction to the control ion, we first transform all of the computational states in this ion except $`|d1_\mathrm{C}`$ to their auxiliary counterparts, conditional on exciting the trap to $`|\underset{¯}{1}`$. We leave $`|d1_\mathrm{C}|\underset{¯}{0}`$ unaffected by turning off the corresponding laser in $`U_\pm `$. We then restore the internal state of the control ion to its original configuration by using a $`\pi `$-pulse of the $`V`$ interaction, which does not affect the trap. The entangled state of the system is then given by
$$|\mathrm{\Psi }_{d1}_\mathrm{C}|\mathrm{\Phi }_\mathrm{T}|\underset{¯}{0}+|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}|\mathrm{\Phi }_\mathrm{T}|\underset{¯}{1},$$
(49)
where $`|\mathrm{\Psi }_\mathrm{C}=|\mathrm{\Psi }_{d1}_\mathrm{C}+|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}`$. We see that all of the control states except $`|d1_\mathrm{C}`$ are entangled with the trap state $`|\underset{¯}{1}`$. Applying a $`\pi `$-pulse of the $`U_\pm `$ interaction to the target ion now, we transform all of its computational states to their auxiliary counterparts, conditional on de-exciting the trap,
$$|\mathrm{\Psi }_{d1}_\mathrm{C}|\mathrm{\Phi }_\mathrm{T}|\underset{¯}{0}+|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}|\mathrm{\Phi }_{\mathrm{aux}}_\mathrm{T}|\underset{¯}{0},$$
(50)
where $`|\mathrm{\Phi }_{\mathrm{aux}}_\mathrm{T}`$ is the original target ion state written in the auxiliary basis. The first term in expression (49) is not affected by this operation since the trap ground state cannot be de-excited. Next, applying $`\widehat{V}`$ in the computational basis of the target ion, we simulate $`Y_d=Z_d`$ or $`X_d`$, transforming $`|\mathrm{\Phi }_\mathrm{T}`$ but not affecting $`|\mathrm{\Phi }_{\mathrm{aux}}_\mathrm{T}`$,
$$|\mathrm{\Psi }_{d1}_\mathrm{C}\{\widehat{Y}_d|\mathrm{\Phi }_\mathrm{T}\}|\underset{¯}{0}+|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}|\mathrm{\Phi }_{\mathrm{aux}}_\mathrm{T}|\underset{¯}{0}.$$
(51)
The target ion state $`|\mathrm{\Phi }_{\mathrm{aux}}_\mathrm{T}`$ is then restored to the computational basis by reversing the operations that took us from expression (48) to expression (50), giving
$$|\mathrm{\Psi }_{d1}_\mathrm{C}\{\widehat{Y}_d|\mathrm{\Phi }_\mathrm{T}\}|\underset{¯}{0}+|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}|\mathrm{\Phi }_\mathrm{T}|\underset{¯}{0}.$$
(52)
This completes the implementation of $`𝚪_\mathrm{𝟐}[Y_d]`$ on expression (48), with the target ion transformed by $`Y_d`$ conditional on the control ion being in $`|d1_\mathrm{C}`$. This two-qudit logic is made possible by the $`\widehat{U}_\pm `$ interaction, which allows the information about whether the control ion is in $`|\mathrm{\Psi }_{d1}_\mathrm{C}`$ or $`|\mathrm{\Psi }_{\mathrm{other}}_\mathrm{C}`$ to be carried to the target ion via entanglement with the trap states, $`|\underset{¯}{0}`$ or $`|\underset{¯}{1}`$. This shows that universal multi-valued computing is feasible in the linear ion trap scheme.
## IV Summary
We conclude with a comparison of the binary and multi-valued approaches to quantum computing. The main advantage of the latter is a logarithmic reduction in the number of separate quantum systems needed to span the quantum memory. For a Hilbert space of $`N`$ dimensions, corresponding to $`n_2=\mathrm{log}_2N`$ qubits, the number of qudits needed to store this information is
$$n=\frac{\mathrm{log}_2N}{\mathrm{log}_2d}=\frac{n_2}{\mathrm{log}_2d}.$$
(53)
Note that this retains the same scaling in $`N`$ and $`d`$ with the inclusion of the auxiliary qudits used in the gate construction of Fig. 1. Using a binary equivalent of this construction, we find that the overall time-complexity of a binary simulation is $`𝒪[n_2^2N^2]`$. That is, this many two-qubit gates are required to simulate an $`N`$-dimensional unitary operator $`𝐔`$. By analogy, the number of two-qudit gates used in the construction of section II scales as $`𝒪[n^2N^2]`$, or
$$𝒪\left[n^2N^2\right]=𝒪\left[\frac{(\mathrm{log}_2N)^2N^2}{(\mathrm{log}_2d)^2}\right]=𝒪[\frac{n_2^2N^2}{(\mathrm{log}_2d)^2}],$$
(54)
where we have used Eq. (53). Eq. (54) represents an upper bound on the time-complexity of a multi-valued simulation, and shows that this has a $`(\mathrm{log}_2d)^2`$ advantage over the binary case. This comes at the cost of larger elementary gates, $`𝚪_\mathrm{𝟐}[Y_d]`$, which require $`d`$ states in each ion to be controlled, not just two. This suggests that we ought to multiply the multi-valued time-complexity in Eq. (54) by $`d`$ for a physically relevant comparison with the binary case. However, the frequency-domain approach taken to constructing the $`V`$ and $`U_\pm `$ interactions in Eqs. (36-38) assumes that the $`d1`$ lasers operate simultaneously on each ion, which allows a two-qudit gate to be implemented without slowing down the computation in real time. The cost of the multi-valued speed-up in this case is the need for multiple lasers to address the corresponding transitions in the $`d`$-level ion.
Finally, we must make note of the non-logarithmic scaling in $`N`$ in Eq. (54), which shows the inefficiency of this construction for simulating arbitrary $`N`$-dimensional unitary transforms. This is analogous to the binary case, where only certain unitary transforms admit an efficient simulation in terms of elementary gates, making them useful for efficient quantum algorithms.
We thank David Aronstein for helpful comments. This work was supported by the Army Research Office through the MURI Center for Quantum Information.
|
warning/0002/cond-mat0002431.html
|
ar5iv
|
text
|
# The puzzle of 90∘ reorientation in the vortex lattice of borocarbide superconductors
\[
## Abstract
We explain 90 reorientation in the vortex lattice of borocarbide superconductors on the basis of a phenomenological extension of the nonlocal London model that takes full account of the symmetry of the system. We propose microscopic mechanisms that could generate the correction terms and point out the important role of the superconducting gap anisotropy.
\]
Abrikosov vortices in type two superconductors repel each other and therefore tend to form two dimensional lattices when thermal fluctuations or disorder are not strong enough to destroy lateral correlations. In isotropic s–wave materials the lattices are triangular, however in anisotropic materials or for ”unconventional” d–wave or p–wave pairing interactions less symmetric vortex lattices (VL) can form as recent experiment on high $`T_c`$ cuprates , $`SrRuO_4`$ and borocarbides showed. The quality of samples in the last kind of superconductors allows detailed reconstruction of the phase diagram by means of small angle neutron scattering, scanning tunnelling microscopy or Bitter decoration technique. For $`H||c`$ the presence of a whole series of structural transformations of VL was firmly established. At first, stable at high magnetic fields square lattice becomes rhombic, or ”distorted triangular”, via a second order phase transition . Then, at lower fields, 45 reorientation of VL relative to crystal axis occurs. For $`H||a`$ a continuous lock-in phase transition was predicted. Above this transition apex angle of elementary rhombic cell of VL does not depend on magnetic field, but below a critical field such a dependence appears.
Theoretically the mixed state in nonmagnetic borocarbide superconductors $`RNi_2B_2C`$, $`R=Y,Lu`$ can be understood in the frame of the extended London model (in regions of the phase diagram close to $`H_{c2}(T)`$ line the extended Ginzburg–Landau model can be used ). So far this theory always provided qualitative and even quantitative description of phase transitions in VL and various other properties such as magnetization behavior , dependence of nonlocal properties on the disorder, etc. However, recently another ”reorientation” phase transition has been clearly observed in neutron scattering experiment on $`LuNi_2B_2C,`$ which cannot be explained by the theory despite considerable efforts. When magnetic field of $`0.3T`$ was applied along the $`a`$ axis of this tetragonal superconductor sudden 90 reorientation of VL has been seen . At this point a rhombic (nearly hexagonal, apex angle $`60^{}`$) lattice, oriented in such a way that the crystallographic axes are its symmetry axes, gets rotated by $`90^{}.`$ Both initial and rotated lattices are found to coexist at the field range of about $`0.1T`$ wide around the transition. Similar observations have been made in magnetic material $`ErNi_2B_2C.`$
In this Letter we explain why the extended London model in its original form cannot generally explain even the existence of the $`90^{}`$ reorientation transition. The reason is that it possesses a ”hidden” spurious fourfold symmetry preventing such a transition. Then we generalize the model to include the symmetry breaking effect and explain why the reorientation take place. Then we search for a microscopic origin of this effect. Using BCS type theory we find that anisotropy of the Fermi surface is ruled out due to smallness of its contribution. It is anisotropy of the pairing interaction that provides the required mechanism. We, therefore, suggest that there exist a correlation between the critical field of 90 reorientation in VL and the value of the anisotropy of the gap.
A convenient starting point of any generalized ”London” model is the linearized relation between the supercurrent $`j_i`$ and the vector potential $`A_j`$:
$$(4\pi /c)j_i(𝐪)=K_{ij}(𝐪)A_j(𝐪).$$
(1)
In the standard London limit the kernel $`K_{ij}(𝐪)`$ is approximated just by its $`q=0`$ limit, inverse mass matrix, while in the extended London model the quadratic terms of the expansion of the kernel near $`q=0`$ are also kept:
$$K_{ij}(𝐪)=m_{ij}^1/\lambda ^2+n_{ij,kl}q_kq_l.$$
(2)
The significance of the quantity $`n_{ij,kl}`$ is that it allows proper account of the symmetry of any crystal system while the first term does not guarantee this. At the same time it expresses nonlocal effects which are inherent to the electrodynamics of superconductors and below we call its component or their combination nonlocal parameters. From its definition, $`n_{ij,kl}\frac{1}{2}\frac{^2}{q_lq_m}K_{ij}(q)|_{q=0}`$ is a tensor with respect to both the first and the second pairs of indices. However, the way $`n_{ij,kl}`$ transforms when the first and the second pairs of indices are interchanged is not obvious because the ”origin” of these indices are quite different. The first pair $`(ij)`$ comes, roughly speaking, from the variation of the free energy of the system ”a superconductor in weakly inhomogeneous magnetic field” with respect to the vector potentia while the second pair $`(kl)`$ comes from the expansion in vector $`𝐪.`$ Below we show that in general no symmetry $`n_{ij,kl}=n_{kl,ij}`$ is expected.
The original derivation Eq. (2) from BCS theory in quasiclassical Eilenberger formulation produced a fully symmetric rank four tensor: $`n_{ij,kl}v_iv_jv_kv_l`$ with $`v_i`$ being components of velocity of electrons at the Fermi surface. In this calculation independence of the gap function on the orientation was assumed. Let us consider vortex lattice problem with this result. Specializing to tetragonal borocarbides, the number of independent component of tensor $`n_{ij,kl}`$ is four: $`n_{aaaa,}`$ $`n_{aabb},`$ $`n_{aacc}`$ and $`n_{cccc}.`$ In the case of external magnetic field oriented along $`a`$ axis the free energy of VL, which is the relevant thermodynamic potential for a thin plate sample in perpendicular external field, reads
$`F`$ $`=`$ $`\left(B^2/8\pi \right){\displaystyle \left[1+D(g_x,g_y)\right]^1},`$ (3)
$`D`$ $`=`$ $`\lambda ^2(m_ag_x^2+m_cg_y^2)+\lambda ^4\left[n(m_ag_x^2+m_cg_y^2)^2+dg_x^2g_y^2\right].`$ (4)
Here $`B`$ is magnetic induction and the summation runs over all vectors $`𝐠`$ of the reciprocal VL. The nonlocal parameters appearing in this equation have the form $`n=n_{aacc}`$ and $`d=n_{cccc}m_c^2+n_{aaaa}m_a^26n_{aacc}m_am_c.`$ The free energy of Eq. (3) has been extensively studied numerically first minimizing it on the class of rhombic lattices with symmetry axes coinciding with the crystallographic axes and more recently by us for arbitrary lattices with one flux per unit cell. Despite the fact that great variety of vortex lattice transformation were identified, no a $`90^{}`$ reorientation has been ever seen. The reason is quite simple: the considered free energy is actually effectively fourfold symmetric. After rescaling the reciprocal lattice vectors
$$g_x\stackrel{~}{g}_xg_x/\sqrt{m_a},g_y\stackrel{~}{g}_yg_y/\sqrt{m_c}$$
(5)
the sum in Eq.(3) becomes fourfold symmetric explicitly. Based on this observation one concludes that energies of the lattices participating in the $`90^{}`$ reorientation are equal exactly. Therefore no phase transition between them is possible in the framework of the extended London model of Eq.(3) and further corrections are necessary to account for this transition.
There might be a slight possibility that the observed $`90^{}`$ reorientation presents the lock-in transition described in the beginning of this paper. For this to happen the rescaled square VL should looks almost hexagonal and, correspondingly, a particular value of masses asymmetry $`m_a/m_c=\left[\mathrm{cos}\left(60^{}\right)/\mathrm{cos}\left(45^{}\right)\right]^2=1/2`$ is required. This is very different from the figures quoted in literature : $`m_a/m_c=0.9/1.22`$ $`=0.74`$. More importantly, according to this scenario one should see two degenerate lattices at small fields below the transition and only a single lattice at high fields above the transition which experimentally is clearly not the case.
To explain $`90^{}`$ reorientation we proceed by correcting the model of Eq. (3). On general symmetry grounds for $`H||a`$ one can expect more terms in the expression for $`D`$ which describes vortex-vortex interactions. Given two fold symmetry of the present case we write down for $`D`$ the expansion in Fourier series up to fourth harmonics, perform rescaling defined by Eq. (5) and obtain
$$D_{eff}=D_0(\stackrel{~}{g})+D_4(\stackrel{~}{g})\mathrm{cos}(4\phi )+D_2(\stackrel{~}{g})\mathrm{cos}(2\phi ),$$
(6)
where $`\phi `$ is the polar angle in the rescaled $`bc`$ plane. The quantity $`D`$ from Eq. (3) produces only fourfold invariant terms:
$`D_0(\stackrel{~}{g})`$ $`=`$ $`\lambda ^2\stackrel{~}{g}^2+\left(n+d/8m_am_c\right)\lambda ^4\stackrel{~}{g}^4,`$ (7)
$`D_4(\stackrel{~}{g})`$ $`=`$ $`\left(d/8m_am_c\right)\lambda ^4\stackrel{~}{g}^4.`$ (8)
The new term $`D_2(\stackrel{~}{g})`$ expresses the effective fourfold symmetry breaking. Experimentally, it should be small as indicated by recent success in qualitative understanding the angle dependence of magnetization of $`LuNi_2B_2C`$ with field lying in the $`ab`$ plane on the basis of the theory without $`D_2`$ term. Accordingly, we can treat it perturbatively:$`F=F^{(0)}+F^{^{(pert)}}`$ with
$`F^{(0)}`$ $`=`$ $`\left(B^2/8\pi \right){\displaystyle \left[1+D_0+D_4\mathrm{cos}(4\phi )\right]^1},`$ (9)
$`F^{(pert)}`$ $`=`$ $`\left(B^2/8\pi \right){\displaystyle \frac{D_2\mathrm{cos}(2\phi )}{\left[1+D_0+D_4\mathrm{cos}(4\phi )\right]^2}}`$ (10)
where the summation is over $`\stackrel{~}{𝐠}`$ (see Eq. (5)). The original degeneracy of two VL rotated by $`90^{}`$ with respect to each other is split now. To explain the $`90^{}`$ reorientation the sign of the perturbation should change at certain field $`B_{reo}`$. Magnetic field influences the sum via constraint that area of the unit cell carries one fluxon. Roughly speaking $`D_2(g)`$ should change sign when $`\stackrel{~}{g}\sqrt{B_{reo}/\mathrm{\Phi }_0}.`$ The simplest way to implement this idea is to write for $`D_2(g)`$ two lowest order terms in $`\stackrel{~}{g}:`$
$`D_2=w_4\stackrel{~}{g}^4+w_6\stackrel{~}{g}^6`$ (11)
Quadratic term is not present since we have already rescaled it out in derivation of Eq. (6). In principle the coefficient $`w_6`$ can be derived from BCS similarly to $`n_{ij,kl}`$ tensor within the framework of original extended London model . Then it is proportional to the Fermi surface average of six components of Fermi velocity. To obtain $`w_4,`$ however, the result $`n_{ij,kl}v_iv_jv_kv_l`$ of Ref. is not sufficient. Indeed, using general expression for $`n_{ij,kl}`$ and repeating derivation of Eq. (3) from Eq. (12) we see that
$`w_4=\left(n_{aa,cc}n_{cc,aa}\right)/2.`$ (12)
In what follows we first demonstrate the presence of the first order phase transition in the model of Eq. (11) and then provide a microscopical derivation of $`w_{4.}`$
The critical magnetic field of the 90 reorientation $`B_{reo}`$ depends only on the ratio $`r=\lambda ^2w_6/w_4.`$ We determined this dependence numerically using standard computational methods. At first, for a fixed $`B`$ the equilibrium form of VL unit cell was obtained by minimization of Eq.(10). Then, the zero of the perturbation energy Eq. (10) was found. As usual during the numerical calculations the cutoff factor $`\mathrm{exp}\left(\xi ^2\stackrel{~}{𝐠}^2\right)`$ was introduced inside the above sums in order to properly account for the failure of the London approach in the vortex core. The calculated critical field is presented on Fig. 1 (we used $`d=0.05`$ and $`n=0.015`$ typical for $`\mathrm{𝐿𝑢𝑁𝑖}_2B_2C`$). We see that within the approximation of Eq. (11) the $`90^{}`$ reorientation cannot happen at very low magnetic fields. For $`\mathrm{𝐿𝑢𝑁𝑖}_2B_2C`$ with $`\lambda 710`$Å, the field unit $`\mathrm{\Phi }_0/(2\pi \lambda )^2`$ is about $`100G`$. From the experimentally observed transition field $`B_{reo}=2.95kOe`$ we estimate the relative strength of sixth and fourth order terms in $`D_2`$ (see Eq. (11)) as $`r=0.036.`$
To obtain $`w_4`$ we start by discussing a general pairing model which includes anisotropies in both the dispersion relation of electrons and the singlet pairing interaction
$`H[\psi ]`$ $`=`$ $`{\displaystyle _x}\psi _\alpha ^{}\left[\epsilon \left(i\right)\mu \right]\psi _\alpha +V,`$ (13)
$`V`$ $`=`$ $`{\displaystyle \frac{\lambda }{4}}{\displaystyle _x}\psi _\alpha ^{}\left[1+\delta \left(i\right)\right]\psi _\alpha ^{}\psi _\alpha \left[1+\delta \left(i\right)\right]\psi _\alpha ,`$ (14)
where the summation over spin indices $`\alpha =,`$ is assumed. Here $`\psi (x)`$ and $`\psi ^{}(x)`$ are the electron destruction and creation operators, $`\mu `$ is chemical potential and $`\lambda `$ is a positive constant factorized from the pairing interaction for convenience. Dispersion relation $`\epsilon (𝐤)`$ and pairing interaction, of which $`\delta (𝐤)`$ is a part, are usually defined in $`𝐤`$-space. To treat magnetic field effects it is advantageous to define them in coordinate space. According to the rules of quantum mechanics, in the above functions of $`𝐤`$ we perform the replacements $`𝐤i`$ or $`𝐤`$ $`i`$ depending on whether derivatives act on $`\psi `$ or $`\psi ^{}`$. Then the standard minimal substitution $`i𝚷i𝐀`$ can be accomplished. This procedure, however, is not unique because the components of $`𝚷`$ do not commute with each other. Therefore, $`\epsilon (𝐤)`$ and $`\delta (𝐤)`$ are presented by their Taylor expansions and in those terms which contain mixed derivatives the symmetrization in $`\mathrm{\Pi }_i`$ is used.
The kernel $`K_{ij}(𝐪)`$ from Eq. (1) is obtained by treating the effect of slowly varying magnetic field in terms of the linear response. The change in the Hamiltonian due to the presence of magnetic field $`H_1[\psi ,𝐀]H[\psi ,𝐀]H[\psi ,0]`$ is taken into account perturbatively. The result reads (see, for example, Ref. ):
$`K_{ij}(𝐱𝐲)={\displaystyle \frac{^2H_1}{A_i(𝐱)A_j(𝐲)}}{\displaystyle \frac{H_1}{A_i(𝐱)}}{\displaystyle \frac{H_1}{A_j(𝐲)}},`$ (15)
where angular brackets denote the statistical average with unperturbed density operator $`\mathrm{exp}\left(H[\psi ,0]/T\right).`$ Thus, we have to expand the functional $`H_1`$ up to the terms quadratic in $`𝐀.`$ Because our aim is to calculate $`w_4`$ we need only the coefficients of this expansion for $`A_z`$ and $`A_z/x`$.
In its full generally the problem of Eq. (13) in magnetic field is quite intractable and below we consider two particular cases which help us to estimate quantitatively the magnitude of different contributions to $`w_4`$: i) isotropic superconducting interaction, $`\delta (𝐤)=0,`$ and arbitrary dispersion relation $`\epsilon (𝐤)`$; ii) an example of weakly $`𝐤`$ dependent superconducting electronic interaction, $`\delta (𝐤)=\delta _0k_z^2,`$ and isotropic dispersion of the standard form $`\epsilon (𝐤)=𝐤^2/(2m)`$. For simplicity in both cases a clean system was investigated.
In the case i) we obtain
$`H_1`$ $`=`$ $`{\displaystyle _𝐱}\psi _\alpha ^{}\left[A_z\epsilon _{,z}i\left(_xA_z\right){\displaystyle \frac{\epsilon _{,zx}}{2}}\left(_x^2A_z\right){\displaystyle \frac{\epsilon _{,zx^2}}{6}}\right]\psi _\alpha `$ (18)
$`+{\displaystyle \frac{1}{2}}{\displaystyle _𝐱}\psi _\alpha ^{}[A_z^2\epsilon _{,z^2}iA_z\left(_xA_z\right)\epsilon _{,z^2x}`$
$`A_z\left(_x^2A_z\right){\displaystyle \frac{\epsilon _{,z^2x^2}}{3}}\left(_xA_z\right)^2{\displaystyle \frac{\epsilon _{,z^2x^2}}{4}}]\psi _\alpha ,`$
where $`\epsilon _{,zx}`$ means the second derivative of $`\epsilon (i)`$ with respect to $`z`$ and $`x`$ components of the argument, and so on. The final results reads
$`w_4`$ $`=`$ $`{\displaystyle \frac{1}{V}}{\displaystyle \underset{𝐤}{}}[{\displaystyle \frac{2}{3}}R\epsilon _{,z}\epsilon _{,zx^2}+{\displaystyle \frac{R}{\epsilon }}\epsilon _{,z}^2\epsilon _{,x^2}(xz)],`$ (19)
$`R`$ $`=`$ $`{\displaystyle \frac{4}{T}}\mathrm{cosh}^2\left[{\displaystyle \frac{E(𝐤)}{2T}}\right],E(𝐤)=\sqrt{(\epsilon (𝐤)\mu )^2+\mathrm{\Delta }^2}.`$ (20)
At zero temperature $`R`$ approaches zero exponentially and $`w_4`$ vanishes. As temperature increases, $`w_4`$ increases monotonically and reaches its maximal value at $`T=T_c`$ where it smoothly joins the corresponding component of $`q`$-dependent magnetic susceptibility tensor of the normal metal. For estimation we considered a simple dispersion relation $`\epsilon (𝐤)=\frac{\mathrm{𝟏}}{2m}𝐤^2+\frac{\stackrel{~}{\alpha }}{4}k_z^4`$ and assumed deviations from spherical Fermi surface to be small: $`\alpha \stackrel{~}{\alpha }m^2\mu 1.`$ Expanding in $`\alpha `$ we obtain at $`T=T_c`$ that
$$w_4^{FS}=2\alpha \mathrm{\Phi }_0^2\sqrt{\mathrm{}^2\mu /2m}.$$
(21)
where $`\mathrm{\Phi }_0=2e/hc.`$ This quantity is very small. Indeed, comparing it with the components of $`n_{ij,kl}`$ producing contributions to Eq.(3) we see that $`w_4^{FS}/n_{xxxx}\alpha \left(\mathrm{\Delta }/\mu \right)^2.`$ Therefore in order to find an origin of 90 reorientation one has to look elsewhere.
The obvious possibility is to relax the assumption of the isotropic gap and turn to the case (ii). We calculated averages in Eq.(15) using the $`1/N`$ expansion rather than the BCS approximation. The Hamiltonian Eq.(13) becomes $`\psi _\alpha ^a\left[\epsilon \left(i\right)\mu \right]\psi _\alpha ^a\frac{\lambda }{4N}\psi _\alpha ^a\left[1+\delta \left(i\right)\right]\psi _\alpha ^a\psi _\alpha ^b\left[1+\delta \left(i\right)\right]\psi _\alpha ^b`$ where $`N`$ is number of (real or auxiliary) copies of the Fermi surface enumerated by $`a,`$ $`b`$. The corresponding perturbation Hamiltonian found by the minimal substitution is
$`H_1`$ $`=`$ $`i{\displaystyle _𝐱}A_z\left[{\displaystyle \frac{1}{2m}}\psi _\alpha ^a_z\psi _\alpha ^a+{\displaystyle \frac{\lambda \delta _0}{4N}}S_{}^{}U_{}cc\right],`$ (22)
$`S_{\alpha \beta }`$ $``$ $`\psi _\alpha ^a_z\psi _\beta ^a+\left(_z\psi _\alpha ^a\right)\psi _\beta ^a,`$ (23)
$`U_{\alpha \beta }`$ $``$ $`\psi _\alpha ^a\psi _\beta ^a+{\displaystyle \frac{\delta _0}{2}}\left[\psi _\alpha ^a_z^2\psi _\beta ^a+\left(_z^2\psi _\alpha ^a\right)\psi _\beta ^a\right].`$ (24)
Here the terms proportional to $`A_z^2`$ are omitted since they are local and cannot contribute to derivatives of $`K_{zz}(q)`$ with respect to $`q_x`$ required to obtain $`w_4`$. For simpler situations like the case (i) the leading order in $`1/N`$ expansion, with $`N`$ set to $`1`$, simply coincides with the BCS approximation. The reason to resort to the $`1/N`$ expansion is twofold. Firstly, the BCS expression for $`w_4`$ contains diagrams up to three loops (see Fig. 2c) which are very complicated. Secondly, unlike BCS, this nonperturbative scheme is systematically improvable. The last property is important when questions of principle are concerned. After observing that the order $`1/N`$ contributions, Fig. 2a, all vanish due to $`kk`$ asymmetry, we calculated the leading $`1/N^2`$ contributions to the magnetic kernel, Fig. 2b. At $`T=0`$ to leading order in $`\delta _0`$ (further reducing number of integrals) the result reads
$$w_4^{gap}=\frac{\delta }{N^2}\frac{8\pi }{105}\left(\frac{\mu }{\mathrm{\Delta }}\right)^2\mathrm{\Phi }_0^2\sqrt{\mathrm{}^2\mu /2m}$$
(25)
where $`\delta \delta _0m\mu `$ is dimensionless gap anisotropy. Therefore in physical case of interest $`N=1`$ we obtain $`w_4^{gap}/n_{xxxx}\delta `$ that is not necessary very small. This value is to be compared with $`w_4^{FS}`$ originated from Fermi surface anisotropy which has huge suppression factor $`(\mu /\mathrm{\Delta })^2.`$ A noticeable angular dependence of the gap was indeed observed in the most recent Raman scattering experiments on $`Y`$ and $`Lu`$ borocarbides .
To conclude, we found that the extended London model is incapable of explaining 90 reorientation in VL for $`H||a`$ because it produces an effective fourfold symmetry of the free energy of VL. This symmetry becomes explicit after a rescaling transformation. We showed that in general case one should include into the extended London model correction terms for which $`n_{ij,kl}n_{kl,ij}`$ (see Eq. (2)). As a result, the true twofold symmetry of the system in magnetic field $`H||a`$ is restored and 90 reorientation can be explained naturally. We demonstrated the two mechanisms that generate the correction terms: anisotropy of the Fermi surface and anisotropy of the superconducting gap, and showed that only the contribution of the latter one can lead to observable consequences. The investigation of vortex matter became recently a very sensitive tool to probe microscopic properties of the superconductors. In this paper we employed it to infer qualitative and even quantitative information about pairing interaction by calculating nonlocal corrections to linear response.
Note that inclusion of the correction terms will not change any conclusions of the extended London model for $`H||c.`$ On the other hand, nonzero ”two fold symmetric” correction will lead to smearing, or even disappearance, of lock-in transition in VL for $`H||a`$ . Most probably it will be not possible to check this prediction in the same samples of $`\mathrm{𝐿𝑢𝑁𝑖}_2B_2C`$ in which 90 reorientation was observed, because in this case the experimentally found opening angle of the unit cell of rhombic VL indicates that the critical field of lock-in transition is far above $`H_{c2}`$.
We are grateful to V.G. Kogan for bringing this problem to our attention and numerous illuminating discussions, to M.R. Eskildsen for discussions and correspondence, and also to R. Joynt and Sung-Ik Lee for valuable comments. The work is supported by NSC of Republic of China, grant $`\mathrm{\#}`$89-2112-M-009-039.
|
warning/0002/astro-ph0002374.html
|
ar5iv
|
text
|
# OBSERVATIONS OF THE CRAB NEBULA AND ITS PULSAR IN THE FAR-ULTRAVIOLET AND IN THE OPTICAL Based on observations with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract No. NAS5-26555. Based on observations obtained at the Nordic Optical Telescope on La Palma, using the Andalucia Focal Reducer and Spectrograph.
## 1. Introduction
The Crab nebula and its pulsar (PSR 0531+21) are among the most studied objects in the sky. The discovery of the Crab pulsar as a fast rotating radio pulsar (Staelin & Reifenstein (1968)Comella et al. (1969)) paved the way for the interpretation of pulsars as neutron stars (Gold (1968)). Also, the position of the Crab pulsar in the center of the Crab nebula, which is the remnant of supernova 1054, clearly supports the supernova - neutron star connection. Soon after the radio detection the pulsar was also shown to emit optical pulsations (Cocke, Disney, & Taylor 1969). This established that the pulsating star was the well known south preceding star in the center of the nebula, which early optical spectroscopy showed to emit a featureless continuum (Minkowski (1942)). To date, more than 1000 radio pulsars are known, but only the following few have optical counterparts known to pulsate also in visible light: the Crab pulsar (Cocke et al. (1969)), the LMC pulsar 0540-69 (Middleditch & Pennypacker (1985)), the Vela pulsar (Wallace et al. (1977)), PSR 0656+14 (Shearer et al. (1997)) and (possibly) the Geminga pulsar (Shearer et al. (1998)). In the near-UV (NUV), pulsations have only been established for the Crab pulsar (Percival et al. (1993)Gull et al. (1998), henceforth G98). Due to the faintness of these objects in the optical and in the ultraviolet, the spectroscopic information is very limited. PSR 0540-69 was observed with the Faint Object Spectrograph (FOS) onboard Hubble Space Telescope (HST) in the $`25005000`$ Å range (Hill et al. (1997)) and showed a rather steep power law spectrum. These observations were, however, contaminated by nebular emission. The Geminga pulsar was observed with the Keck telescope (Martin, Halpern, & Schiminovich 1998), but the spectrum has very low signal-to-noise because this pulsar is exceedingly faint in the optical. The only pulsar for which good signal-to-noise spectroscopy in the optical and ultraviolet can be obtained is the Crab pulsar. Surprisingly enough, very little has been done in this respect since the optical observations of Oke (1969). In particular, until the study by G98, no UV spectroscopy of the Crab pulsar had been published since the first attempts by the International Ultraviolet Explorer (IUE) (Benvenuti et al. (1980)). The IUE data cover only the NUV region ($`20003150`$ Å) and have poor signal-to-noise. The HST/STIS (Space Telescope Imaging Spectrograph) data from G98, and the new data presented here, clearly supersede these early attempts.
The Crab pulsar has been extensively studied over a very broad wavelength range, from the radio up to $`\gamma `$-rays (e.g., Lyne & Graham-Smith (1998)). The high energy emission, from infrared (IR) to $`\gamma `$-rays, is believed to be the result of the same emission mechanism (e.g., Lyne & Graham-Smith (1998)). It is therefore of interest to fill in the gaps in the observed spectrum of the pulsar in this range. Although the pulsar is relatively bright in the optical, UV observations are difficult due to the large extinction toward the Crab, $`E(BV)0.5`$ mag (e.g., Davidson & Fesen (1985), and references therein). Here, we present UV observations of the Crab pulsar further into the far-UV (FUV) ($`11401720`$ Å) than has previously been obtained. These are presented together with our previous NUV-data ($`16003200`$ Å)(G98) and new optical data from the Nordic Optical Telescope (NOT). Due to the large extinction correction, great care must be taken to draw conclusions about the intrinsic spectrum, and thus the emission mechanism of the pulsar. However, this procedure might also give a hint on the absorption properties of the dust in the direction toward the pulsar.
In addition to the pulsar emission, we detect emission lines from the Crab nebula itself in the FUV. In particular, the strength of the C IV $`\lambda `$1550 emission can be of interest for abundance determinations. Even more interesting is the broad C IV $`\lambda `$1550 absorption line from the nebula detected against the pulsar continuum. This line provides information on the nature of the SN 1054 event.
Although the Crab nebula has been studied extensively, the nature of the progenitor remains unknown. According to models based on the existence of the central neutron star, as well as on nucleosynthesis arguments, the zero-age main sequence (ZAMS) mass of the progenitor was probably in the range $`813M_{}`$ (Nomoto 1985). The amount of material observed in the nebula ($`4.6\pm 1.8M_{}`$) seems too low to account for this (Fesen, Shull, & Hurford 1997). Furthermore, the velocities of the filaments ($`1400\mathrm{km}\mathrm{s}^1`$, Davidson & Fesen 1985) give an uncomfortably low kinetic energy ($`1\times 10^{50}`$ ergs) compared to other supernova remnants, i.e., at least an order of magnitude less than the canonical energy of supernovae, $`10^{51}`$ ergs.
The “missing mass” could either be in a slow progenitor wind, or in a fast, hitherto undetected, shell ejected at the explosion (Chevalier 1977). If the latter is true, as is hinted by the observations of the outer \[O III\] skin of the nebula (Hester et al. 1996; Sankrit & Hester 1997), this shell might account for the missing mass and kinetic energy of the nebula. The question remains, however, why such a shell has escaped detection despite many efforts to observe it (see, e.g., Fesen et al. (1997)). One possibility is that the low density of the surrounding gas is not high enough to give rise to detectable circumstellar emission when interacting with the ejecta; neither X-ray nor radio searches have indicated any evidence of circumstellar interaction between fast ejecta and ambient gas (Mauche & Gorenstein 1989; Frail et al. 1995). If a fast shell is absent, the birth of the Crab was definitely a low energy event. This would call for a revision of our understanding of supernova explosions, especially since SN 1054 apparently was not unusually dim according to historical records (Chevalier 1977; Wheeler 1978).
It is thus of great interest to further investigate whether there is a stellar wind or supernova ejecta outside the observed nebula, and what velocity this gas may have. Lundqvist, Fransson, & Chevalier (1986, henceforth LFC86) proposed to search for a fast shell by looking in the UV toward the Crab pulsar. Their time dependent photoionization calculations showed that C IV $`\lambda `$1550 could show up in blueshifted absorption if the ionization history of the shell was as predicted in some models of Reynolds & Chevalier (1984). Here, we present the detection of this broad absorption line, and discuss its implications for the fast shell around the Crab nebula.
First we discuss the observations and reductions (§2). We then (§3) discuss the pulse profile, the amount of reddening toward the pulsar and the intrinsic pulsar spectrum in the optical/UV. In §3 we also discuss the lines originating from the interstellar gas toward the pulsar and from the Crab nebula itself. Some of these observations constrain the properties of a possible outer shell. In §4 we summarize our conclusions.
## 2. Observations, Reductions and Results
### 2.1. HST Far-UV Observations
The Crab pulsar was observed on January 22, 1999 using HST/STIS (Kimble et al. (1998)) with the FUV Multi Anode Micro-channel Array (MAMA) detector. The low resolution grating G140L was used, which covers the wavelength interval $`11401720`$ Å. These observations were made in the time-tag mode and used a slit of $`52\mathrm{}\times 0\stackrel{}{\mathrm{.}}5`$. The spectral resolution is 0.58 Å pixel<sup>-1</sup>, and the plate scale is $`0\stackrel{}{\mathrm{.}}0244`$ pixel<sup>-1</sup>. This means that only 25″ of the long-slit is actually projected onto the detector. In total, six orbits of observations, including target acquisition, were used. These were divided into two visits. A log of the observations is shown in Table 1. The total on-target exposure time amounted to 14,040 seconds. The orientation of the slit is shown in Figure 1.
#### 2.1.1 Time-resolved emission
The time-tag mode on the STIS allows us to resolve the emission from the Crab pulsar both in wavelength and time. The time resolution obtained in this mode is 125 $`\mu `$s. As these observations were the first to utilize the time-tag capabilities of HST/STIS in the FUV for a known periodic variable, special software, developed at Goddard Space Flight Center (GSFC) was used to obtain the pulse profile for the pulsar. The analysis followed the procedures outlined in G98. For each of the six datasets, a time averaged image was produced to trace the position of the pulsar spectrum. A 13 pixel wide window was used to extract events in the pulsar spectrum as well as in the background emission at both sides of the pulsar emission. The arrival time of each 125 $`\mu `$s sample was converted to a solar system barycenter arrival time. The position of the HST with respect to earth center was computed by the Flight Dynamics Facility at GSFC with errors less than 200 meters. The position of the earth with respect to the solar system barycenter was computed using a routine, SOLSYS, supplied by the U. S. Naval Observatory (Kaplan et al. 1989). All events in the pulsar spectrum (and in the background regions) were assigned a (barycentric) arrival time with respect to the start of the first exposure. As we found a small drift in the times recorded in the FITS headers, we used the internal clock from the engineering mode header to do this. To determine the period we folded the arrival times modulo a grid of test frequencies, $`f_i`$, and corrected for the slowdown rate of the pulsar. Pulse profiles were calculated as histograms of the function $`f(t)=f_it+\dot{f}t^2/2`$, where the data were coadded into 512 phase bins. The appropriate value of $`f_i`$ was then determined by maximizing the sum of squares of the values in the pulse profile.
In this procedure we used a value for $`\dot{f}`$ from radio observations at Jodrell bank (Lyne, Pritchard, & Roberts 1999), while the second time derivative is unimportant for this purpose. This resulted in a measured period of P=33.492675 ms at Modified Julian Date (MJD) 51200.549, the time of the beginning of our first observation (number O4ZP01010). As the data were obtained during a time period of eight hours, we could also determine the pulsar slowdown rate from our observations. We obtained $`\dot{P}=(4.0\pm 0.4)\times 10^{13}`$ s s<sup>-1</sup>, which is consistent with the value used above from the radio observations, $`4.2\times 10^{13}`$ s s<sup>-1</sup>.
In Figure 2 we show the Crab pulsar pulse profile in the FUV regime. It was obtained by subtracting the background from the pulse profile obtained for the period given above. For comparison, the figure also shows the NUV pulse profile from G98.
#### 2.1.2 Phase-averaged spectrum
Averaging over the pulse we obtain a phase-averaged spectrum of the pulsar which covers the region $`11401720`$ Å. This spectrum probes the emission of the Crab pulsar further into the UV than has previously been done and is shown in Figure 3, together with the NUV spectrum of G98. These spectra overlap nicely and cover together the whole range from $`11403200`$ Å. The combined spectrum offers the possibility to deduce the amount and characteristics of the interstellar reddening, as well as to determine the dereddened pulsar spectrum itself.
The FUV-spectrum was extracted with a 13 pixel wide window. The reductions were made using the CALSTIS software developed at the GSFC. These IDL-routines flatfield the images and then the point source spectrum is localized and traced on the detector. The extracted pulsar spectrum is background subtracted and converted to absolute flux units using the G140L sensitivity table. The accuracy of the absolute flux calibration is $`15\%`$ over the full wavelength scale. Wavelengths are assigned from a library dispersion solution, while the zero point adjustments are determined from arc frames taken through the $`0\stackrel{}{\mathrm{.}}05`$ slit for each science observation. The wavelengths are then converted to heliocentric wavelengths. The accuracy of the wavelength solution is about 0.4 Å.
### 2.2. Optical observations
In addition to the UV spectrum, we have also collected data in the optical regime. During several nights in December 1998 we did spectroscopy of the Crab pulsar using the Andalucia Focal Reducer and Spectrograph (ALFOSC) at the 2.56m NOT on La Palma. In total we obtained 11.25 hours of data in five different grisms. The 1$`\stackrel{}{\mathrm{.}}`$2 slit was used for all observations (see Table 2 for more details). Not all nights were photometric and the seeing was generally just above 1″. The data were bias subtracted and flatfielded. Wavelength calibrations were done using arc frames obtained with a helium lamp. Flux calibration of the spectra was accomplished by comparison to the spectrophotometric standard stars Feige 34 and G191-B2B. To avoid systematic errors due to background subtraction the slit was put at two different position angles. All observations were made at low air masses and close to the parallactic angle to reduce the effects of atmospheric dispersion; the Crab pulsar passes just $`7\mathrm{°}`$ from zenith as viewed from La Palma in December. The slit positions for the NOT observations are shown in Figure 1.
### 2.3. Combined UV/optical spectrum
The combined optical and UV observations (both NUV and FUV) are shown in Figure 4. It covers the region $`11409250`$ Å. Although great care was given to the background subtraction of the nebula, the optical spectrum of the Crab pulsar was contaminated by over- and under-subtractions of strong nebular emission lines. These had to be taken out by hand, and we used the IRAF task SPLOT to interactively clean the spectrum. Points that deviated more than 4$`\sigma `$ from a smooth continuum fit were rejected. This procedure was robust enough to exclude only clear cases of nebular contamination. The optical spectrum used is a combination of all the different spectra in Table 2. Note that an absolute flux calibration was not applied to the optical spectrum, as the observing conditions were often non-photometric. Instead we have applied a grey shift to the spectrum to match the $`V`$-band observations of Percival et al. (1993) and Nasuti et al. (1996). As seen in Figure 4, this is well matched with the UV spectrum from the HST.
## 3. Discussion
### 3.1. Pulse profile
The work of Percival et al. (1993) showed small differences in the optical versus NUV pulse profile shapes. They observed the Crab pulsar with the High Speed Photometer (HSP) onboard HST and found that the main pulse is slightly narrower in the UV than in the optical. Eikenberry et al. (1997) extended the analysis into the near-IR and found that the trend for a decreasing Full Width Half Maximum (FWHM) with decreasing wavelength seems to hold over the full UV-IR range.
The time-tag mode of STIS/FUV-MAMA allows us to examine if the pulse profile of the emission is different in this wavelength region from that in the NUV. The most striking feature of the FUV pulse profile shown in Figure 2 is certainly that it is very similar to the profile previously obtained in the NUV (G98). It appears, however, that the primary peak is slightly narrower in the FUV than in the NUV, as indicated by the blow-up of that region in Figure 2. We measured the FWHM of the FUV and NUV primary peak to be 0.0405 and 0.0426 periods, respectively. The position of the peak was determined by a polynomial fit to the central 10 phase bins. The phase at half-maximum was then simply determined by linear interpolation between the two closest phase bins. To estimate a statistical error on the procedure used to measure the FWHM, we computed the FWHM for each of our six FUV observations. The standard deviation obtained in this procedure was 0.001 phase bins. The measured difference in the FWHM of the primary peak can therefore be considered marginally significant. The secondary peak in Figure 2 might actually appear broader in the FUV. It is, however, much noisier than the primary peak and the same procedure as above could not determine any significant difference to the 13$`\%`$ (3$`\sigma `$) level. The above findings are in agreement with the trend seen in Percival et al. (1993) and Eikenberry et al. (1997). The pulse period obtained from the FUV observations is P=33.492675 ms. Radio data from Jodrell Bank (Lyne et al. (1999)) determined the pulse period to be P=33.492402 ms on January 15 1999. Using their values for the pulse period and its first time derivative on this date we can calculate the period at the time of our HST observations. The result is P=33.492676 ms. This agrees with our estimate to 7 significant digits. The limiting errors in our computation of the period are the accuracy of the SOLSYS routine which has a barycentric velocity error of less than $`8.0\times 10^7`$ AU/day and the unknown accuracy of the rate of the STIS onboard clock.
### 3.2. The UV extinction curve
The phase averaged UV spectrum of the Crab pulsar must be corrected for a substantial amount of interstellar reddening. The value for $`E(BV)`$ has been estimated by a number of authors; Wu (1981) obtained $`E(BV)=0.50\pm 0.03`$ mag by using the 2200 Å dust absorption feature, the nebular synchrotron continuum and an extinction curve derived from eight stars. Blair et al. (1992) also used the best fit to the UV nebular continuum and obtained $`E(BV)=0.51_{0.03}^{+0.04}`$ mag. A different method was used by Miller (1973), who determined the reddening of the Crab nebula from observations of \[S II\] lines. Using modern values for the atomic parameters (Keenan et al. 1993; Ramsbottom, Bell, & Stafford 1996), and the extinction curve of Fitzpatrick (1999), his measurements gives $`E(BV)=0.50_{0.06}^{+0.04}`$ mag.
Our data of the pulsar itself allows us to estimate the value for $`E(BV)`$ by “ironing out” the 2200 Å bump. To do this we assumed a standard value $`R=3.1`$ and dereddened the UV spectrum for different values of $`E(BV)`$. The dereddened spectra were then fitted by a power law and we chose the value for $`E(BV)`$ that minimized $`\sigma _{\alpha _\nu }`$, the standard deviation of the power law fit in the region log $`\nu `$=\[14.98, 15.41\], where the region including the Ly$`\alpha `$ absorption was excluded. This procedure used the galactic mean extinction curve from Fitzpatrick (1999) and gave $`E(BV)=0.52`$ mag. This is in excellent agreement with the previous results stated above. As our data have better signal-to-noise and sampling than previous continuum fits, we will use this value as the best estimate of the extinction toward the Crab nebula throughout this paper.
From the NUV spectrum taken of the Crab pulsar by IUE, claims were made for a peculiar extinction curve (Benvenuti et al. (1980)). In particular, the 2200 Å bump was reported to be substantially narrower than for the galactic mean extinction curve, a finding that could indicate that the supernova event itself had altered the grain composition in the Crab nebula. In principle both the extinction curve and the intrinsic pulsar spectrum are unknown, which of course makes it troublesome to disentangle these quantities. We take the following approach to this problem: theoretical models favor a power law spectrum (e.g., Lyne & Graham-Smith (1998)) and dereddening with $`E(BV)=0.52`$, $`R=3.1`$ indeed gives a power law pulsar spectrum (Fig. 4), so we will simply assume the intrinsic spectrum of the Crab pulsar to follow a power law $`F_\nu `$ = $`K(\nu /\nu _0`$)$`^{\alpha _\nu }`$ ergs s<sup>-1</sup> cm<sup>-2</sup> Hz<sup>-1</sup>. Here $`\nu `$ is the frequency of the radiation and $`K`$ is a constant that is nearly independent of $`\alpha _\nu `$ when $`\nu _0`$ is the logarithmic mean frequency of the fitted bandpass (Percival et al. (1993)). Using the extinction parameters above, $`E(BV)=0.52`$ and $`R=3.1`$, the spectral index is $`\alpha _\nu =0.035`$ in the UV range.
By assuming that the intrinsic pulsar spectrum is indeed well represented by the obtained power law we derived an extinction curve toward the pulsar. This is shown in Figure 7 together with the mean galactic extinction curve for $`R=3.1`$ from Fitzpatrick (1999). The derived extinction curve has a moderately narrower dip (15$`\%`$) and a somewhat shallower rise in the extreme FUV. Apart from this, it overlaps nicely with the galactic mean extinction curve. Considering the large variety of measured UV extinction curves (see Fig. 2 in Fitzpatrick 1999), the derived extinction curve toward the Crab can hardly be claimed to be peculiar. Assuming that the pulsar spectrum follows a power law, we conclude that we find no evidence for a non-standard extinction curve toward the Crab pulsar.
### 3.3. The spectral index of the pulsar continuum
According to models, the high energy pulsar emission, from IR to $`\gamma `$-rays, is produced by (curvature) synchrotron radiation (e.g., Lyne & Graham-Smith (1998), and references therein). A number distribution of electrons following a power law $`N(E)dE=CE^\gamma dE`$, where $`E`$ is the energy, $`C`$ is a constant and $`\gamma `$ is the electron spectral index, will produce synchrotron radiation that has a power-law distribution in flux density $`F_\nu =K(\nu /\nu _0)^{\alpha _\nu }`$ ergs s<sup>-1</sup> cm<sup>-2</sup> Hz<sup>-1</sup>. The photon spectral index, $`\alpha _\nu `$, is related to the electron spectral index, $`\gamma `$, via $`\alpha _\nu =(\gamma 1)/2`$ for synchrotron radiation.
The early low resolution spectroscopy of Oke (1969) appears to peak in the middle of the observed region ($`34008000`$ Å). He reports a slope of $`\alpha _\nu =0.2`$ with no stated errors, although he cautions that the uncertainty in the reddening correction could allow also for a positive slope. Much of the theoretical work on the optical emission mechanism of pulsars has been based on this finding (cf. Ginzburg & Zheleznyakov 1975; Lyne & Graham-Smith (1998)). More recently, Percival et al. (1993) used ground-based optical broadband photometry together with a NUV photometric point from HST/HSP to determine a slope of $`\alpha _\nu =0.11\pm 0.13`$, while Nasuti et al. (1996) obtained a spectrum in the limited wavelength range $`49007000`$ Å (see below) and determined $`\alpha _\nu =0.10\pm 0.01`$.
Applying $`R=3.1`$ and $`E(BV)=0.52`$ from §3.2 to our UV spectrum gives a spectral index of $`\alpha _\nu =0.035\pm 0.040`$, where the error is simply the RMS around the fit. Obviously, the main uncertainty in this procedure is the extinction parameters. For $`E(BV)`$ in the range \[0.48, 0.55\], the uncertainty due to reddening becomes $`\alpha _\nu =0.035_{0.13}^{+0.094}`$. An accurate estimate of the spectral index obtained in the UV, for the given extinction range, can be obtained by $`\alpha _\nu =0.035+3.12[E(BV)0.52]`$. Note that the spectral index for $`E(BV)=0.55`$ was erroneously reported in G98. The analysis in this paper supersedes this previous report.
Including the optical spectrum from the NOT gives a wider wavelength range for the fit. The dereddened spectrum shown in Figure 4 was obtained with $`R=3.1`$ and $`E(BV)=0.52`$. The best power law fit to the complete spectrum (excluding Ly$`\alpha `$) gives $`\alpha _\nu =0.11`$.
Using the full wavelength range we have also tried to constrain both $`E(BV)`$ and $`R`$. This can be done using the extinction curves from Fitzpatrick (1999), which are a one-parameter family in $`R`$. By assuming an intrinsic power law we thus allowed both $`R`$ and $`E(BV)`$ to vary, and chose the values that minimized $`\sigma _{\alpha _\nu }`$, the standard deviation of the power law fit. This procedure gives $`R=3.0`$ and $`E(BV)=0.51`$, which is consistent with the values used above. In Figure 6 we show the dereddened spectrum of the Crab pulsar for several different values of $`R`$ and $`E(BV)`$. In this figure we have also included IR data from Eikenberry et al. (1997). These plots clearly show the ambiguity of the reddening correction.
We can express the spectral index of the power law fit for the values $`E(BV)=0.52`$ and $`R=3.1`$ as $`\alpha _\nu =0.11\pm 0.04_{0.22}^{+0.21}`$. The first error represents the RMS around the power law fit, and the last errors include all power law fits in the extinction intervals $`R=[2.9,3.3]`$, $`E(BV)=[0.48,0.55]`$. A linear fit, $`\alpha _\nu =0.110.38(R3.1)+3.88[E(BV)0.52]`$ reproduces the obtained $`\alpha _\nu `$ in this interval to better than 0.02 units.
We can only echo Oke (1969) in his conclusion that the uncertainties in reddening corrections are large enough to allow both for negative and positive slopes. The inferred energy distribution for the electrons is given by $`\gamma =0.8\pm 0.5`$. It is worth noting that the only other young pulsar for which a spectrum has been obtained in the optical, PSR B0540-69, had $`\alpha _\nu =1.6\pm 0.4`$ (Hill et al. (1997)). This is clearly steeper than the Crab pulsar spectrum.
Finally, the time tag mode of our FUV observations also allows us to extract phase-resolved spectra, and we have looked for spectral differences during the pulse phase. We found no significant differences above the $`5\%`$ level, neither in the spectrum of the primary versus the secondary peak, nor in the leading versus trailing part of the primary peak.
### 3.4. The $`5900`$ Å feature
In the first report on the optical spectrum of the Crab pulsar, Minkowski (1942) reported a featureless continuum with no absorption or emission lines. The observations of Oke (1969) indicated a similar conclusion, although the spectral resolution was not very good. Since then, few attempts have been made to obtain a better optical spectrum of the Crab pulsar, despite that technical development certainly admits improvements. The only modern optical spectrum of the Crab pulsar was obtained with the New Technology Telescope (NTT) by Nasuti et al. (1996). This phase-averaged spectrum covered a rather small wavelength region ($`49007000`$ Å) and was flat with $`\alpha _\nu =0.10\pm 0.01`$ if dereddened with $`E(BV)=0.51`$. The spectrum was also reported to show a large dip at $`5900`$ Å, the width being $`100`$ Å. According to the authors this feature probably originates close to the pulsar itself, but no detailed physical mechanism was proposed.
We have searched for this feature in our optical spectra, but found no sign for a dip. This is true for all the different gratings covering this region during all of the nights of data obtained in December 1998 (see Fig. 4). Neither do we see the feature in unpublished data taken by us at NOT one year earlier (December 1997). Nasuti et al. (1996) noted that the feature became enhanced after flux calibration. This suggests that the dip may be an artefact of the reduction procedure. While their investigation used a single spectrophotometric standard star with flux sampled only every 100 Å, we have used two standard stars with 2 Å sampling in the relevant wavelength region. Although the feature could be time dependent, we propose to regard it as an artefact until confirmed by other observations.
### 3.5. Absorption lines
As in the NUV spectrum of the pulsar (G98), the FUV spectrum contains several absorption lines. The identified lines are shown in Figures 7 and 8, and their equivalent widths (EWs) and corresponding column densities, $`N`$, are listed in Table 3. For completeness, we have also included the lines in the NUV spectrum identified and measured by G98.
The strongest line is Ly$`\alpha `$, and its damping wings can be used to estimate $`N`$(H I). To do this we assume that the optical depth in the damping wings is defined by $`\sigma (\lambda )N`$(H I) (Shull & Van Steenberg, 1985). Here the absorption cross section $`\sigma (\lambda )=(4.26\times 10^{20}\mathrm{cm}^2)/(\lambda \lambda _0)^2`$, where the wavelengths are in Å and $`\lambda _0=1215.67`$ Å. We have not considered the instrumental profile, which is much narrower ($`1`$ Å) than the line width (see Fig. 7). From this analysis we obtain $`N`$(H I)$`=(3.0\pm 0.5)\times 10^{21}`$ cm<sup>-2</sup>. The spectrum corrected for Ly$`\alpha `$ absorption is shown in Figure 7. Our estimated uncertainty in $`N`$(H I) reflects the uncertainty in the continuum fit of the corrected spectrum. The column density of free electrons toward the Crab is measured from pulsar dispersion to be $`N_\mathrm{e}`$ $`=(0.1755\pm 0.0007)\times 10^{21}`$ cm<sup>-2</sup> (Comella et al. (1969)). The amount of atomic hydrogen is thus $`N`$(H)$`=(3.2\pm 0.5)\times 10^{21}`$ cm<sup>-2</sup>. This value agrees well with both the result of Schattenburg & Canizares (1986) who obtained $`N`$(H) $`=(3.45\pm 0.42)\times 10^{21}`$ cm<sup>-2</sup> from an estimate of the X-ray absorption toward the Crab, and the value we estimate from $`E(BV)`$ using the relation in de Boer, Jura, & Shull (1987), which gives $`N`$(H I)$`3.0\times 10^{21}`$ cm<sup>-2</sup> for $`E(BV)=0.52`$.
The other absorption lines we identify around zero velocity in the FUV spectrum are: C I $`\lambda \lambda `$ 1277,1329,1561,1658, C II $`\lambda `$1335, O I $`\lambda `$1303, Al II $`\lambda `$1671, Si II $`\lambda \lambda `$ 1260,1527 and Si IV $`\lambda `$1394. The Si IV line is close to noise level, which explains why the Si IV $`\lambda `$1403 component is not seen. The C I $`\lambda \lambda `$ 1277,1329 lines are also marginal detections, but their strengths correspond to those expected when compared with the strengths of C I $`\lambda \lambda `$ 1561,1658. The C IV $`\lambda \lambda `$ 1548,1551 doublet is absent at zero velocity, but shows a blueshifted absorption with a maximum shift of $`2500\mathrm{km}\mathrm{s}^1`$. We return to this line in §3.7.
All lines except Ly$`\alpha `$, and the C IV doublet which is not interstellar or from a slow wind (§3.7), are unresolved by the moderate spectral resolution in the STIS spectra ($`1.2`$ Å in FUV and $`3.2`$ Å in NUV, corresponding to $`250\mathrm{km}\mathrm{s}^1`$ and $`400\mathrm{km}\mathrm{s}^1`$ for the central wavelengths of the two gratings, respectively). The measured EWs of the absorption lines are rather large, which means that we cannot assume that the lines are resolved and optically thin (i.e., we cannot use “weak-line” theory \[e.g., Morton 1991\]) to derive abundances of the various species, as this would severely underestimate the abundances.
In the absence of a proper model for the distribution of intervening matter, we assume that the absorption is dominated by a single cloud component. We set the intrinsic “Doppler” width in this component to $`1\mathrm{km}\mathrm{s}^1`$, and assume that this is the same for all lines. Furthermore, we consider only one spectral component of the lines listed in Table 3. All these assumptions cause us to systematically overestimate the column densities and abundances so that our estimates will be upper limits to these. With this in mind, we calculate Voigt line profiles and EWs as functions of column density, using the atomic data in Morton (1991). From the measured EWs we then obtain column densities and abundances for the interstellar gas toward the Crab. The abundances in Table 3 are presented in the standard logarithmic form where the value for hydrogen is set to 12.0.
In Table 4, where we have simply coadded the abundances of the different ionization stages of each element, we give the abundances of C, O, Mg, Al, Si and Fe from our analysis. We also list the solar values for these elements according to Anders & Grevesse (1989). There appears to be no extreme depletion of any element, contrary to what was stated in G98, where “weak-line” theory was used. This is in agreement with the X-ray observations by Schattenburg & Canizares (1986) which were consistent with a solar abundance of oxygen. Our oxygen abundance from the single O I $`\lambda `$1303 line is, however, rather uncertain due to geocoronal airglow corrections, and a possible blending with Si II $`\lambda `$1304. As pointed out above, our method is likely to systematically overestimate abundances. Broader lines, or more cloud components, would lower the abundances derived in accordance with depletion seen in the normal interstellar medium (e.g., 0.65 dex for carbon at 2 kpc, Welty et al. 1999). Spectra with higher resolution are required to refine this analysis.
### 3.6. Emission lines
The FUV-spectrum contains also information about the Crab nebula. The long slit covers $``$ 25″$`\times `$ 0$`\stackrel{}{\mathrm{.}}`$5 of the nebula, along position angle (PA) $`40\mathrm{°}.5`$. The orientation of the slit is shown in Figure 1. To extract the pulsar spectrum we used only 13 of the 1024 pixels of the MAMA detector. To obtain a spectrum of the nebular emission outside this extraction window we again used CALSTIS to produce rectified, wavelength- and flux-calibrated images. As the count rates were very low, we summed the emission from two 8$`\stackrel{}{\mathrm{.}}`$2 long regions positioned 0$`\stackrel{}{\mathrm{.}}`$66 above and below the pulsar position. This safely excludes any contribution from the point spread function (PSF) wings of the pulsar. All emission from all 6 observations was combined and averaged to increase the signal-to-noise. We present this nebular spectrum in Figure 9, which shows only the wavelength region $`14001700`$ Å. This is to exclude the geocoronal lines of Ly$`\alpha `$ and O I $`\lambda \lambda `$ 1303,1356. The spectrum is the average of the regions above and below the pulsar. For the spectrum extracted above (North-East of) the pulsar, the continuum level was flattened artificially. This is because we observed a rising continuum in the red part of this spectrum, which is probably due to contamination of the nearby star seen in Figure 1. Moreover, the count rates in the continuum are only $`7.5\times 10^5`$ counts pixel<sup>-1</sup> s<sup>-1</sup>. At such low count rates, trends in the dark currents might influence the continuum. No dark images are subtracted from the FUV-MAMA observations, because the dark images are known to be variable and to have low count statistics. Therefore, we will not discuss further the slope of the continuum emission from the nebula.
Only two emission lines intrinsic to the Crab nebula can be seen, C IV $`\lambda `$1550 and He II $`\lambda `$1640. The measured intensities are $`(2.1\pm 0.8)\times 10^{15}`$ and $`(1.4\pm 0.8)\times 10^{15}`$ ergs s<sup>-1</sup> cm<sup>-2</sup>, respectively. This is an average for the two 8$`\stackrel{}{\mathrm{.}}`$2 $`\times `$ 0$`\stackrel{}{\mathrm{.}}`$5 regions, just above and below the pulsar. Our detection of these two lines is in accordance with the two previous spectroscopic observations of the Crab nebula in the FUV. Davidson et al. (1982) used the IUE to detect these two lines, as well as C III\] $`\lambda 1909`$. Blair et al. (1992) recalibrated the IUE data and complemented these with observations from the Hopkins Ultraviolet Telescope (HUT).
The widths of the lines are approximately 13 Å and 10 Å, for C IV $`\lambda `$1550 and He II $`\lambda `$1640, respectively. The width is due to the velocity distribution of the material, the doublet nature of C IV $`\lambda `$1550 as well as to the spectral resolution for extended sources for the spectrograph. The bulk of the emission is redshifted by $`1000\mathrm{km}\mathrm{s}^1`$, although a narrower zero-velocity component is seen from the region above the pulsar.
The flux ratio of C IV $`\lambda `$1550 to He II $`\lambda `$1640 is thus $`1.5`$, which is in accordance with the findings of Davidson et al. (1982), who sampled the fluxes over a larger field, $`20\mathrm{}\times 10\mathrm{}`$. They used this to argue that the Crab nebula has no overabundance of carbon. However, the subsequent investigation by Blair et al. (1992) showed that variations in the He II/C IV line ratio exist in the nebula. They cautioned on conclusions regarding the carbon abundance, as differences in physical parameters (e.g., ionization, density, temperature, clumping) are difficult to disentangle from abundance variations. To determine the carbon abundance is important because it holds the potential to provide information about the ZAMS mass of the progenitor. Comparisons to detailed photoionization models for the large apertures used by IUE and HUT are hampered by the fact that ionization conditions are known to change over very small spatial scales (Sankrit et al. (1998)). The observations we present have finer spatial resolution and are not biased toward bright filaments. HST observations of the same regions in the optical, to establish the ionization conditions, could make more quantitative estimates of the carbon abundance possible.
### 3.7. The outer shell
As was mentioned in §3.5, and shown in Figure 8, C IV $`\lambda `$1550 is the only line in the UV spectrum which shows clear evidence of blueshifted absorption. No evidence for absorption can be seen at zero velocity. The greatest absorption appears to arise in material moving at $`1200\mathrm{km}\mathrm{s}^1`$, and can thus be due to “normal” Crab nebula material. However, the absorption seems to continue to higher velocities, and can be traced out to $`2500\mathrm{km}\mathrm{s}^1`$ (see Fig. 10). To estimate the significance of the detection we have calculated an average and RMS from two 50 Å regions redward and blueward of the C IV line. Although the line is too noisy for individual pixels to be significant, it consists of many pixels that are consistently below the average. The eleven pixels between $`16502780\mathrm{km}\mathrm{s}^1`$ give a detected absorption with a significance of $`5.8\sigma `$. For the five points in the region $`23302780\mathrm{km}\mathrm{s}^1`$ the significance is $`3.0\sigma `$.
Together with Clark et al. (1983), who also reported velocities in excess of $`2000\mathrm{km}\mathrm{s}^1`$, this is the highest velocity ever measured in the Crab, and can be interpreted as evidence for the existence of the long sought fast outer shell (cf. §1). The column density of C IV in Table 4, $`N(`$C IV$`)=(3.0\pm 1.1)\times 10^{14}`$ cm<sup>-2</sup>, assumes the line to be resolved and optically thin. This should be the case for velocity broadening caused by the tentative fast shell, though we caution that the material moving at $`1200\mathrm{km}\mathrm{s}^1`$ may not be spectrally resolved. The total column density of C IV is therefore likely to be higher, while that of the proposed fast shell is lower than the value given in Table 4. We will now investigate whether the detection is consistent with a fast outer shell model and what information can be provided about the supernova ejecta.
#### 3.7.1 Constraints from C IV $`\lambda `$1550
We adopt a model similar to that of LFC86, i.e., outside the observed filaments we attach a massive, freely coasting, spherically symmetric shell. The inner radius, $`R_{\mathrm{in}}`$, is set to $`5.0\times 10^{18}`$ cm, which agrees with the “mean” inner radius of the presumed shell used by Sankrit & Hester (1997, see their Fig. 7). For free expansion this corresponds to a velocity of $`1680\mathrm{km}\mathrm{s}^1`$. Furthermore, we assume that the mass of the shell, $`M_{\mathrm{sh}}`$, is $`4M_{}`$, and that the outer radius of the shell, $`R_{\mathrm{out}}`$, is $`1.9\times 10^{19}`$ cm. The maximum velocity, $`V_{\mathrm{out}}`$, is then $`6370\mathrm{km}\mathrm{s}^1`$, and the total kinetic energy of the shell, $`E_{\mathrm{sh}}`$, is $`1.0\times 10^{51}`$ ergs, if the density is constant in the shell. For a density which decreases with radius, $`E_{\mathrm{sh}}`$ is lower, if $`M_{\mathrm{sh}}`$ is held constant. This is shown in Table 5 for density slopes up to $`\eta =9`$, where $`\eta `$ is defined as $`\rho (R)=\rho (R_{\mathrm{in}})(R/R_{\mathrm{in}})^\eta `$.
In our model we have used the relative abundances $`X(`$H$`)`$: $`X(`$He$`)`$: $`X(`$C$`)`$ = 1.0: 0.1: $`3.5\times 10^4`$, where the $`X(`$C$`)`$/$`X(`$H$`)`$ ratio corresponds to the solar value. With these abundances we have calculated the absorption in C IV $`\lambda \lambda `$ 1548,1551 as a function of $`\eta `$ and the relative fraction of carbon in C IV, $`X(`$C IV$`)`$. The parameter $`X(`$C IV$`)`$ is unity when carbon is all in C IV.
In Figure 10 we show results for two models where $`X(`$C IV$`)`$ has been kept constant throughout the shell. The dotted line shows the C IV line in the case of $`\eta =0`$ and $`X(`$C IV$`)=1`$, while the dashed line is for $`\eta =3`$ and $`X(`$C IV$`)=0.14`$. Both models are described in Table 5. The photoionization models of LFC86 show a case similar to our $`\eta =0`$ case (their model 1). There, carbon is ionized beyond C IV, and only a very small fraction, concentrated to the outer edge of the shell, remains in C IV. This gives maximum absorption in the $`\eta =0`$ case around $`6000\mathrm{km}\mathrm{s}^1`$ (see Fig. 1 of LFC86 for such a case). This is clearly not what is seen in the FUV data. On the contrary, the observed absorption peaks at lower velocities, meaning that the C IV number density must be the highest closer to $`R_{\mathrm{in}}`$. The photoionization models of LFC86 make the $`\eta =0`$ case rather unlikely from this point of view. Instead we turn our attention to steeper density profiles.
Such a situation is highlighted by the $`\eta =3`$ case in Figure 10. Our assumption of constant $`X(`$C IV$`)`$ in the shell is probably more realistic for $`\eta =3`$ than for $`\eta =0`$. This is because the ionization parameter, $`\xi =\mathrm{n}_\gamma /\mathrm{n}_\mathrm{e}`$, where $`\mathrm{n}_\gamma `$ and $`\mathrm{n}_\mathrm{e}`$ are number densities of ionizing photons and electrons, respectively, has the radial dependence $`\xi R^{\eta 2}`$, if absorption can be neglected. With absorption included, $`\eta `$ must be $`>2`$ to obtain a near-constant $`\xi (R)`$. Our choice of $`X(`$C IV$`)=0.14`$ has only been made to fit the data. Any combination of $`X(`$C IV$`)`$ and $`M_{\mathrm{sh}}`$, so that $`X(`$C IV$`)M_{\mathrm{sh}}0.56M_{}`$, would fit the data equally well. In our model, for $`\eta =3`$, this gives a minimum mass (for $`X(`$C IV$`)=1`$) of the shell of $`M_{\mathrm{sh}}0.6M_{}`$, and a minimum kinetic energy of $`E_{\mathrm{sh}}8\times 10^{49}`$ ergs.
While a constant $`X(`$C IV$`)`$ in the shell gives a good fit to the line profile for $`\eta =3`$, cases with a steeper density dependence must have an ionization structure where $`X(`$C IV$`)`$ increases outward through the shell. From the radial dependence on the ionization parameter this could be possible, but may require a fine-tuning between the $`R^{\eta 2}`$ and exp$`(\tau )`$ parts of $`\xi `$. Time dependent effects are also likely to be important, and the likelihood of an increasing $`X(`$C IV$`)`$ with radius can only be explored by numerical models. We postpone this to a future study. To give the same absorption at $`2500\mathrm{km}\mathrm{s}^1`$ as the $`\eta =3`$ model in Figure 10, models with other $`\eta `$ must have $`X(`$C IV$`)1.49^{\eta 3}X(`$C IV$`)_{\mathrm{in}}`$. The parameter $`X(`$C IV$`)_{\mathrm{in}}`$ is the value of $`X(`$C IV$`)`$ at $`R_{\mathrm{in}}`$, and is the value given in Table 5. For example, for $`\eta =9`$, $`X(`$C IV$`)`$ must be $`11`$ times higher at the radius of $`2500\mathrm{km}\mathrm{s}^1`$ compared to $`X(`$C IV$`)_{\mathrm{in}}`$.
It thus appears as if the observed line profile of the C IV $`\lambda 1550`$ absorption is consistent with the outer shell being fast and massive. That is, the C IV line cannot exclude that the shell could carry an energy of $`10^{51}`$ ergs, because the absorption in the gas with the highest velocities ($`6000\mathrm{km}\mathrm{s}^1`$) in the outer shell model would simply disappear in the noise of our rather poor signal-to-noise spectrum.
To limit ourselves to the observed velocities, we adopt a maximum velocity of $`3000\mathrm{km}\mathrm{s}^1`$ to obtain lower limits on the mass and energy of the shell. Table 5 shows that $`M_{\mathrm{sh}}`$ could be very low (i.e., $`M_{\mathrm{sh}}=M_{\mathrm{trunc}}X(`$C IV$`)`$, with $`M_{\mathrm{trunc}}`$ defined in Table 5) if $`\eta `$ is large and if we only use the absorption at $`R_{\mathrm{in}}`$ as a criterion. However, if we also require that the absorption at $`2500\mathrm{km}\mathrm{s}^1`$ should be the same for any $`\eta `$ as in the $`\eta =3`$ model shown in Figure 10, the minimum mass for models with $`\eta 3`$ is given by $`M_{\mathrm{sh}}1.49^{\eta 3}X(`$C IV$`)_{\mathrm{in}}M_{\mathrm{trunc}}M_{}`$, which for $`\eta =3`$ (4, 5, 7, 9) becomes $`M_{\mathrm{sh}}0.27(0.31,0.37,0.55,0.90)M_{}`$. The kinetic energy corresponding to this $`M_{\mathrm{sh}}`$ (i.e., for $`V_{\mathrm{out}}=3000\mathrm{km}\mathrm{s}^1`$) is $`E_{\mathrm{sh}}1.5(1.7,1.8,2.4,3.4)\times 10^{49}`$ ergs. The limits on $`M_{\mathrm{sh}}`$ and $`E_{\mathrm{sh}}`$ for shallower density profiles are fixed by the product $`X(`$C IV$`)M_{\mathrm{trunc}}`$ (cf. Table 5), and are $`M_{\mathrm{sh}}0.48(0.36)M_{}`$ and $`E_{\mathrm{sh}}3.1(2.2)\times 10^{49}`$ ergs, respectively, for $`\eta =1(2)`$. All these limits scale inversely with the overall abundance of carbon.
To summarize the constraints from C IV $`\lambda `$1550, we first emphasize that the line shows that an outer shell with maximum velocity of $`V_{\mathrm{out}}2500\mathrm{km}\mathrm{s}^1`$ appears to be present. We have a $`5\sigma `$ detection of velocities between $`16502780\mathrm{km}\mathrm{s}^1`$. The kinetic energy of the shell depends on its mass, density structure and extent. For a relatively shallow density profile ($`\rho R^3`$, or shallower) the kinetic energy of the shell could be as high as $`10^{51}`$ ergs, if the shell is spherically symmetric and extends to velocities higher than those we can detect with the obtained signal-to-noise. For a maximum velocity of $`6000\mathrm{km}\mathrm{s}^1`$, the mass of the shell required to obtain this kinetic energy would be $`48M_{}`$. A completely flat density profile, however, seems unlikely from the results of LFC86, both in the fast case (LFC86, model 1), and for a model similar to our $`V_{\mathrm{out}}3000\mathrm{km}\mathrm{s}^1`$ case (LFC86, model 4).
For a density profile steeper than $`\rho R^3`$ the kinetic energy is most likely $`<10^{51}`$ ergs since the shell mass should be lower than $`8M_{}`$ to be compatible with progenitor models. The lowest mass and energy we estimate for a shell with $`V_{\mathrm{out}}=3000\mathrm{km}\mathrm{s}^1`$ is for $`\eta 3`$, and are $`0.3M_{}`$ and $`1.5\times 10^{49}`$ ergs, respectively. These values are approximate as they depend on spherical symmetry, the value of $`R_{\mathrm{in}}`$, and a model fit to the line profile of C IV $`\lambda 1548`$ at $`2500\mathrm{km}\mathrm{s}^1`$ where the line profile is rather uncertain. To distinguish between models with different density slopes (see Table 5), photoionization calculations are needed.
#### 3.7.2 Other constraints on the shell
There are, unfortunately, no other lines detected in the UV data that can constrain our current analysis further. Two potentially useful doublets are Si IV $`\lambda \lambda `$ 1394,1403 and N V $`\lambda \lambda `$ 1239,1243, but we cannot identify absorption at high velocities in any of these two doublets. The absence of the two doublets is, however, not surprising. Silicon should be more highly ionized than carbon, and LFC86 found that C IV $`\lambda 1550`$ should produce significantly stronger absorption than N V $`\lambda 1240`$. The spectral region around N V $`\lambda 1240`$ is also rather noisy and the line sits in the damping wing of Ly$`\alpha `$. Although the absence of the lines cannot constrain models in this simple analysis, it can be used in conjunction with photoionization calculations to test different models.
The tentative outer shell has been previously searched for also in the optical. Searches in \[O III\] have been negative for the region outside the observed \[O III\] skin (cf. Fesen et al. 1997, and references therein). This is not surprising from the point of view of the models of LFC86, where oxygen is more highly ionized than O III.
A highly ionized massive shell is bound to give rise to H$`\alpha `$ emission. The deepest search for such emission was done by Fesen et al. (1997) who found a surface brightness limit in H$`\alpha `$ of $`1.5\times 10^7`$ ergs cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup>. With the dereddening suggested by Fesen et al. (1997), i.e., $`A_{\mathrm{H}\alpha }=2.536E(BV)`$, and $`E(BV)=0.52`$ (cf. above), the dereddened surface brightness limit becomes $`5.1\times 10^7`$ ergs cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup>. We have calculated the surface brightness in our models to see how it compares with the observed limit. We use a temperature of the tentative shell of $`2\times 10^4`$ K, which is nearly three times higher than that used by Fesen et al. (1997) and Murdin (1994), but in accordance with the models of LFC86.
The maximum (dereddened) surface brightness in H$`\alpha `$, $`\mathrm{\Sigma }_{\mathrm{H}\alpha }`$, occurs at the impact parameter, $`p=R_{\mathrm{in}}`$, i.e., just at the edge of the observed nebula. The value of $`\mathrm{\Sigma }_{\mathrm{H}\alpha }`$ for this impact parameter is given as a function of $`\eta `$ in Table 5 for the model with $`M_{\mathrm{sh}}=4M_{}`$. It is seen that the modeled surface brightness exceeds the observed limit for $`\eta 4`$. However, $`\mathrm{\Sigma }_{\mathrm{H}\alpha }`$ decreases rapidly with $`p`$ for large $`\eta `$. Table 5 shows that at $`p=1.1\times R_{\mathrm{in}}`$ (corresponding to $`17\mathrm{}`$ outside the observed nebula for a distance of 2 kpc, and roughly where the search by Fesen et al. (1997) was conducted), the surface brightness exceeds the observed limit only for $`\eta 5`$. For a shell mass as low as $`0.30.9M_{}`$, which we found to be likely lower limits to the shell mass for the density slopes investigated, the shell would easily have escaped detection in H$`\alpha `$.
A method to derive parameters for the outer shell was devised by Sankrit & Hester (1997). They estimated the density needed to form a radiative shock at the interface between the nebula and the presumed outer shell, as such a shock is needed in their model to explain the observed \[O III\] skin. They estimate that a minimum density of $`\rho /\mathrm{m}_\mathrm{H}12`$ cm<sup>-3</sup> is needed, at least in the presumed equatorial plane of the nebula. If this is true also in the direction toward the pulsar, the models of Sankrit & Hester yield $`N(`$C IV$`)10^{14}`$ cm<sup>-2</sup> for the radiative tail of the shock. This would escape detection in our data since the intrinsic line width in their model should be small (much less than our spectral resolution). We can therefore not distinguish between a radiative or adiabatic shock (or no shock at all) in the direction to the pulsar. This also means that it is unlikely that any of the absorption we detect occurs in a region similar to the radiative region in the model of Sankrit & Hester.
Sankrit & Hester (1997) estimate that a mean density at the inner edge of the shell, averaged over all polar angles, should be $`8`$ cm<sup>-3</sup>. This would correspond to $`n_\mathrm{H}(R_{\mathrm{in}})5.7\mathrm{cm}^3`$ for the He/H ratio used in Table 5. To see if we can make a consistent model including the C IV line, the H$`\alpha `$ surface brightness limit by Fesen et al. (1997) and the model by Sankrit & Hester (1997), we have assumed an upper limit to the shell mass of $`8M_{}`$ (inside $`V=3000\mathrm{km}\mathrm{s}^1`$), and used the information in Table 5. We then find that $`n_\mathrm{H}(R_{\mathrm{in}})5.7(M_{\mathrm{trunc}}/8M_{})\mathrm{cm}^3`$ is required to get a high enough density at $`R_{\mathrm{in}}`$ to agree with the model of Sankrit & Hester (1997). According to the values in Table 5 this is fulfilled for $`\eta 3`$. The mass and kinetic energy for a shell with such a high $`n_\mathrm{H}(R_{\mathrm{in}})`$ and with $`V_{\mathrm{out}}=3000\mathrm{km}\mathrm{s}^1`$ would be $`6.8(5.0,3.8,2.4,1.7)M_{}`$ and $`3.9(2.7,1.8,1.0,0.7)\times 10^{50}`$ ergs, respectively, for $`\eta =3`$ (4, 5, 7, 9). A caveat for this model is that $`\mathrm{\Sigma }_{\mathrm{H}\alpha }`$ at $`p=1.1\times R_{\mathrm{in}}`$ then becomes $`2.0(1.4,1.0,0.60,0.35)\times 10^6`$ ergs cm<sup>-2</sup> s<sup>-1</sup> sr<sup>-1</sup> for $`\eta =3`$ (4, 5, 7, 9), which is close to, or higher than, the observed limit. In this scenario, a limit on the H$`\alpha `$ surface brightness improved by a factor of a few, close to the observed nebula, should be able to distinguish between models with different density slopes even if the He/H ratio is higher than we have assumed. Photoionization models to accurately calculate the temperature and to check the radial dependence on $`X(`$C IV$`)`$, are also needed.
## 4. Conclusions
Using STIS onboard the HST we have observed the Crab nebula and its pulsar in the far-UV ($`11401720`$ Å). We have obtained the pulse profile of the pulsar, which is very similar to our previous near-UV profile, although the primary peak appears to be marginally narrower than in the near-UV data ($`5\%,2\sigma `$). Combining the far- and near-UV data, and assuming an intrinsic power law for the pulsar continuum, we have derived an extinction of $`E(BV)=0.52`$ mag toward the Crab. No evidence for a non-standard extinction curve was found. We have also added optical spectra taken with the NOT to obtain a spectrum of the pulsar from 1140 Å to 9250 Å. We have shown that the pulsar spectrum can be well fitted over the full UV/optical range by a power law with spectral index $`\alpha _\nu =0.11`$. The exact value of the spectral index is, however, sensitive to the amount and characteristics of the interstellar reddening, and we have investigated this dependence for a likely range of $`E(BV)`$ and $`R`$. In the optical, we find no evidence for the dip in the pulsar spectrum around 5900 Å reported by Nasuti et al. (1996).
The interstellar absorption lines detected in the UV have been analyzed, and are consistent with normal interstellar abundances. The column density of neutral hydrogen is $`(3.0\pm 0.5)\times 10^{21}`$ cm<sup>-2</sup>, which corresponds well to the value derived for $`E(BV)`$. From the Crab nebula itself we detect the emission lines C IV $`\lambda `$1550 and He II $`\lambda `$1640. The ratio of the fluxes of these lines is similar to what has been derived previously, although obtained with much improved spatial resolution.
C IV $`\lambda `$1550 is also seen in absorption toward the pulsar. The line is broad and blueshifted with a maximum velocity of $`2500\mathrm{km}\mathrm{s}^1`$, and there is no absorption at zero velocity. These are the highest velocities measured in the Crab and shows that there exists material outside the visible nebula. This can be interpreted as evidence for the fast shell that has been predicted to surround the Crab nebula (Chevalier 1977). We have used a simple, spherically symmetric model in which the density in the shell falls off with radius as $`R^\eta `$ from $`5\times 10^{18}`$ cm (corresponding to $`1680\mathrm{km}\mathrm{s}^1`$) to derive the mass and energy of such a shell. The conclusions from our model depend on how we tie our model into other observations and models. From the C IV line alone, we find that the minimum mass and kinetic energy of the fast gas are $`0.3M_{}`$ and $`1.5\times 10^{49}`$ ergs, respectively. This occurs for a density slope $`\eta 3`$. A model with a flat ($`\eta =0`$) density profile appears unlikely as the required ionization structure disagrees with the modeling of Lundqvist et al. (1986). The maximum mass of the shell is set by progenitor models, and is unlikely to be much larger than $`8M_{}`$. With a high shell mass, and the velocity extending to velocities much higher than we can detect, the shell could carry an energy of $`10^{51}`$ ergs. The signal-to-noise of C IV $`\lambda `$1550 is too low at high velocities to reject or confirm such a conclusion.
Adding constraints from the model of Sankrit & Hester (1997) to those from the C IV line narrows down the parameter space for the shell. In particular, a density slope of $`\eta 3`$ is required to agree with the interpretation of the observed \[O III\]-skin being a radiative shock. For $`\eta 9`$, the shell mass is then $`1.7M_{}`$ and the kinetic energy $`7\times 10^{49}`$ ergs. Although the limit on the H$`\alpha `$ surface brightness from the search of Fesen et al. (1997) tend to favor models with steep density profiles, a model with $`\eta =3`$ might still be possible, if the He/H ratio is higher than solar also in the fast shell, and the asymmetry of the outer shell different from that in the model of Sankrit & Hester (1997).
We thank Rob Fesen for help during preparations of the HST observations, and for discussions and comments on the manuscript. We also thank Phil Plait for help with barycentric corrections, and Stefan Larsson for advice on period determination. We thank the Swedish National Space Board, and GSFC/NASA for support which enabled JS and PL to visit GSFC. We are also grateful to The Swedish Natural Science Research Council for support. JS was also supported by grants from the Holmberg, Hierta and Magn. Bergvall foundations. The research of RAC is supported through grant NAG5-8130.
|
warning/0002/astro-ph0002489.html
|
ar5iv
|
text
|
# Present and Near-Future Reflected Light Searches for Close-In Planets
## 1. Introduction and Motivation
Radial velocity surveys of nearby Sun-like stars have uncovered a population of close-in orbiting companions of roughly Jupiter mass. The ten such objects with semi-major axes $`a<0.1`$ AU are listed in Table 1, along with the values for the semi-major axes, (minimum) masses, spectral types of the stars, and equilibrium temperatures (calculated from the estimated values for the semi-major axes, stellar radii and effective temperatures, and assuming a Bond albedo of $`A`$).
A successful spectroscopic detection of an extrasolar planet in reflected light would yield the inclination, and hence the planetary mass, and would also measure a combination of the planetary radius and albedo. Furthermore, it would open the way to direct investigation of the spectrum of the planet itself. Conversely, a low enough upper limit would provide useful constraints on the radius and albedo of the companion.
The predicted albedo of a close-in extrasolar giant planet has been the focus of recent theoretical work (Marley et al. 1999; Seager, Whitney, & Sasselov 2000), and the possible values range by several orders of magnitude based on the atmospheric temperature, chemical composition (see, e.g., Burrows & Sharp 1999), and the presence (or absence) of atmospheric condensates, and their respective size distributions.
## 2. The Method of Reflected Light
The amplitude, relative to the star, of the observed flux reflected from a close-in planet viewed at a phase angle $`\alpha `$ (where $`\alpha `$ is the angle between the star and the observer as viewed from the planet) is given by
$$f_\lambda (\alpha )=ϵ\varphi _\lambda (\alpha )=p_\lambda \left(\frac{R_p}{a}\right)^2\varphi _\lambda (\alpha ),$$
(1)
where $`R_p`$ is the planetary radius, $`a`$ is the semi-major axis, and $`p_\lambda `$ is the wavelength-dependent geometric albedo. The quantity $`ϵ`$ is the contrast ratio at opposition. The phase function $`\varphi _\lambda (\alpha )`$ is the brightness of the planet viewed at an angle $`\alpha `$ relative to its value at opposition (where by definition $`\varphi _\lambda (0)=1`$), and is a monotonically decreasing function of $`\alpha `$ for all physically reasonable atmospheres. For planets detected by the radial velocity technique alone, $`a`$ is determined, and $`\alpha `$ at a given time is prescribed by the known orbital phase $`\mathrm{\Phi }`$ and the unknown orbital inclination $`i`$, but $`R_p`$, $`p_\lambda `$ and $`\varphi _\lambda (\alpha )`$ remain unknown.
A distant observer viewing the solar system in the $`V`$ band would measure a tiny reflected light ratio of $`f4\times 10^9`$ from Jupiter viewed at opposition. This ratio increases dramatically for close-in, giant planets, due to the small orbital separation. Scaling equation (1), we find
$$f_\lambda (\alpha )9.1\times 10^5p_\lambda \left(\frac{R_p}{R_{\mathrm{Jup}}}\right)^2\left(\frac{0.05\mathrm{AU}}{a}\right)^2\varphi _\lambda (\alpha ).$$
(2)
This photometric modulation is most likely beyond the reach of ground-based observations, but will be accessible to upcoming photometric satellite missions, such as COROT (see, e.g., Rouan et al. (1999)), provided such missions can achieve a precision of $`20\mu \mathrm{mag}`$ with stability over timescales of a few days.
Photometry alone will not yield the orbital inclination, and hence the planetary mass will remain constrained only by the lower limit imposed by the radial velocity observations. However, a spectrum of a close-in planet system will contain a secondary component that varies in brightness according to equation (1) and in Doppler shift (relative to the star) according to
$$v_p(\mathrm{\Phi })=K_s\frac{M_s+M_p}{M_p}\mathrm{cos}2\pi \mathrm{\Phi },$$
(3)
where $`M_s`$ and $`M_p`$ are the stellar and planetary masses. The stellar radial velocity amplitude $`K_s`$ and orbital phase $`\mathrm{\Phi }`$ are determined from the observed radial velocity orbit of the star. For the close-in planets, $`|v_p|100\mathrm{km}\mathrm{s}^1`$, which is two orders of magnitude larger than the resolution of current spectrographs. If the reflected light spectrum of the planet is detected, then the planetary mass is determined by equation (3).
Furthermore, as first discussed in Charbonneau, Jha & Noyes (1998), if the star has been tidally spun-up so that its rotation period is equal to the planetary orbital period, then the planet would reflect a non-rotationally-broadened spectrum, further distinguishing these features from the stellar lines.
In summary, the observational challenge is similar to that of transforming a single-lined binary system into a double-lined system, in the case of an exceptionally large contrast ratio between the two components. See Charbonneau et al. (1999), hereafter C99, for more details.
## 3. Results for $`𝝉`$ Boö
Figure 1 summarizes our search for the reflected light spectrum from the planet orbiting the star $`\tau `$ Boö, as described in C99. At the 99% confidence level, we find no evidence for a reflected flux ratio in excess of $`1\times 10^4`$. For edge-on values of the inclination ($`i>70^{}`$), this ratio is further restricted to be less than $`5\times 10^5`$. Assuming a planetary radius of $`1.2R_{\mathrm{Jup}}`$ (Guillot et al. 1996), this limits the geometric albedo to $`p_\lambda 0.3`$ for 466 nm $`\lambda `$ 498 nm.
These conclusions are marginally in conflict with the recent claim of a possible detection of reflected light for the same system by Cameron et al. (1999). They find a reflected light ratio of $`(1.9\pm 0.4)\times 10^4`$ in the wavelength range from 456 nm to 524 nm, and do not detect the planet in wavelength bands on either side of this region (385 nm to 456 nm, and 524 nm to 611 nm). Figure 2 demonstrates the confidence intervals that we would have obtained on the two unknowns $`\{ϵ,i\}`$ for inclinations near the one claimed by Cameron et al. (1999) and for a selection of contrast ratios.
Taking into account the different descriptions of the assumed phase function $`\varphi _\lambda (\alpha )`$, and the respective errors stated in the two papers, it is possible that the two results are consistent, i.e. the inclination claimed by Cameron et al. (1999) is correct, but the amplitude of the signal is less than their most probable value by 1.5 times their quoted error. However, Cameron et al. (1999) caution that there is a 5% possibility that their detection is spurious. This controversy should be resolved by observations in the near future that can now be tailored (in selection of wavelength region and orbital phase) to verify the claimed detection.
## 4. Additional Targets
There are two principal considerations that enter into the choice of which planets may be most easily detected by reflected light: An ideal system is one with a small semi-major axis and a large stellar apparent brightness, and thus a large reflected light signal relative to the photon noise of the star. There are two other systems that satisfy these criteria, and we compare below the prospects for detecting each of these with those for $`\tau `$ Boö ($`a=0.046`$ AU, $`V=4.5`$).
The planet orbiting $`\upsilon `$ And is further from its star ($`a=0.059`$ AU), but the star has a greater apparent brightness ($`V=4.1`$). Combining these two effects, the detection threshold that could be established (given an equivalent amount of observing time) would be 1.36 times higher relative to the primary star than that for $`\tau `$ Boö. For the planet orbiting 51 Peg, the larger semi-major axis ($`a=0.051`$ AU) and fainter star ($`V=5.5`$) result in a detection threshold that is 1.93 times higher than that for $`\tau `$ Boö.
Near edge-on inclinations ($`i90^{}`$) are desirable for reflected light observations since they allow the planet to be viewed at the full range of phase angles. There are additional considerations that may constrain the inclination for these two systems. Assuming that the orbital planes of the three planets orbiting $`\upsilon `$ And (Butler et al. 1999) are co-aligned, stability arguments (Laughlin & Adams 1999) favor $`\mathrm{sin}i>0.75`$. However, Hipparcos astrometry favors (Mazeh et al. 1999) $`\mathrm{sin}i0.4`$. The rotation periods of $`\upsilon `$ And and 51 Peg indicate that the stellar rotation has not become tidally locked to the orbital period, and this in turn might imply planetary masses which are near the minimum mass values (Drake et al. 1998).
We expect 51 Peg b and $`\upsilon `$ And b to have larger radii than the more massive $`\tau `$ Boö b (Guillot et al. 1996). If $`\tau `$ Boö b has been detected, it has a very large surface area. If 51 Peg b and $`\upsilon `$ And b have comparable albedos to $`\tau `$ Boö b, they should have a comparable or larger reflected light ratio.
## 5. The Transiting System HD 209458
In the case of a planet that is observed to transit its star, the situation improves considerably. Since transit observations yield $`R_p`$ and $`i`$, then only remaining unknowns in equation (1) are $`p_\lambda `$ and $`\varphi _\lambda (\alpha )`$.
The first transiting extrasolar planet, HD 209458 b, was detected by Charbonneau et al. (2000) and Henry et al. (2000). The dominant uncertainty in deriving the planetary radius and orbital inclination is the estimation of the stellar radius, and not the photometric precision. A detailed analysis of the stellar modeling, and the resulting derivation of the planetary parameters and the systematic uncertainties is presented in Mazeh et al. (2000). They derive $`R_p=1.40\pm 0.17R_{\mathrm{Jup}}`$, $`a=0.467`$ AU, and $`i=86.^{}1\pm 1.^{}6`$. Substituting these values into equation (3), we find $`f_\lambda (\alpha )=2.1\times 10^4p_\lambda \varphi _\lambda (\alpha )`$.
Transiting systems such as HD 209458 are particularly desirable targets for reflected light measurements since by securing measurements near opposition (where $`\varphi _\lambda (0)1`$), an observed reflected light amplitude is a direct measurement of the geometric albedo. Furthermore, one can use the spectra obtained while the planet is behind the star (rather than near inferior conjunction) to create the required stellar template spectrum. One difficulty is that HD 209458 is significantly fainter ($`V=7.6`$) than the systems described above.
We have simulated what could be achieved in a modest observing run using the Keck-1 telescope and HIRES spectrograph (Vogt et al. 1994), based on experience gained from our earlier observing run on $`\tau `$ Boö (C99). In three nights when the planet is near opposition, we would obtain a detection threshold of $`4\times 10^5`$ for the reflected flux ratio. Thus we would constrain the geometric albedo to be $`p_\lambda 0.2`$, or directly detect the planet in reflected light.
### Acknowledgments.
We thank our collaborators in our reflected light study of $`\tau `$ Boö, Sylvain Korzennik, Peter Nisenson, Saurabh Jha, Steven Vogt, and Robert Kibrick. We also thank Timothy Brown for many discussions regarding HD 209458.
## References
Baliunas, S. A., Henry, G. W., Donahue, R. A., Fekel, F. C., & Soon, W. H. 1997, ApJ, 474, L119
Burrows, A. & Sharp, C. M. 1999, ApJ, 512, 843
Butler, R. P. Marcy, G. W., Fischer, D. A., Brown, T. M., Contos, A. R., Korzennik, S. G., Nisenson, P., Noyes, R. W. 1999, ApJ, 526, 916
Cameron, A. C., Horne, K., Penny, A., & James, D. 1999, Nature, 402, 751
Charbonneau, D., Brown, T. M., Latham, D. W., & Mayor, M. 2000, ApJ, 529, L45
Charbonneau, D., Jha, S., & Noyes, R. W. 1998, ApJ, 507, L153
Charbonneau, D., Noyes, R. W., Korzennik, S. G., Nisenson, P., Jha, S., Vogt, S. S., & Kibrick, R. I. 1999, ApJ, 522, L145 (C99)
Drake, S. A., Pravdo, S. H., Angelini, L., & Stern, R. A. 1998, AJ, 115, 2122
Gray, D. F. 1982, ApJ, 261, 259
Guillot, T., Burrows, A., Hubbard, W. B., Lunine, J. I., & Saumon, D. 1996, ApJ, 459, L35
Henry, G. W., Marcy, G. W., Butler, R. P., & Vogt, S. S. 2000, ApJ, 529, L41
Laughlin, G. & Adams, F. C. 1999, ApJ, 526, 881
Marley, M. S., Gelino, C., Stephens, D., Lunine, J. I., & Freedman, R. 1999, ApJ, 513, 879
Mazeh, T., Zucker, S., Dalla Torre, A., van Leeuwen, F. 1999, ApJ, 522, L149
Mazeh, T., et al. 2000, ApJ, in press
Rouan, D., et al. 1999, Phys. Chem. Earth, 24, 567
Seager, S., Whitney, B. A., & Sasselov, D. D. 2000, ApJ, submitted
Vogt, S. S. et al. 1994, Proc. SPIE, 2198, 362
|
warning/0002/quant-ph0002058.html
|
ar5iv
|
text
|
# An unentangled Gleason’s theorem
## 1. Introduction.
Let $`H`$ be a Hilbert space with unit sphere $`S(H)`$. Following Gleason (\[Gleason\]) we will call a function $`f:S(H)`$ a frame function of weight $`w`$ if for every orthonormal basis $`\{v_i\}`$ of $`S(H)`$
(1)
$$\underset{i}{}f(v_i)=w\text{.}$$
In \[Gleason\] the following theorem was proved
###### Theorem 1.
If $`dimH3`$ and $`f`$ is a frame function that takes non-negative real values then there exists a self adjoint trace class operator $`T:HH`$ such that
$$f(v)=v|T|v,vS(H)\text{.}$$
This theorem is of importance to quantum mechanics because it allows a significant weakening of the axioms, showing that the Born probability rule \[Born\] provides the unique class of probability assignments for measurement outcomes so long as those probabilities are specified by frame functions \[Pitowsky\]. The theorem also rules out a large class of hidden-variable explanations for quantum statistics, the so-called noncontextual hidden variables, in dimension $`3`$ or greater. The interested reader should consult \[Bell\] for a discussion of this point. If the Hilbert space is of dimension $`2`$, then the statement in the theorem is easily seen to be false.
The purpose of this note is to give a generalization of Gleason’s theorem inspired by recent work in quantum information theory. In that context the issue of *local* measurements and operations on multipartite quantum systems (as opposed to the full set of operations) is of the utmost importance \[BDFMRSSW\]. For instance, it has been pointed out that probabilities for the outcomes of local measurements are enough to uniquely specify the quantum state from which they arise if the field of the Hilbert space is complex, though this fails for real and quaternionic Hilbert spaces \[Araki,Wootters\].
Chris Fuchs has asked to what extent local and semi-local measurements not only uniquely specify the quantum state, but also a Born-like rule as in Gleason’s result \[Fuchs\]. In this regard, the following formalization appears natural. We confine our attention to finite dimensional Hilbert spaces for the sake of simplicity. Let $`H_1,\mathrm{},H_n`$ be Hilbert spaces. Set $`H=H_1H_2\mathrm{}H_n`$. Let $`\mathrm{\Sigma }=\mathrm{\Sigma }(H_1,\mathrm{},H_n)`$ denote the subset of $`S(H)`$ consisting of those elements of the form $`a_1\mathrm{}a_n`$ with $`a_iS(H_i)`$ for $`i=1,\mathrm{},n`$. In the jargon of quantum information theory such states are called *unentangled* or *product states*. The ones that are not of this form are said to be *entangled*. An orthonormal basis $`\{v_i\}`$ of $`H`$ is said to be unentangled if $`v_i\mathrm{\Sigma }`$ for all $`i`$. We say that $`f:\mathrm{\Sigma }`$ is an unentangled frame function of weight $`w`$ if whenever $`\{v_i\}`$ is an unentangled orthonormal basis of $`H`$ then $`f`$ satisfies (1) above. We establish the following result.
###### Theorem 2.
If $`dimH_i3`$ for all $`i`$ and if $`f:\mathrm{\Sigma }`$ is a non-negative unentangled frame function then there exists $`T:HH`$ a self adjoint trace class operator such that $`f(v)=v|T|v`$ for all $`v\mathrm{\Sigma }`$.
This theorem is an almost direct consequence of Gleason’s original theorem. We will give a proof of it in the next section. The second result in this paper shows that the dimensional condition is necessary.
It should be noted however, that despite the absence of entangled or “nonlocal” states in $`\mathrm{\Sigma }`$, in \[BDFMRSSW\] it is asserted that not all unentangled bases correspond to quantum measurements that can be carried out by local means alone (even with iterative procedures based on weak local measurements and unlimited amounts of classical communication between the measurers at each site). The simplest kind of purely local measurement is given by an alternative type of basis adapted to the tensor product structure. This is a *product basis* and is defined as to be a basis of the form $`\{u_{i_11}u_{i_22}\mathrm{}u_{i_nn}\}`$ where $`u_{1j},\mathrm{},u_{n_jj}`$ is an orthonormal basis of $`H_j`$. We could define a product frame function in the same way as we did for an unentangled frame function except that we only assume that there exists a weight $`w`$ such that $`_{i_1,i_2,\mathrm{},i_n}f(u_{i_11}u_{i_22}\mathrm{}u_{i_nn})=w`$ for every product basis. One can ask whether this is all that us necessary for the conclusion of the theorem above. The answer is no and a method of “finding” a large class of examples will be given at the end of the next section (see the proposition at the end of the section). This result amasses some evidence that the structure of local measurements alone is not enough to establish the Born rule for multipartite systems, but a full answer would require consideration of the largest class of local measurements in \[BDFMRSSW\].
These issues also spawn another theorem.
###### Theorem 3.
Let $`dimH_1=2`$ and let $`f:S(H_1)`$ be a frame function of weight $`w_1`$ and $`g:\mathrm{\Sigma }(H_2,\mathrm{},H_n)`$ be an unentangled frame function of weight $`w_2`$. We set $`h(v_1u)=f(v_1)g(u)`$ for $`u\mathrm{\Sigma }(H_2,\mathrm{},H_n)`$. Then $`h`$ is an unentangled frame function of weight $`w_1w_2`$.
This result is a bit harder and the proof involves a method (see Theorem 5) that describes a combinatorial scheme for finding all unentangled orthonormal bases where all of the spaces, $`H_i`$, have dimension $`2`$. This analysis in turn leads to a natural question. Given and unentangled orthonormal set can it be extended to an unentangled orthonormal basis? Or even stronger: Can it be a proper subset of an unentangled orthonormal set? This question was studied in \[BDMSST\]. We conclude the paper by giving a proof based on simple algebraic geometry of the following theorem which is related to the bound that occurs in \[BDMSST\].
###### Theorem 4.
Let $`V`$ be a subspace of $`H_1\mathrm{}H_n`$ such that if $`vV`$ and $`v0`$ then $`v`$ is entangled. Then $`dimVdim(H_1)\mathrm{}dim(H_n)(dimH_i1)1`$. Furthermore, the upper bound is attained.
## 2. The unentangled Gleason theorem.
In this section we will give a proof of Theorem 2. If $`n=1`$ the statement is just Gleason’s theorem. We consider the situation of $`H=H_0V`$ with $`V=H_1H_2\mathrm{}H_n`$ and $`dimH_i3`$ for all $`i`$. We prove Theorem 1 by induction (i.e. assume the result for $`n)`$. We note that if $`\{v_i\}`$ is an orthonormal basis of $`H_0`$ and if for each $`i`$, $`\{u_{ij}\}`$ is an unentangled orthonormal basis of $`V`$ then the set $`\{v_iu_{ij}\}`$ is an unentangled orthonormal basis of $`H`$. Thus if $`w`$ is the weight of $`f`$ then we have
$$\underset{j}{}f(v_1u_{1j})=w\underset{i2,j}{}f(v_iu_{ij}).$$
Thus for each $`vS(H_0)`$ the function $`f_v(u)=f(vu)`$ is an unentangled frame function. The inductive hypothesis implies that for each $`vS(H_0)`$ there exists a self adjoint (due to the reality of $`f`$) linear operator $`T(v)`$ such that $`f(vu)=u|T(v)|u`$ for $`u\mathrm{\Sigma }(H_1,\mathrm{},H_n)`$. Similarly, if $`\{u_i\}`$ is an unentangled orthonormal basis of $`V`$ and for each $`i`$, $`\{v_{ij}\}`$ is an orthonormal basis of $`H_0`$ then $`\{v_{ij}u_i\}`$ is an unentangled orthonormal basis of $`H`$. We therefore conclude as above that if $`u\mathrm{\Sigma }(H_1,\mathrm{},H_n)`$ then there exists $`S(u)`$ a self adjoint linear operator on $`H_0`$ so that $`f(vu)=v|S(u)|v`$ for all $`vH_0`$.
Let $`\{u_i\}`$ be an unentangled orthonormal basis of $`V`$ and let $`\{v_j\}`$ be an orthonormal basis of $`H_0`$. Set
$$a_{ij}(v)=u_i|T(v)|u_j$$
and
$$b_{ij}(u)=v_i|S(u)|v_j,u\mathrm{\Sigma }(H_1,\mathrm{},H_n).$$
We now observe that if $`v=_ix_iv_i`$ and if $`u=_jy_ju_j`$ then we have
$$\underset{p,q}{}a_{p,q}(v)\overline{y}_py_q=\underset{r,s}{}b_{r,s}(u)\overline{x}_rx_s.$$
If we substitute $`v=v_r`$ then we have
$$b_{rr}(u)=\underset{p,q}{}a_{p,q}(v_r)\overline{y}_py_q\text{.}$$
Now assuming that $`rs`$ and taking $`v=\frac{1}{\sqrt{2}}(v_r+v_s)`$ we have
$`\mathrm{Re}b_{rs}(u)`$ $`={\displaystyle \underset{p,q}{}}a_{p,q}({\displaystyle \frac{1}{\sqrt{2}}}(v_r+v_s))\overline{y}_py_q`$
$`{\displaystyle \frac{1}{2}}\left({\displaystyle \underset{p,q}{}}a_{p,q}(v_r)\overline{y}_py_q+{\displaystyle \underset{p,q}{}}a_{p,q}(v_s)\overline{y}_py_q\right).`$
Also if we take $`v=\frac{1}{\sqrt{2}}(v_r+iv_s)`$ then we have
$`\mathrm{Im}b_{rs}(u)`$ $`={\displaystyle \underset{p,q}{}}a_{p,q}({\displaystyle \frac{1}{\sqrt{2}}}(v_r+iv_s))y_p\overline{y}_q`$
$`{\displaystyle \frac{1}{2}}\left({\displaystyle \underset{p,q}{}}a_{p,q}(v_r)y_p\overline{y}_q+{\displaystyle \underset{p,q}{}}a_{p,q}(v_s)y_p\overline{y}_q\right).`$
Thus if we set
$$c_{rrpq}=a_{pq}(v_r)$$
and if $`rs`$ then
$`c_{rspq}`$ $`=a_{p,q}\left({\displaystyle \frac{1}{\sqrt{2}}}(v_r+v_s){\displaystyle \frac{1}{2}}(a_{p,q}(v_r)+a_{p,q}(v_s))\right)+`$
$`a_{p,q}\left({\displaystyle \frac{1}{\sqrt{2}}}(v_r+iv_s){\displaystyle \frac{1}{2}}(a_{p,q}(v_r)+a_{p,q}(v_s))\right)`$
Then
$$f(vu)=\underset{rspq}{}c_{rspq}\overline{x}_r\overline{y}_px_sy_q.$$
This is the content of the theorem.
We will now give a counterexample to the analogous assertion for product bases.
###### Proposition 5.
Let $`H_1`$ and $`H_2`$ be finite dimensional Hilbert spaces of dimension greater than $`1`$. Then there exists $`f:\mathrm{\Sigma }(H_1,H_2)[0,\mathrm{})`$ such that $`_{i,j}f(u_iv_j)=w`$, with $`w`$ fixed, for all choices $`\{u_i\}`$ and $`\{v_j\}`$ of orthonormal bases of $`H_1`$ and $`H_2`$ respectively but there is no linear endomorphism, $`T`$, on $`H_1H_2`$ such that $`f(uv)=uv|T|uv`$ for $`uS(H_1)`$ and $`vS(H_2)`$.
Proof. Let for $`w>0`$, $`𝒫_w`$ denote the set of all Hermitian positive semi-definite endomorphisms, $`A`$, of $`H_2`$ such that $`tr(A)=w`$. Fix $`w_o=\frac{w}{dimH_1}`$. Let $`\phi :S(H_1)𝒫_{w_o}`$ be a mapping (completely arbitrary). Set $`f(uv)=v|\phi (u)|v`$, for $`uS(H_1)`$ and $`vS(H_2)`$ . If $`\{u_i\}`$ is an orthonormal basis of $`H_1`$ and if $`\{v_j\}`$ is an orthonormal basis of $`H_2`$ then
$$\underset{i,j}{}f(u_iv_j)=\underset{i}{}\left(\underset{j}{}v_j|\phi (u_i)|v_j\right)=\underset{i}{}tr(\phi (u_i))=dim(H_1)w_o.$$
Note: In this argument only one factor need be finite dimensional. Also note that $`f`$ can be chosen to be continuous.
## 3. Unentangled Bases
In this section we will develop the material on “unentangled bases” that we will need to prove Theorem 3 (in fact as we shall see a generalization). Let $`V`$ be a $`2`$-dimensional Hilbert space and let $`H`$ be an $`n`$-dimensional Hilbert space. Fix $`\mathrm{\Sigma }S(H)`$ such that $`\lambda \mathrm{\Sigma }=\mathrm{\Sigma }`$ for all $`\lambda `$ with $`|\lambda |=1`$. We will use the notation $`S(V)\mathrm{\Sigma }=\{vw|vS(V),w\mathrm{\Sigma }\}`$.
If $`aS(V)`$ then up to scalar multiple there is exactly one element of $`S(V)`$ that is perpendicular to $`a`$. We will denote a choice of such an element by $`\widehat{a}`$. The main result of this section is
###### Theorem 6.
If $`\{u_j\}_{j=1}^{2n}`$ is an orthonormal basis of $`VH`$ with $`u_jS(V)\mathrm{\Sigma }`$ for $`j=1,\mathrm{},2n`$ then there exists a partition
$$n_1n_2\mathrm{}n_r>0$$
of $`n`$, an orthogonal decomposition
$$H=U_1\mathrm{}U_r,$$
elements $`a_1,\mathrm{},a_rS(H)`$, and for each $`i=1,\mathrm{},r`$ orthonormal bases $`\{b_{i1},\mathrm{},b_{in_i}\}`$ and $`\{c_{i1},\mathrm{},c_{in_i}\}`$ of $`U_i`$ such that
$$\{u_i|i=1,\mathrm{},2n\}=\underset{i=1}{\overset{r}{}}\left(\{a_ib_{ij}|j=1,\mathrm{},n_i\}\{\widehat{a}_ic_{ij}|j=1,\mathrm{},n_i\}\right).$$
Before we prove the theorem we will make several preliminary observations. Let $`\{u_i\}`$ be as in the statement of the theorem. Then each $`u_i=a_ih_i`$ with $`a_iS(V)`$ and $`h_i\mathrm{\Sigma }`$.
1. For each $`i`$ there exists $`j`$ such that $`a_j`$ is a multiple of $`\widehat{a}_i`$.
If not then we would have $`a_i|a_j0`$ for all $`j`$. Since $`a_ih_i|a_jh_j=a_i|a_jh_i|h_j`$, $`h_i|h_j=0`$ for all $`ji`$. This implies that $`\{u_j\}_{ji}V\{h_i^{}\}`$. This space has dimension equal to $`2(n1)`$. So it could not contain $`2n1`$ orthonormal elements. This contradiction implies that assertion 1. is true.
2. Assume that $`ij`$. If $`a_i|a_j0`$ then $`h_i|h_j=0`$. If $`h_i|h_j0`$ then $`a_i|a_j=0`$.
This is clear (see the proof of 1.)
We will now prove the theorem by induction on $`n`$. If $`n=1`$ the result is trivial. We assume the result for all $`H`$ with $`dimH<n`$ and all possible choices for $`\mathrm{\Sigma }`$. We now prove it for $`n`$.
For each $`i`$ let $`m_i`$ denote the number of $`j`$ such that $`a_j`$ is a multiple of $`a_i`$. Let $`m=\mathrm{max}\{m_i|i=1,\mathrm{},2n\}`$. If we relabel we may assume that the first $`m`$ of the $`a_i`$ are equal to $`a_1`$ (we may have to multiply $`h_i`$ by a scalar of norm $`1`$). By 1. above we may assume that the next $`k`$ of the $`a_i`$ are equal to $`\widehat{a}_1`$ with $`1km`$ and if $`i>m+k`$ then $`a_i`$ is not a multiple of either $`a_1`$ or $`\widehat{a}_1`$. This implies by 2. above that $`h_i|h_j=0`$ for $`j>m+k`$ and $`i=1,\mathrm{},m`$. Also $`\{h_1,\mathrm{},h_m\}`$ is an orthonormal set. Thus $`u_iV(\{h_1,\mathrm{},h_m\}^{})`$ for $`i>m+k`$. This implies that $`V(\{h_1,\mathrm{},h_m\}^{})`$ contains $`2n(m+k)`$ orthonormal elements. Since $`dimV(\{h_1,\mathrm{},h_m\}^{})=2(nm)`$ this implies that $`k=m`$. We now rewrite the first $`2m`$ elements of the basis as
$$a_1b_1,\mathrm{},a_1b_m,\widehat{a}_1c_1,\mathrm{},\widehat{a}_1c_m.$$
If we apply observation 2. again we see that the elements $`h_i`$ for $`i>2m`$ must be orthogonal to $`\{b_1,\mathrm{},b_m\}`$ and to $`\{c_1,\mathrm{},c_m\}`$. A dimension count says that they must span the orthogonal complements of both $`\{b_1,\mathrm{},b_m\}`$ and $`\{c_1,\mathrm{},c_m\}`$. But then $`\{b_1,\mathrm{},b_m\}`$ and $`\{c_1,\mathrm{},c_m\}`$ must span the same space, $`UH`$. We have therefore shown that $`\{u_i\}_{i>2m}`$ is an orthonormal basis of $`VU^{}`$. We may thus apply the inductive hypothesis to $`U^{}`$ and $`\mathrm{\Sigma }U^{}`$. This completes the inductive step and hence the proof.
If $`W`$ is a Hilbert space and if $`\mathrm{\Xi }`$ is a subset of $`S(W)`$ that is invariant under multiplication by scalars of absolute value $`1`$ then a function $`f:\mathrm{\Xi }`$ is said to be a $`\mathrm{\Xi }`$-frame function of weight $`w=w_f`$ if whenever $`\{u_i\}`$ is an orthonormal basis of $`W`$ with $`u_i\mathrm{\Xi }`$ (i.e. $`\{u_i\}`$ is a $`\mathrm{\Xi }`$-frame) we have $`_if(u_i)=w`$. We note
3. Let $`f`$ be a $`\mathrm{\Xi }`$-frame function. If $`\{u_i\}`$ is a $`\mathrm{\Xi }`$-frame for $`W`$ and if $`F`$ is a subset of $`\{u_i\}`$ then $`f_{|F^{}\mathrm{\Xi }}`$ is a $`F^{}\mathrm{\Xi }`$-frame function of weight $`w_f_{u_iF}f(u_i)`$.
This is pretty obvious. Let $`\{v_j\}`$ be a $`\mathrm{\Xi }F^{}`$-frame for $`F^{}`$. Then $`\{\nu _j\}F`$ is a $`\mathrm{\Xi }`$-frame for $`W`$.
###### Proposition 7.
Let $`V`$ be a two dimensional Hilbert space and let $`H`$ be an $`n`$-dimensional Hilbert space. Let $`\mathrm{\Sigma }S(H)`$ be as in the rest of this section and let $`g:S(V)`$ and $`h:\mathrm{\Sigma }`$ be respectively a frame function and a $`\mathrm{\Sigma }`$-frame function. Then if $`f(vw)=g(v)h(w)`$ for $`vS(H)`$ and $`w\mathrm{\Sigma }`$ then $`f`$ is an $`S(V)\mathrm{\Sigma }`$-frame function of weight $`w_gw_h`$.
Proof. Let $`\{u_i\}`$ be an $`S(V)\mathrm{\Sigma }`$-frame. Then Theorem 5 implies that we may assume that there is partition $`n_1n_2\mathrm{}n_r>0`$ of $`n`$ and elements $`a_i,b_{ij}`$and $`c_{ij}`$ as in the statement so that
$$\{u_i\}=\underset{i=1}{\overset{r}{}}\left(\{a_ib_{ij}|j=1,\mathrm{},n_i\}\{\widehat{a}_ic_{ij}|j=1,\mathrm{},n_i\}\right).$$
Thus
$$\underset{i}{}f(u_i)=\underset{i}{}g(a_i)\underset{j=1}{\overset{n_i}{}}h(b_{ij})+\underset{i}{}g(\widehat{a}_i)\underset{j=1}{\overset{n_i}{}}h(c_{ij}).$$
Observation 3. above implies that for each $`i`$ we have $`_{j=1}^{n_i}h(b_{ij})=_{j=1}^{n_i}h(c_{ij})`$. Now $`g(a_i)+g(\widehat{a}_i)=w_g`$. Hence since $`\{b_{ij}\}`$ is a $`\mathrm{\Sigma }`$-frame the result follows.
Theorem 3 is an immediate consequence of the above proposition.
## 4. Entangled subspaces.
Let $`H_1,\mathrm{},H_n`$ be finite dimensional Hilbert spaces and set $`H=H_1H_2\mathrm{}H_n`$. If $`VH`$ is a subspace than we will say then $`V`$ is *entangled* if whenever $`vV`$ and $`v0`$ then $`v`$ is entangled (i.e. $`v`$ cannot be written in the form $`v=h_1h_2\mathrm{}h_n`$ for any choice of $`h_iH_i`$). The purpose of this section is to give a proof of Theorem 4 using basic algebraic geometry. That is, we will prove that
$$dimVdim(H_1)\mathrm{}dim(H_n)(dimH_i1)1$$
and that this estimate is best possible. The reader should consult \[Hartshorne\] for the algebraic geometry used in the proof of this result.
Let $`L=\{\lambda H^{}|\lambda (V)=0\}`$ ($`H^{}`$ the complex dual space of $`H`$). Let $`X=\{h_1\mathrm{}h_n|h_iH_i\}`$. We consider the map $`\mathrm{\Phi }:H_1\times \mathrm{}\times H_nX`$ given by $`\mathrm{\Phi }(h_1,\mathrm{},h_n)=h_1\mathrm{}h_n`$. Then $`\mathrm{\Phi }`$ is a surjective polynomial mapping. If we denote by $`\overline{\mathrm{\Phi }}`$ the corresponding mapping of projective spaces we have $`\overline{\mathrm{\Phi }}:P(H_1)\times \mathrm{}\times P(H_n)P(H)`$. General theory implies that the image of $`\overline{\mathrm{\Phi }}`$ is Zariski closed in $`P(H)`$. Since $`X`$ is clearly the cone on that image we see that $`X`$ is Zariski closed and irreducible. Also the map $`\overline{\mathrm{\Phi }}`$ is injective so the dimension over $``$ of its image is $`(dimH_i1)`$. Thus the dimension over $``$ of $`X`$ is $`d=(dimH_i1)+1`$.
Since $`V`$ is entangled $`XV=\{0\}`$. This implies that $`\{xX||\lambda (x)=0,\lambda L\}=\{0\}`$. Thus $`dimLdimX=d`$. Hence $`dimV=dimHdimLdimHd`$. This is the asserted upper bound. The fact that this upper bound is best possible follows from the Noether normalization theorem which implies that there exist $`\lambda _1,\mathrm{},\lambda _dH^{}`$ such that $`\{xX||\lambda _i(x)=0`$ for all $`i\}=\{0\}`$ (i.e. a linear system of parameters).
References
\[Gleason\] A. M. Gleason, Measures on the closed subspaces of a Hilbert space, J. Math. and Mech. 6 (1957), 885–893.
\[Born\] M. Born, Zur Quantenmechanik der Stossvorgänge, Zeits. Phys. 37 (1926), 863–867. Reprinted and translated in *Quantum Theory and Measurement*, edited by J. A. Wheeler and W. H. Zurek (Princeton U. Press, Princeton, NJ, 1983), pp. 52–55.
\[Pitowsky\] I. Pitowsky, Infinite and finite Gleason’s theorems and the logic of indeterminacy, J. Math. Phys. 39 (1998), 218–228.
\[Bell\] J. S. Bell, On the problem of hidden variables in quantum mechanics, Rev. Mod. Phys. 38 (1966), 447–452.
\[BDFMRSSW\] C. H. Bennett, D. P. DiVincenzo, C. A. Fuchs, T. Mor, E. Rains, P. W. Shor, J. A. Smolin, and W. K. Wootters, Quantum nonlocality without entanglement, Phys. Rev. A 59 (1999), 1070–1091.
\[Araki\] H. Araki, On a characterization of the state space of quantum mechanics, Comm. Math. Phys 75 (1980), 1–24.
\[Wootters\] W. K. Wootters, Local accessibility of quantum states, in *Complexity, Entropy and the Physics of Information*, edited by W. H. Zurek (Addison-Wesley, Redwood City, CA, 1990), pp. 39–46.
\[Fuchs\] C. A. Fuchs, private communication.
\[BDMPSST\] C. H. Bennett, D. P. DiVincenzo, T.Mor, P. W. Shor, J. A. Smolin, B. M. Terhal, Unextendible product bases and bound entanglement, Phys. Rev. Lett. 82 (1999) 5385–5388.
\[Hartshorne\] R. Hartshorne, *Algebraic Geometry*, Graduate Texts in Mathematics, 52, Springer-Verlag, New York, 1977.
Nolan R. Wallach
University of California, San Diego
*E-mail address:* nwallach@ucsd.edu
|
warning/0002/nlin0002027.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In a seminal paper Maillet and Sanchez de Santos revealed the uses of factorizing Drinfel’d twists for inhomogeneous statistical spin chain models for which the method of the algebraic Bethe ansatz is available. Those authors used as paradigmata of their argumentation the rational XXX and the trigonometric XXZ models being realized on tensor products of two-dimensional (fundamental) representations of the underlying group $`sl(2)`$. They showed that the similarity transformation provided by the Drinfel’d twist gives rise to a completely symmetric representation of the respective monodromy matrices and implies simplifying features in the new basis - to be described in detail below - for the various operators in the grid of the monodromy matrix.
The results of have been generalized to any finite dimensional irreducible representation of the Yangian $`Y(sl(2))`$ and have been used to achieve substantial simplifications in the calculation of form factors , in the determination of thermodynamic quantities such as the spontaneous magnetization , and to solve the so called quantum inverse problem , , that is, to express the local spin operators of the microscopic model through the operators figuring in the algebraic Bethe ansatz.
The most striking aspect of the results in is, as we think, related to the fact that no polarization clouds are attached to quasiparticle creation and annihilation operators in the basis in which the monodromy matrix is completely symmetric. This means in terms of a particle notation that no virtual particle–antiparticle pairs are present in the wave vectors generated by the action of the creation operators to the ground state (the reference state of the Bethe ansatz), or in spin chain terminology that the creation and annihilation operators are exclusively built from local spin raising and spin lowering operators respectively (that is, there are no compensating pairs of local raising and lowering spin operators). It was noted in that this latter feature underscores the neat connection between the quantum spin chain models and their respective quasiclassical limits, which are Gaudin magnets , insofar as the appearance of the quasi particle operators of the quantum models in the particular basis differ from the corresponding operators in the quasiclassical limit models only by a “diagonal dressing” (see below).
This connection motivated us to attempt a generalization of the work of Maillet and Sanchez de Santos towards models based on higher rank groups. We will deal here with the simplest conceivable extension in the form of the rational XXX model with $`sl(n)`$ as underlying group.
A notorious technical difficulty of integrable models with underlying higher rank group arises from the intricacies of the recursive procedure of the hierarchical Bethe ansatz . It has been known for some time that the recursion of the hierarchical ansatz can be resolved in the case of the quasiclassical limit of the rational models, i.e., the rational Gaudin magnets. Constructing the analogue of the factorizing twist of for higher rank models one may hope - in view of the affinity of the special basis rendered by the factorizing twist with the quasiclassical limit model - for an explicit resolution of the Bethe ansatz hierarchy. This will indeed be our main result for the spin model under consideration: an explicit representation of the $`sl(n)`$ Bethe wave vectors, solving therewith (for the wave vectors) the hierarchy.
The plan of the paper is as follows: section 2 sets the notation, section 3 is devoted to the construction of the factorizing twist. In section 4 we give the expressions for the $`sl(n)`$ generators and for the operators contained in the monodromy matrix in the basis mediated by the factorizing twist. In section 5 we discuss the resolution of the Bethe hierarchy. Section 6 contains our conclusions. Some technical details are relegated to appendices.
## 2 Basic definitions and notation
Below we shall use many of the notations of references , . We consider the $`sl(n)`$ Yangian $`R`$-matrix depending on a spectral parameter $`\lambda `$ and a quantum deformation parameter $`\eta `$:
$$R_{12}(\lambda )=b(\lambda )1\mathrm{I}_{12}+c(\lambda )P_{12}$$
(1)
where
$$b(\lambda )=\frac{\lambda }{\lambda +\eta },c(\lambda )=\frac{\eta }{\lambda +\eta }.$$
(2)
The matrix $`R_{12}`$ is meant to represent a map $`_{(1)}^n_{(2)}^n_{(1)}^n_{(2)}^n(_{(1)}^n_{(2)}^n^n)`$ and $`P_{12}`$ is the permutation operator acting in $`_{(1)}^n_{(2)}^n`$. Local spectral parameters attached to vectorspaces $`_{(i)}^n`$ isomorphic to $`^n`$ will be called $`z_i`$. We will also use the notation
$`b_{ij}=b(z_iz_j),c_{ij}=c(z_iz_j).`$ (3)
It is well known that $`R`$-matrices defined by (1) satisfy the Yang-Baxter equation in vertex form:
$$R_{12}(z_1z_2)R_{13}(z_1z_3)R_{23}(z_2z_3)=R_{23}(z_2z_3)R_{13}(z_1z_3)R_{12}(z_1z_2)$$
(4)
and the unitarity relation
$$R_{12}R_{21}=1\mathrm{I}.$$
(5)
where $`R_{ij}=R_{ij}(z_iz_j)`$ acts non-trivially on the tensor product $`_{(i)}^n_{(j)}^n`$.
Our convention for the matrix indices is as follows:
$`\left(Z\right)_{\beta \alpha }^{\gamma \delta }=\left(XY\right)_{\beta \alpha }^{\gamma \delta }=\left(X\right)_{j_1j_2}^{\gamma \delta }\left(Y\right)_{\beta \alpha }^{j_1j_2}.`$ (6)
With the notation $`T_{0,23}=R_{03}R_{02},R_{0i}R_{0i}(z_i)`$, where the index $`0`$ refers to an auxilliary space $`_{(0)}^n`$, one may rewrite Eq. (4) in the form of a Faddeev–Zamolodchikov relation
$$R_{23}^{\sigma _{23}}T_{0,23}=T_{0,32}R_{23}^{\sigma _{23}}$$
(7)
with $`\sigma _{23}`$ the transposition of space labels $`(2,3)`$.
We use here and subsequently a notation (which may not be in line with common use) that the labels in the upper row are permuted relative to lower indices according to the permutation inscribed, which reads in the example at hand as $`\left(R^{\sigma _{23}}\right)_{\beta _3\beta _2}^{\alpha _2\alpha _3}`$.
It is straightforward to generalize Eq. (7) to a N-fold tensor product of spaces:
With the definition $`T_{0,1\mathrm{}N}=R_{0N}\mathrm{}R_{01}`$ the generalization reads
$$R_{1\mathrm{}N}^\sigma T_{0,1\mathrm{}N}=T_{0,\sigma (1)\mathrm{}\sigma (N)}R_{1\mathrm{}N}^\sigma $$
(8)
where $`\sigma `$ is now an element of the symmetric group $`S_N`$ and $`R_{1\mathrm{}N}^\sigma `$ denotes a product of $`R`$-matrices occuring in (7), the product corresponding to a decomposition of $`\sigma `$ into elementary transpositions.
The order of the upper matrix indices $`\alpha _i`$ of the $`R^\sigma `$ reads according to the above prescription as follows:
$`\left(R_{1\mathrm{}N}^\sigma \right)_{\beta _N\mathrm{}\beta _1}^{\alpha _{\sigma (N)}\mathrm{}\alpha _{\sigma (1)}}.`$ (9)
Eq. (8) implies the composition law (note the difference to the composition law used in ref. )
$`R_{1\mathrm{}N}^{\sigma ^{}\sigma }=R_{\sigma ^{}(1)\mathrm{}\sigma ^{}(N)}^\sigma R_{1\mathrm{}N}^\sigma ^{}`$ (10)
for a product of two elements in $`S_N`$. The factor $`R_{\sigma (1)\mathrm{}\sigma (N)}^\sigma ^{}`$ on the r.h.s.of Eq. (10) satisfies for itself the relation
$`R_{\sigma (1)\mathrm{}\sigma (N)}^\sigma ^{}T_{0,\sigma (1)\mathrm{}\sigma (N)}=T_{0,\sigma \sigma ^{}(1)\mathrm{}\sigma \sigma ^{}(N)}R_{\sigma (1)\mathrm{}\sigma (N)}^\sigma ^{}.`$ (11)
## 3 The $`F`$-matrix and some of its properties
The starting point of paper is the Drinfel’d factorizing twists of the elementary $`sl(2)`$ $`R`$-matrix:
$$R_{12}=F_{21}^1F_{12}$$
where $`F_{12}`$ is given by the formula (90) of
$$F_{12}=\left(\begin{array}{cccc}1& \mathrm{\hspace{0.33em}0}& \mathrm{\hspace{0.33em}0}& \mathrm{\hspace{0.33em}0}\\ 0& \mathrm{\hspace{0.33em}1}& \mathrm{\hspace{0.33em}0}& \mathrm{\hspace{0.33em}0}\\ 0& c(z_1z_2)& b(z_1z_2)& 0\\ 0& \mathrm{\hspace{0.33em}0}& \mathrm{\hspace{0.33em}0}& \mathrm{\hspace{0.33em}1}\end{array}\right).$$
(12)
The generalization of this formula to the $`sl(n)`$ case is of the form
$$F_{12}=\underset{n\alpha _2\alpha _1}{}P_{\alpha _1}^1P_{\alpha _2}^2\mathrm{\hspace{0.17em}1}\mathrm{I}_{12}+\underset{n\alpha _1>\alpha _2}{}P_{\alpha _1}^1P_{\alpha _2}^2R_{12}^{\sigma _{12}}.$$
(13)
Here $`[P_\alpha ^i]_{k,l}=\delta _{k,\alpha }\delta _{l,\alpha }`$ is the projector on the $`\alpha `$ component acting in $`i`$-th space.
Generalizing this factorization matrix to the $`N`$-site problem one has to satisfy at least three properties for the $`F`$-matrix (see ,):
factorization, that is
$$F_{\sigma (1)\mathrm{}\sigma (N)}(z_{\sigma (1)},\mathrm{},z_{\sigma (N)})R_{1\mathrm{}N}^\sigma (z_1,\mathrm{},z_N)=F_{1\mathrm{}N}(z_1,\mathrm{},z_N)$$
(14)
for any permutation $`\sigma S_N`$;
lower-triangularity;
non-degeneracy.
Proposition 3.1 The following expression for the $`F`$-matrix:
$$F_{1\mathrm{}N}=\underset{\sigma S_N}{}\underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}R_{1\mathrm{}N}^\sigma (z_1,\mathrm{},z_N)$$
(15)
satisfies the properties A,B and C. The sum $`^{}`$ is to be taken over all non-decreasing sequences of the labels $`\alpha _{\sigma (i)}`$ which are increasing at places where the permuted index is decreasing ($`\sigma (i+1)<\sigma (i)`$), namely, labels $`\alpha _i`$ should satisfy one of two inequalities for each pair of neighbouring spaces labels:
$`\alpha _{\sigma (i+1)}\alpha _{\sigma (i)}\text{if}\sigma (i+1)>\sigma (i)`$ (16)
$`\alpha _{\sigma (i+1)}>\alpha _{\sigma (i)}\text{if}\sigma (i+1)<\sigma (i)`$ .
Proof First of all let us note that the lower-triangularity can be traced back to the form of the elementary $`R`$-matrix using the definition of $`F`$, Eq. (15). Indeed, the ordering (16) just corresponds to the lower-triangularity of the matrix $`F`$. Non-degeneracy follows from the lower-triangularity and the fact that all diagonal elements are non-zero. Apart from that we shall give below the explicit form of $`F^1`$.
To prove the factorization property A let us, as above, represent the arbitrary permutation $`\sigma `$ in the form the composition of $`k`$ elementary transpositions $`\sigma _i`$ i.e.
$$\sigma =\sigma _1\mathrm{}\sigma _k.$$
The important structural feature of equation (15) is that it can be decomposed stepwise into elementary transpositions:
$$F_{\sigma (1)\mathrm{}\sigma (N)}R_{1\mathrm{}N}^\sigma =$$
$$=F_{\sigma _1\sigma _2\mathrm{}\sigma _k(1,\mathrm{},N)}R_{\sigma _1\sigma _2\mathrm{}\sigma _{k1}(1,\mathrm{},N)}^{\sigma _k}R_{\sigma _1\sigma _2\mathrm{}\sigma _{k2}(1,\mathrm{},N)}^{\sigma _{k1}}\mathrm{}R_{1\mathrm{}N}^{\sigma _1}$$
$$=F_{\sigma _1\sigma _2\mathrm{}\sigma _{k1}(1,\mathrm{},N)}R_{\sigma _1\sigma _2\mathrm{}\sigma _{k2}(1,\mathrm{},N)}^{\sigma _{k1}}R_{\sigma _1\sigma _2\mathrm{}\sigma _{k3}(1,\mathrm{},N)}^{\sigma _{k2}}\mathrm{}R_{1\mathrm{}N}^{\sigma _1}$$
$$=\mathrm{}\mathrm{}\mathrm{}=F_{\sigma _1(1,\mathrm{},N)}R_{1\mathrm{}N}^{\sigma _1}=F_{1\mathrm{}N}$$
where the composition law (10) was used. So we have to prove equation (15) for elementary transpositions only.
Let $`\sigma _i`$ be the elementary transposition $`\{i,i+1\}\{i+1,i\}`$. We consider the product $`F_{1\mathrm{}i+1i\mathrm{}N}R_{1\mathrm{}N}^{\sigma _i}`$. With the help of Eq.’s (15) and (10) we obtain
$`F_{1\mathrm{}i+1i\mathrm{}N}R_{1\mathrm{}N}^{\sigma _i}`$ $`=`$ $`F_{\sigma _i(1\mathrm{}ii+1\mathrm{}N)}R_{1\mathrm{}N}^{\sigma _i}`$ (17)
$`=`$ $`{\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{\alpha _{\sigma _i\sigma (1)}\mathrm{}\alpha _{\sigma _i\sigma (N)}}{\overset{(i)}{}}}{\displaystyle \underset{j=1}{\overset{N}{}}}P_{\alpha _{\sigma _i\sigma (j)}}^{\sigma _i\sigma (j)}R_{\sigma _i(1,\mathrm{},N)}^\sigma R_{1\mathrm{}N}^{\sigma _i}`$
$`=`$ $`{\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{\alpha _{\sigma _i\sigma (1)}\mathrm{}\alpha _{\sigma _i\sigma (N)}}{\overset{(i)}{}}}{\displaystyle \underset{j=1}{\overset{N}{}}}P_{\alpha _{\sigma _i\sigma (j)}}^{\sigma _i\sigma (j)}R_{1\mathrm{}N}^{\sigma _i\sigma }`$
with the $`^{(i)}`$ being defined by the restricting conditions
$`\alpha _{\sigma _i\sigma (j+1)}\alpha _{\sigma _i\sigma (j)}\text{if}\sigma (j+1)>\sigma (j)`$ (18)
$`\alpha _{\sigma _i\sigma (j+1)}>\alpha _{\sigma _i\sigma (j)}\text{if}\sigma (j+1)<\sigma (j)`$ $`.`$
(It may be helpful to keep in mind that the ordering prescription has to be executed according to the shifted labels $`\stackrel{~}{j}=\sigma _i(j)`$.) Substituting in (17) $`\stackrel{~}{\sigma }`$ for $`\sigma _i\sigma `$ one arrives at
$`F_{1\mathrm{}i+1i\mathrm{}N}R_{1\mathrm{}N}^{\sigma _i}`$ $`=`$ $`{\displaystyle \underset{\stackrel{~}{\sigma }S_N}{}}{\displaystyle \underset{\alpha _{\stackrel{~}{\sigma }(1)}\mathrm{}\alpha _{\stackrel{~}{\sigma }(N)}}{\overset{}{}}}{\displaystyle \underset{j=1}{\overset{N}{}}}P_{\alpha _{\stackrel{~}{\sigma }(j)}}^{\stackrel{~}{\sigma }(j)}R_{1\mathrm{}N}^{\stackrel{~}{\sigma }}`$ (19)
with the defining restrictions of $`^{}`$ now of the form
$`\alpha _{\stackrel{~}{\sigma }(j+1)}\alpha _{\stackrel{~}{\sigma }(j)}\text{if}\sigma _i\stackrel{~}{\sigma }(j+1)>\sigma _i\stackrel{~}{\sigma }(j)`$
$`\alpha _{\stackrel{~}{\sigma }(j+1)}>\alpha _{\stackrel{~}{\sigma }(j)}\text{if}\sigma _i\stackrel{~}{\sigma }(j+1)<\sigma _i\stackrel{~}{\sigma }(j)`$ (20)
which has a slightly different appearance in comparison to (16). Elementary combinatorical considerations lead to the conclusion that the stipulations (16) and (20) give the same result as long as $`\sigma ^1(i)`$ and $`\sigma ^1(i+1)`$ do not happen to be on neighbouring places, that is if not
$`\sigma ^1(i)=\sigma ^1(i+1)\pm 1.`$ (21)
If (21) holds we have to appeal to the specific form of the $`R`$-matrix to complete the argument.
Comparing the r.h.s. of Eq. (19) in connection with (20) to the r.h.s. of Eq. (15) in connection with (16) one notes that a discrepancy is certainly excluded if the strict inequality is implied in the step from $`\sigma ^1(i)`$ to $`\sigma ^1(i+1)`$ (if $`\sigma ^1(i+1)`$ is larger than $`\sigma ^1(i)`$), or from $`\sigma ^1(i+1)`$ to $`\sigma ^1(i)`$ if the reversed order is assumed. But for equal group labels at the two neighbouring places in question the representation of the additional transposition of $`i`$ and $`i+1`$ in (15) as compared to (19) has no effect, since it supplies a unit factor due to the projectors.
It completes the proof of the proposition.
Remark: The most general matrix $`\stackrel{~}{F}`$ satisfying the above conditions A and C differs from the special solution of the preceding theorem by a non-degenerate, completely symmetric matrix factor ,
$`\stackrel{~}{F}_{1\mathrm{}N}(z_1,\mathrm{},z_N)`$ $`=`$ $`X_{1\mathrm{}N}(z_1,\mathrm{},z_N)F_{1\mathrm{}N}(z_1,\mathrm{},z_N),`$
$`X_{1\mathrm{}N}(z_1,\mathrm{},z_N)`$ $`=`$ $`X_{\sigma (1)\mathrm{}\sigma (N)}(z_{\sigma (1)},\mathrm{},z_{\sigma (N)})\sigma S_N.`$
Indeed it is easy to see that $`\stackrel{~}{F}`$ satisfies together with $`F`$ the factorization equation (14). Conversely, suppose that both $`F`$ and $`\stackrel{~}{F}`$ satisfy (14). It follows that
$`F_{\sigma (1)\mathrm{}\sigma (N)}^1F_{1\mathrm{}N}=\stackrel{~}{F}_{\sigma (1)\mathrm{}\sigma (N)}^1\stackrel{~}{F}_{1\mathrm{}N}`$
and therefrom
$`F_{1\mathrm{}N}\stackrel{~}{F}_{1\mathrm{}N}^1=F_{\sigma (1)\mathrm{}\sigma (N)}\stackrel{~}{F}_{\sigma (1)\mathrm{}\sigma (N)}^1.`$
Hence it follows that
$`X_{1\mathrm{}N}=F_{1\mathrm{}N}\stackrel{~}{F}_{1\mathrm{}N}^1`$
is nondegenerate and completely symmetric and transforms $`\stackrel{~}{F}`$ into $`F`$, $`X_{1\mathrm{}N}\stackrel{~}{F}_{1\mathrm{}N}=F_{1\mathrm{}N}`$.
We need furthemore the inverse operator $`F^1`$. To find its expression we have to prove the following
Proposition 3.2 The operator $`F^{}`$ defined by the formula
$$F_{1\mathrm{}N}^{}=\underset{\sigma S_N}{}\underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}R_{1\mathrm{}N}^{(t)\sigma }(z_1,\mathrm{},z_N)\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}$$
(22)
with the shorthand notation
$`R_{1\mathrm{}N}^{(t)\sigma }R_{\sigma (1,\mathrm{},N)}^{\sigma ^1}`$ (23)
and $`^{}`$ is taken over all possible $`\alpha _i`$ which satisfy one of two inequalities for each neighbouring pair of spaces $`i`$ and $`i+1`$:
$`\alpha _{\sigma (i+1)}\alpha _{\sigma (i)}\text{if}\sigma (i+1)<\sigma (i)`$
$`\alpha _{\sigma (i+1)}<\alpha _{\sigma (i)}\text{if}\sigma (i+1)>\sigma (i)`$ (24)
satisfy the relation
$$F_{1\mathrm{}N}F_{1\mathrm{}N}^{}=\underset{i<j}{}\mathrm{\Delta }_{ij}$$
(25)
where the diagonal matrix
$$[\mathrm{\Delta }_{ij}]_{\alpha _i,\alpha _j}^{\beta _i,\beta _j}=\delta _{\alpha _i\beta _i}\delta _{\alpha _j\beta _j}\{\begin{array}{cc}1\text{if}\alpha _i=\alpha _j,\hfill & \\ b_{ij}\text{if}\alpha _i>\alpha _j,\hfill & \\ b_{ji}\text{if}\alpha _j>\alpha _i\hfill & \end{array}$$
(26)
acts in the pair of spaces $`i`$ and $`j`$.
Proof Taking into account the conditions (16) and (24) in sums $`^{}`$ and $`^{}`$ of the expressions (15) and (22) respectively one can write down the expression for the product $`F_{1\mathrm{}N}F_{1\mathrm{}N}^{}`$ in the following form:
$$F_{1\mathrm{}N}F_{1\mathrm{}N}^{}=\underset{\sigma S_N}{}\underset{\sigma ^{}S_N}{}\underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}\underset{\beta _{\sigma ^{}(1)}\mathrm{}\beta _{\sigma ^{}(N)}}{\overset{}{}}\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}R_{1\mathrm{}N}^\sigma R_{\sigma ^{}(1)\mathrm{}\sigma ^{}(N)}^{\sigma ^1}\underset{i=1}{\overset{N}{}}P_{\beta _{\sigma ^{}(i)}}^{\sigma ^{}(i)}$$
$$=\underset{\sigma S_N}{}\underset{\sigma ^{}S_N}{}\underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}\underset{\beta _{\sigma ^{}(1)}\mathrm{}\beta _{\sigma ^{}(N)}}{\overset{}{}}\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}R_{\sigma ^{}(1,\mathrm{},N)}^{\sigma _{}^{}{}_{}{}^{1}\sigma }\underset{i=1}{\overset{N}{}}P_{\beta _{\sigma ^{}(i)}}^{\sigma ^{}(i)}$$
(27)
$$=\underset{\sigma S_N}{}\underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}R_{\sigma (N,\mathrm{},1)}^{\overline{\sigma }}\underset{i=1}{\overset{N}{}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}$$
(28)
where the permutation $`\overline{\sigma }`$ reverses the order of the labels:
$$\overline{\sigma }(1,\mathrm{},N)=(N,\mathrm{},1).$$
In the line above (27) we have inserted the definitions of $`F`$ and $`F^{}`$, Eq’s (15) and (22) resp. Equality (27) is obtained by applying the compositon rule (10). To prove equality (28) we note first of all that any matrix $`R^\sigma `$ provides maps s.t. the sets of $`sl(n)`$ labels of the incoming and outgoing states are connected by a permutation. (This property is easily verified for matrices $`R^\sigma `$ corresponding to elementary transpositions and it is preserved under the composition of several transpositons.) But the labels $`\left\{\alpha _{\sigma (i)}\right\}`$ represent according to the prescription (16) a non-decreasing series (in ($`i`$)) of labels whereas the $`\left\{\beta _{\sigma ^{}(i)}\right\}`$ \- being related to $`\left\{\alpha _{\sigma (i)}\right\}`$ by a permutation - are according to (24) a non-increasing series. For these two requirements to be fulfilled the equalities
$`\beta _{\sigma ^{}(N)}=\alpha _{\sigma (1)},\mathrm{},\beta _{\sigma ^{}(1)}=\alpha _{\sigma (N)}`$ (29)
are a necessity.
Let us assume momentarily that all the labels $`\beta _{\sigma ^{}(i)}`$ (and hence the $`\alpha _{\sigma (i)}`$) are different from each other. We want to show that Eq. (29) implies the equality
$`\sigma \overline{\sigma }=\sigma ^{}`$ (30)
for the matrix element
$`\left(R_{\sigma ^{}(1,\mathrm{},N)}^{\sigma _{}^{}{}_{}{}^{1}\sigma }\right)_{\beta _{\sigma ^{}(N)}\mathrm{}\beta _{\sigma ^{}(1)}}^{\alpha _{\sigma (N)}\mathrm{}\alpha _{\sigma (1)}}`$
to be non-vanishing. Viewing $`R_{\sigma ^{}(1,\mathrm{},N)}^{\sigma _{}^{}{}_{}{}^{1}\sigma }`$ as a product of elementary $`R`$-matrices one observes that the group label $`\beta _{\sigma ^{}(N)}=\alpha _{\sigma (1)}`$ can be transported from the lower left corner to the upper right place only if the space labels $`\sigma ^{}(N)`$ and $`\sigma (1)`$ are identical. Assume to the contrary that $`\sigma ^{}(N)`$ is identical to some other element $`\sigma (x)\sigma (1)`$. It would follow that the group label $`\beta _{\sigma ^{}(N)}`$ could appear in the upper row only at the place with space label $`\sigma (x)`$ or on the l.h.s. of it. (This restriction on the flow of group labels is a straightforward consequence of the form of the elementary $`R`$-matrix, Eq. (1).) We conclude that we have indeed to identify $`\sigma ^{}(N)`$ with $`\sigma (1)`$ to obtain a non-vanishing matrix element of $`R`$. The identification of $`\sigma ^{}(N1)`$ with $`\sigma (2)`$ etc. follows analogously and therefrom Eq. (30).
A glance on (16) and (24) affirms that Eq. (30) remains valid under general circumstances, i.e., if some group lables $`\beta _{\sigma ^{}(i)}`$ and therefore $`\alpha _{\sigma (j)}`$ occur repeatedly, since the order of the space labels attached to the same group label is uniquely specified by these prescriptions.
One deduces from (28) that $`FF^{}`$ is a diagonal matrix. A simple calculation leads to the expression for the diagonal elements quoted in Eq. (26). (The product appearing on the r.h.s. of Eq. (25) is related to $`\overline{\sigma }`$ as the latter is a maximal element of $`S_N`$ and as such is representable as a product of $`N(N1)/2`$ elementary transpositions. Each transposition is reflected in one factor of the product in Eq. (25).)
This completes the proof of proposition 3.2.
We get from the formula (25) the expression for $`F_{1\mathrm{}N}^1`$:
$$F_{1\mathrm{}N}^1=F_{1\mathrm{}N}^{}\underset{i<j}{}\mathrm{\Delta }_{ij}^1.$$
(31)
For the case of the $`sl(2)`$ Yangian the formula (31) corresponds to the result of proposition 4.6 of .
## 4 $`sl(n)`$ generators and the monodromy matrix in the F-basis
We will first determine the simple root $`sl(n)`$ generators $`\stackrel{~}{E}_{\alpha ,\alpha \pm 1}=F_{1\mathrm{}N}E_{\alpha \pm \alpha +1}F_{1\mathrm{}N}^1`$ and the element $`\stackrel{~}{T}_{nn}=F_{1\mathrm{}N}T_{nn}F_{1\mathrm{}N}^1`$ of the monodromy matrix. The remaining $`sl(n)`$ generators can then be obtained from the simple ones through multiple commutators. The examination of the full algebra can be found in Appendix A. One may exploit the $`sl(n)`$ invariance of the monodromy matrix (with respect to its combined action in the quantum spaces and the auxiliary space, see e.g. ) to derive expressions for all elements $`\stackrel{~}{T}_{\alpha \beta }`$ given $`\stackrel{~}{T}_{nn}`$ and the $`sl(n)`$ generators.
One has in particular the relation
$`\stackrel{~}{T}_{n\alpha }=[\stackrel{~}{E}_{\alpha ,n},\stackrel{~}{T}_{nn}].`$ (32)
The l.h.s. of the latter equation originates from the action of the $`sl(n)`$ generator in the auxiliary space whereas the r.h.s. evidently reflects the corresponding action in the quantum space.
We will content ourselves to derive the explicit form of the $`\stackrel{~}{T}_{n\alpha }`$ using Eq. (32), since this is all we need to build $`sl(n)`$ Bethe wave vectors.
The simple root generators in the new basis differ from those in the original basis by a diagonal dressing factor. We have
Proposition 4.1
$$\stackrel{~}{E}_{\gamma ,\gamma \pm 1}=\underset{i=1}{\overset{N}{}}E_{\gamma ,\gamma \pm 1}^{(i)}_{ji}G^{\pm \gamma }(i,j)_{[j]}$$
(33)
where
$$G^\gamma (i,j)_{k,l}=\delta _{kl}\{\begin{array}{cc}b_{ij}^1\text{if}k=\gamma ,\hfill & \\ 1\text{otherwise}\hfill & \end{array}$$
$$G^\gamma (i,j)_{k,l}=\delta _{kl}\{\begin{array}{cc}b_{ji}^1\text{if}k=\gamma +1\hfill & \\ 1\text{otherwise}.\hfill & \end{array}$$
(34)
Proof Eq’s (33) and (34) specialized to the rational $`sl(2)`$ case have been presented in propositions 5.1 and 5.2 of ref. . The proof of these equations for the $`sl(n)`$ model with arbitrary $`n`$ can be reduced to that of the $`sl(2)`$ model. One has to note for this purpose that one obtains due to the $`sl(n)`$ invariance of the elementary R-matrices the vanishing result
$`[R_{1\mathrm{}N}^\sigma ,E_{\alpha ,\alpha \pm 1}]=0`$ (35)
for any permutation $`\sigma S_N`$.
This allows us to write (cf. Eq. (27))
$`\stackrel{~}{E}_{\gamma ,\gamma \pm 1}={\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{\sigma ^{}S_N}{}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}}{\displaystyle \underset{\beta _{\sigma ^{}(1)}\mathrm{}\beta _{\sigma ^{}(N)}}{\overset{}{}}}{\displaystyle \underset{i=1}{\overset{N}{}}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}E_{\gamma ,\gamma \pm 1}R_{\sigma ^{}(1,\mathrm{},N)}^{\sigma _{}^{}{}_{}{}^{1}\sigma }{\displaystyle \underset{i=1}{\overset{N}{}}}P_{\beta _{\sigma ^{}(i)}}^{\sigma ^{}(i)}{\displaystyle \underset{i<j}{}}\mathrm{\Delta }_{ij}.`$ (36)
The collapse of the double sum $`_{\sigma ,\sigma ^{}}`$ into a single sum proceeds here along the same pattern as above (in the transition from Eq. (27) to Eq. (28)). One further has to note that group indices $`\gamma `$ and $`(\gamma +1)`$ ($`(\gamma 1)`$ resp.) only occur in neighbouring positions what concerns ingoing and outgoing matrix indices because of the monotonicity prescription incorporated into the sums $`^{}`$ and $`^{}`$ resp. The rearrangement of the neighbouring labels $`\gamma `$ and $`(\gamma +1)`$ ($`(\gamma 1)`$ resp.) goes on according to $`sl(2)`$ rules and produces the result quoted in Eq. (34) and in ref. . Rearrangements involving group indices different from $`\gamma `$ and $`(\gamma +1)`$ ($`(\gamma 1)`$ resp.) are not affected by the presence of the generator $`E_{\gamma ,\gamma \pm 1}`$, since for those rearrangements the difference of $`\gamma `$ and $`(\gamma +1)`$ ($`(\gamma 1)`$ resp.) is immaterial.
Proposition 4.2
$$\stackrel{~}{T}_{nn}(\lambda )=_{i=1}^N\text{diag}\{b(\lambda z_i),\mathrm{},b(\lambda z_i),1\}.$$
(37)
Proof Let us consider the action of the matrix $`F`$ on $`T_{nn}`$
$`F_{1\mathrm{}N}T_{nn}`$ $`=`$ $`{\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}}{\displaystyle \underset{i=1}{\overset{N}{}}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}R_{1\mathrm{}N}^\sigma P_n^0T_{0,1\mathrm{}N}P_n^0`$ (38)
$`=`$ $`{\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{}{}}}{\displaystyle \underset{i=1}{\overset{N}{}}}P_{\alpha _{\sigma (i)}}^{\sigma (i)}P_n^0T_{0,\sigma (1)\mathrm{}\sigma (N)}P_n^0R_{1\mathrm{}N}^\sigma .`$
The specialization to the entry $`(n,n)`$ of the auxiliary space is here achieved by the projectors $`P_n^0`$. For the second equality in (38) we have used relation (8) and the obvious fact that $`P_n^0`$ commutes with $`R_{1\mathrm{}N}^\sigma `$. To simplify the following argument we distinguish in the sum $`^{}`$ cases of various multiplicities of the occurence of the group index $`n`$:
$`F_{1\mathrm{}N}T_{nn}`$ $`=`$ $`{\displaystyle \underset{\sigma S_N}{}}{\displaystyle \underset{k=0}{\overset{N}{}}}{\displaystyle \underset{\alpha _{\sigma (1)}\mathrm{}\alpha _{\sigma (N)}}{\overset{^{}}{}}}{\displaystyle \underset{j=Nk+1}{\overset{N}{}}}\delta _{\alpha _{\sigma (j)},n}{\displaystyle \underset{j=1}{\overset{Nk}{}}}P_{\alpha _{\sigma (j)}}^{\sigma (j)}P_n^0T_{0,\sigma (1)\mathrm{}\sigma (N)}P_n^0R_{1\mathrm{}N}^\sigma .`$ (39)
Let us consider the prefactor of $`R_{1\mathrm{}N}^\sigma `$ on the r.h.s. of Eq. (39) more closely. Using specific features of the $`R`$-matrices we can rewrite it as follows:
$`{\displaystyle \underset{j=1}{\overset{Nk}{}}}P_{\alpha _{\sigma (j)}}^{\sigma (j)}{\displaystyle \underset{j=Nk+1}{\overset{N}{}}}P_n^{\sigma (j)}P_n^0T_{0,\sigma (1)\mathrm{}\sigma (N)}P_n^0`$ (40)
$`=`$ $`{\displaystyle \underset{j=1}{\overset{Nk}{}}}P_{\alpha _{\sigma (j)}}^{\sigma (j)}\left(R_{0,\sigma (N)}\right)_{nn}^{nn}\left(R_{0,\sigma (N1)}\right)_{nn}^{nn}\mathrm{}\left(R_{0,\sigma (Nk+1)}\right)_{nn}^{nn}P_n^0T_{0,\sigma (1)\mathrm{}\sigma (Nk)}P_n^0{\displaystyle \underset{j=Nk+1}{\overset{N}{}}}P_n^{\sigma (j)}`$
$`=`$ $`{\displaystyle \underset{j=1}{\overset{Nk}{}}}P_{\alpha _{\sigma (j)}}^{\sigma (j)}P_n^0T_{0,\sigma (1)\mathrm{}\sigma (Nk)}P_n^0{\displaystyle \underset{j=Nk+1}{\overset{N}{}}}P_n^{\sigma (j)}`$
$`=`$ $`{\displaystyle \underset{i=1}{\overset{Nk}{}}}\left(R_{0i}\right)_{n,\alpha _{\sigma (i)}}^{n,\alpha _{\sigma (i)}}{\displaystyle \underset{j=1}{\overset{Nk}{}}}P_{\alpha _{\sigma (j)}}^{\sigma (j)}{\displaystyle \underset{j=Nk+1}{\overset{N}{}}}P_n^{\sigma (j)}`$
Inserting the r.h.s. of (40) into Eq. (39) one sees that the product $`_i\left(R_{0i}\right)_{n,\alpha _{\sigma (i)}}^{n,\alpha _{\sigma (i)}}`$ provides the desired diagonal dressing factor of $`T_{nn}`$ and the product of projectors applied to $`R^\sigma `$ restores $`F_{1\mathrm{}N}`$.
This completes the proof of proposition 4.2.
Given the simple root generators $`\stackrel{~}{E}_{\alpha ,\alpha \pm 1}`$ it is a straightforward task to evaluate the generators corresponding to non-simple roots.
One finds in particular
$`\stackrel{~}{E}_{n\alpha ,n}={\displaystyle \underset{k=1}{\overset{\alpha }{}}}{\displaystyle \underset{i_1\mathrm{}i_k}{}}{\displaystyle \underset{\gamma =1}{\overset{k1}{}}}{\displaystyle \frac{\eta }{z_{i_\gamma }z_{i_{\gamma +1}}}}{\displaystyle \underset{\alpha =\beta _0>\beta _1\mathrm{}>\beta _k=0}{}}_{l=1}^kE_{n\beta _{l1},n\beta _l}^{(i_l)}_{ji_1\mathrm{}i_k}\mathrm{\Gamma }_{j;\underset{\beta _0\beta _1}{\underset{}{i_1..i_1}}\underset{\beta _1\beta _2}{\underset{}{i_2..i_2}}\mathrm{}\underset{\beta _{k1}\beta _k}{\underset{}{i_k..i_k}}}^{(j)}`$ (41)
where $`\mathrm{\Gamma }_{j;i_1,\mathrm{},i_\alpha }=\text{diag}\{1,\mathrm{},1,b_{i_1j}^1,\mathrm{},b_{i_\alpha j}^1,1\}`$.
Exploiting the last equation and Eq. (32) one finally arrives at
$`\stackrel{~}{T}_{nn\alpha }={\displaystyle \underset{k=1}{\overset{\alpha }{}}}{\displaystyle \underset{i_1\mathrm{}i_k}{}}c(\lambda z_{i_k}){\displaystyle \underset{\gamma =1}{\overset{k1}{}}}{\displaystyle \frac{\eta }{z_{i_\gamma }z_{i_{\gamma +1}}}}b(\lambda z_{i_\gamma })\times `$
$`{\displaystyle \underset{\alpha =\beta _0>\beta _1\mathrm{}>\beta _k=0}{}}_{l=1}^kE_{n\beta _{l1},n\beta _l}^{(i_l)}_{ji_1\mathrm{}i_k}\mathrm{\Delta }_{j;\underset{\beta _0\beta _1}{\underset{}{i_1..i_1}}\underset{\beta _1\beta _2}{\underset{}{i_2..i_2}}\mathrm{}\underset{\beta _{k1}\beta _k}{\underset{}{i_k..i_k}}}^{(j)}`$ (42)
where $`\mathrm{\Delta }_{j;i_1,\mathrm{},i_\alpha }^{(j)}=\text{diag}\{b(\lambda z_j),\mathrm{},b(\lambda z_j),b(\lambda z_j)b_{i_1j}^1,\mathrm{},b(\lambda z_j)b_{i_\alpha j}^1,1\}`$ is a diagonal dressing matrix acting in $`j`$-th space.
## 5 Bethe wave vectors
What concerns the description of the hierarchical Bethe ansatz we will be rather sketchy, referring for more details to and .
The operators $`T_{n\alpha }(\lambda )(1\alpha <n1)`$ serve in the $`sl(n)`$ problem as quasiparticle creation operators and the corresponding operators $`T_{\alpha n}(\lambda )`$ have the role of annihilation operators.
The $`T_{n\alpha }(\lambda )`$ satisfy the Faddeev–Zamolodchikov algebra
$$[T_{n\alpha }(\lambda _1),T_{n\alpha }(\lambda _2)]=0$$
$$T_{n\alpha }(\lambda _1)T_{n\beta }(\lambda _2)=\frac{1}{b(\lambda _2\lambda _1)}T_{n\beta }(\lambda _2)T_{n\alpha }(\lambda _1)\frac{c(\lambda _2\lambda _1)}{b(\lambda _2\lambda _1)}T_{n\beta }(\lambda _1)T_{n\alpha }(\lambda _2)$$
(43)
where in the last relation $`\alpha \beta `$.
An ansatz for a Bethe vector $`\mathrm{\Psi }_n`$ is given in terms of a linear superposition of products of operators $`T_{n\alpha }`$ acting on a reference state $`\mathrm{\Omega }_N^{(n)}`$:
$$\mathrm{\Psi }_n(N;\lambda _1,\mathrm{},\lambda _p)=\underset{\alpha _1,\mathrm{},\alpha _p}{}\mathrm{\Phi }_{\alpha _1,\mathrm{},\alpha _p}T_{n\alpha _1}(\lambda _1)\mathrm{}T_{n\alpha _p}(\lambda _k)\mathrm{\Omega }_N^{(n)}$$
(44)
where the reference state $`\mathrm{\Omega }_N^{(n)}`$ is constituted as a $`N`$-fold tensor product of lowest weight states $`v_n^{(i)}`$ in $`_n^{(i)}`$
$`\mathrm{\Omega }_N=_{i=1}^Nv_n^{(i)}`$
and the $`\mathrm{\Phi }_{\alpha _1,\mathrm{},\alpha _p}`$ denote some c-number coefficients.
It is important to note that the reference state is invariant under the $`F`$-transformation:
$`F\mathrm{\Omega }_N^{(n)}=\mathrm{\Omega }_N^{(n)}`$
since it is immediate from the definition (15) of $`F`$ that from the sum over the permutation group only the term with the unit element inscribed gives a non-vanishing result when applied to $`\mathrm{\Omega }_N^{(n)}`$.
It can be shown, , , that $`\mathrm{\Psi }_n`$ is eigenvector of the transfer matrix $`t(\lambda )=_iT_{ii}(\lambda )`$ if
$`i)`$ the parameters $`\lambda _1,\mathrm{},\lambda _p`$ satisfy a certain system of rational equations, the famous Bethe ansatz equations
and if
$`ii)`$ the c-number coefficients are chosen s.t. they constitute the components of a rational $`sl(n1)`$ transfer matrix.
One establishes therewith a recursive procedure leading finally to a $`sl(2)`$ eigenvalue problem. We will keep the spectral parameters arising in the various stages of the procedure in general position instead of specializing them to solutions of the Bethe ansatz equations. We keep in other words the Bethe vector “off-shell” . Our goal in this paper is to figure out the functional form of the Bethe wave vectors.
To start with we recall the form of the $`sl(2)`$ wave vectors in the basis provided by Maillet and Sanchez de Santos . The creation operators with respect to the lowest weight reference state (in the special basis) are of the form
$$\stackrel{~}{T}_{21}(\lambda )=\underset{i=1}{\overset{N}{}}c(\lambda z_i)\sigma _+^{(i)}_{ji}\left(\begin{array}{cc}b(\lambda z_j)b_{ij}^1& 0\\ 0& 1\end{array}\right)_{[j]}$$
(45)
The ensuing Bethe wave vectors are given by
$`\mathrm{\Psi }_2(N;\lambda _1,\mathrm{},\lambda _p)`$ $`=`$ $`\stackrel{~}{T}_{21}(\lambda _1)\mathrm{}\stackrel{~}{T}_{21}(\lambda _p)\mathrm{\Omega }_N^{(2)}`$ (46)
$`=`$ $`{\displaystyle \underset{i_1\mathrm{}i_p}{}}B_p^{(2)}(\lambda _1,\mathrm{},\lambda _p|z_{i_1},\mathrm{},z_{i_p})\sigma _+^{(i_1)}\mathrm{}\sigma _+^{(i_p)}\mathrm{\Omega }_N^{(2)}.`$
The c-number coefficients $`B^{(2)}(\{\lambda _i\}|\{z_i\})`$ of the last equation can easily be worked out - taking into account the “diagonal dressing” factors of the spin raising operators $`\sigma _+^i`$ in (45) - to be of the form
$`B_p^{(2)}(\lambda _1,\mathrm{},\lambda _p|z_1,\mathrm{},z_p)`$ $`=`$ $`{\displaystyle \underset{\sigma S_p}{}}{\displaystyle \underset{m=1}{\overset{p}{}}}c(\lambda _mz_{\sigma (m)}){\displaystyle \underset{l=m+1}{\overset{p}{}}}{\displaystyle \frac{b(\lambda _mz_{\sigma (l)})}{b(z_{\sigma (m)}z_{\sigma (l)})}}.`$ (47)
A concise alternative representation of the coefficients $`B_p^{(2)}`$ has been derived in :
$$B_p^{(2)}(\lambda _1,\mathrm{},\lambda _p|z_{i_1},\mathrm{},z_{i_p})=\frac{\underset{i,j}{}(\lambda _iz_j)}{_{i>j}(\lambda _i\lambda _j)(z_jz_i)}\underset{<i,j>}{det}(\frac{1}{\lambda _iz_j}\frac{1}{\lambda _iz_j+\eta }).$$
(48)
The vectors $`\mathrm{\Psi }_p^{(2)}(N;\lambda _1,\mathrm{},\lambda _p)`$ are invariant under arbitrary exchanges of the variables $`\lambda _1,\mathrm{},\lambda _p`$ since operators $`\stackrel{~}{T}_{21}`$ with different values of the attached spectral parameters do commute with each other.
It has been shown in , that this symmetric appearance of the spectral parameters in the wave vectors $`\mathrm{\Psi }^{(n)},n>2`$ persists - despite of the Faddeev-Zamolodchikov relations (43) - under the assumption that the coefficients $`\mathrm{\Phi }_{\alpha _1,\mathrm{},\alpha _p}`$ in (44) are components of a $`sl(n1)`$ Bethe wave vector. Our argumentation below will heavily rely on this exchange symmetry.
We discuss now the $`sl(3)`$ model. The generalization to $`sl(n),n>3`$ will afterwards be rather obvious. Eq. (42) specialized to the case of $`sl(3)`$ renders the creation operators in the $`F`$-basis as
$`\stackrel{~}{T}_{32}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}c(\lambda z_i)E_{23}^{(i)}_{ji}\text{diag}\{b(\lambda z_j),b(\lambda z_j)b_{ij}^1,1\}_{[j]}`$ (49)
$`\stackrel{~}{T}_{31}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}c(\lambda z_i)E_{13}^{(i)}_{ji}\text{diag}\{b(\lambda z_j)b_{ij}^1,b(\lambda z_j)b_{ij}^1,1\}_{[j]}+`$
$`{\displaystyle \underset{ij}{}}c(\lambda z_i)`$ $`b(\lambda z_j){\displaystyle \frac{\eta }{z_iz_j}}E_{23}^{(i)}E_{12}^{(j)}_{ki,j}\text{diag}\{b(\lambda z_k)b_{jk}^1,b(\lambda z_k)b_{ik}^1,1\}_{[k]}.`$ (50)
The strategy employed in determining the form of the Bethe wave vector (44) will be as follows:
– We select a particular order in which the operators $`T_{n\alpha }`$ act on the reference state s.t. the eventual explicit evaluation becomes as simple as possible. (This particular order can always be achieved by the use of the Faddeev–Zamolodchikov relations (43).)
– The c-number coefficient $`\mathrm{\Phi }^{(2)}`$ has to be taken in the original basis and not in the $`F`$ basis, but fortunately a particular coefficient in the sum (44) (specialized to $`sl(3)`$) is invariant under the similarity transformation induced by the $`F`$-matrices. This enables one to compute the explicit form of this special coefficient $`\mathrm{\Phi }^{(2)}`$ using the result (47) and relate it to the order of operators alluded to in the preceding point by an appropriate factor $`_{ij}b^1(\lambda _i\mu _j)`$.
– One uses the permutation symmetry to determine all other terms in the sum.
Following this line of thought we arrive at the following
Proposition 5.1
$`\stackrel{~}{\mathrm{\Psi }}_3(N,\lambda _1,\mathrm{},\lambda _{p_0};\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1})=`$
$`{\displaystyle \underset{\sigma S_{p_0}}{}}B_{p_1}^{(2)}(\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1}|\lambda _{\sigma (1)},\mathrm{},\lambda _{\sigma (p_1)}){\displaystyle \underset{k=1}{\overset{p_1}{}}}{\displaystyle \underset{l=p_1+1}{\overset{p_0}{}}}b(\lambda _{\sigma (k)}\lambda _{\sigma (l)})^1`$
$`\stackrel{~}{T}_{32}(\lambda _{\sigma (p_1+1)})\mathrm{}\stackrel{~}{T}_{32}(\lambda _{\sigma (p_0)})\stackrel{~}{T}_{31}(\lambda _{\sigma (1)})\mathrm{}\stackrel{~}{T}_{31}(\lambda _{\sigma (p_1)})\mathrm{\Omega }_N^{(3)}`$ (51)
Proof The proof of this formula procceeds as mentioned above:
We have specialized the form of the ansatz in Eq. (51) as compared to Eq. (44) so that operators $`\stackrel{~}{T}_{32}`$ are placed to the left of all operators $`\stackrel{~}{T}_{31}`$. The latter order can be achieved by moving the operators $`\stackrel{~}{T}_{32}`$ in the general ansatz (44) to the wanted position with the help of the Faddeev–Zamolodchikov relations (43). Let us consider in particular the vector contributing in (44) of the form
$`\stackrel{~}{T}_{31}(\lambda _1)\mathrm{}\stackrel{~}{T}_{31}(\lambda _{p_1})\stackrel{~}{T}_{32}(\lambda _{p_1+1})\mathrm{}\stackrel{~}{T}_{32}(\lambda _{p_0})\mathrm{\Omega }_N^{(3)}`$ (52)
and let us relate it to the vector contributing in (51) of the form
$`\stackrel{~}{T}_{32}(\lambda _{p_1+1})\mathrm{}\stackrel{~}{T}_{32}(\lambda _{p_0})\stackrel{~}{T}_{31}(\lambda _1)\mathrm{}\stackrel{~}{T}_{31}(\lambda _{p_1})\mathrm{\Omega }_N^{(3)}.`$ (53)
A diligent appreciation of the Faddeev–Zamolodchikov relations leads to the conclusion that (53) has its unique origin in (52) and that moreover only the first term on the r.h.s. of (43) supplies contributions in the transition from (52) to (53). It follows that the transition from (52) to (53) is accompanied by an additional factor
$`{\displaystyle \underset{x=1}{\overset{p_1}{}}}{\displaystyle \underset{y=p_1+1}{\overset{p_0}{}}}{\displaystyle \frac{1}{b(\lambda _x\lambda _y)}}`$ (54)
The c-number coefficients $`\mathrm{\Phi }_{\alpha _1\mathrm{}\alpha _p}`$ in (44) (when specialized to the case $`n=3`$) refer to a $`sl(2)`$ Bethe wave vector in the familiar basis commonly used for the algebraic Bethe ansatz - not the one of Maillet and Sanchez de Santos.
But we want to argue that the special coefficient $`\mathrm{\Phi }_{1\mathrm{}12\mathrm{}2}^{(2)}`$ (the factor which accompanies the vector (53)) is in fact the same in both frames. One has to note first that the similarity transformation by the $`F`$-matrices (specialized to the case of $`sl(3)`$) respects the $`sl(2)`$ structure. This means among other things that components only with the same number of labels $`1`$ and $`2`$ are related to each other through the similarity transformation. One has secondly to observe that in the transformation of $`\mathrm{\Phi }_{1\mathrm{}12\mathrm{}2}^{(2)}`$ no other components with a different order of labels can appear due to the lower triangularity of $`F`$. (The matrix $`F`$ would otherwise not be lower triangular).
One finds thirdly through a direct examination of the definition of F that its diagonal elements relating the coefficients $`\mathrm{\Phi }_{1\mathrm{}12\mathrm{}2}^{(2)}`$ in the two frames to each other is equal to unity. Therefore we know the coefficient $`\mathrm{\Phi }_{1\mathrm{}12\mathrm{}2}^{(2)}`$ to be of the Maillet – Sanchez de Santos form.
Invoking the above mentioned exchange symmetry one completes the proof.
The expression (51) for $`\stackrel{~}{\mathrm{\Psi }}_3`$ can be worked out further by inserting the definitions (50) and (49) of $`\stackrel{~}{T}_{31}`$ and $`\stackrel{~}{T}_{32}`$ resp. to yield
$$\stackrel{~}{\mathrm{\Psi }}_3(N,\lambda _1,\mathrm{},\lambda _{p_0};\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1})=\underset{i_1\mathrm{}i_{p_0}}{}B_{p_0,p_1}^{(3)}(\lambda _1,\mathrm{},\lambda _{p_0};\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1}|z_{i_1},\mathrm{},z_{i_{p_0}})$$
$$E_{23}^{(i_{p_1+1})}\mathrm{}E_{23}^{(i_{p_0})}E_{13}^{(i_1)}\mathrm{}E_{13}^{(i_{p_1})}\mathrm{\Omega }_N^{(3)}$$
(55)
The order of operators adopted in Eq. (51) yields the bonus that the second term on the r.h.s. (the twofold sums) of (50) do not appear in (55), since those drop out if applied to the reference state $`\mathrm{\Omega }_N`$.
The sets of operators $`\stackrel{~}{T}_{31}`$ and $`\stackrel{~}{T}_{32}`$ generate through their respective diagonal dressing the structure of two $`sl(2)`$ wave vectors together with a factor which accounts for the way the operators $`\stackrel{~}{T}_{32}`$ respond to operators $`\stackrel{~}{T}_{31}`$ on their right hand side (cf. Eq. (52)).
This completes our goal to reduce the $`sl(3)`$ Bethe wave vectors to $`sl(2)`$ structures:
$$B_{p_0,p_1}^{(3)}(\lambda _1,\mathrm{},\lambda _{p_0};\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1}|z_{i_1},\mathrm{},z_{i_{p_0}})=$$
$$\underset{\sigma S_{p_0}}{}\underset{k=1}{\overset{p_1}{}}\underset{l=p_1+1}{\overset{p_0}{}}\frac{b(\lambda _{\sigma (l)}z_{i_k})}{b(\lambda _{\sigma (k)}\lambda _{\sigma (l)})}B_{p_0p_1}^{(2)}(\lambda _{\sigma (p_1+1)},\mathrm{},\lambda _{\sigma (p_0)}|z_{i_{p_1+1}},\mathrm{},z_{i_{p_0}})$$
$$B_{p_1}^{(2)}(\lambda _{p_0+1},\mathrm{},\lambda _{p_0+p_1}|\lambda _{\sigma (1)},\mathrm{},\lambda _{\sigma (p_1)})B_{p_1}^{(2)}(\lambda _{\sigma (1)},\mathrm{},\lambda _{\sigma (p_1)}|z_{i_1},\mathrm{},z_{i_{p_1}})$$
(56)
All ingredients of our argumentation for the case of $`sl(3)`$ can be straightforwardly generalized to $`sl(n);n>3`$.
We collect all operators $`\stackrel{~}{T}_{nn\alpha }`$ to the left of operators $`\stackrel{~}{T}_{nn\beta }`$ if $`\alpha <\beta `$. Once again only the first term in the expression (42) of the respective operators $`\stackrel{~}{T}_{nni}`$ contributes in this special ordering.
The wave function $`\stackrel{~}{\mathrm{\Psi }}_n`$ is then expressed in analogy to Eq. (55) by:
$$\stackrel{~}{\mathrm{\Psi }}_n(N,p_0,p_1,\mathrm{},p_{n2})=$$
$$\underset{i_1\mathrm{}i_{p_0}}{}B_{p_0,p_1,\mathrm{},p_{n2}}^{(n)}(\lambda _1,\mathrm{},\lambda _{p_0+\mathrm{}p_{n2}}|z_{i_1},\mathrm{},z_{i_{p_0}})\underset{\alpha =1}{\overset{n1}{}}\underset{j=p_\alpha +1}{\overset{p_{\alpha 1}}{}}E_{n\alpha n}^{(i_j)}\mathrm{\Omega }_N^{(n)}$$
(57)
with the following recursion relation for the function $`B^{(n)}`$:
$$B_{p_0p_1\mathrm{}p_{n2}}^{(n)}(\lambda _1,\mathrm{},\lambda _{p_0+p_1+\mathrm{}p_{n2}}|z_1,\mathrm{},z_{p_0})$$
$$=\underset{\sigma S_{p_0}}{}\underset{\alpha =1}{\overset{n2}{}}\underset{k_\alpha =p_{\alpha +1}+1}{\overset{p_\alpha }{}}\underset{l_\alpha =p_\alpha +1}{\overset{p_0}{}}\frac{b(\lambda _{\sigma (l_\alpha )}z_{k_\alpha })}{b(\lambda _{\sigma (k_\alpha )}\lambda _{\sigma (l_\alpha )})}$$
$$\underset{\gamma =0}{\overset{n2}{}}B_{p_\gamma p_{\gamma +1}}^{(2)}(\lambda _{\sigma (p_{\gamma +1}+1)}\mathrm{}\lambda _{\sigma (p_\gamma )}|z_{p_{\gamma +1}+1}\mathrm{}z_{p_\gamma })$$
$$B_{p_1\mathrm{}p_{n2}}^{(n1)}(\lambda _{p_0+1}\mathrm{}\lambda _{p_0+p_1+\mathrm{}+p_{n2}}|\lambda _{\sigma (1)}\mathrm{}\lambda _{\sigma (p_1)})$$
(58)
The resolution of the recursion gives
$`B_{p_0p_1\mathrm{}p_{n2}}^{(n)}(\lambda _1,\mathrm{},\lambda _{p_0+p_1+\mathrm{}+p_{n2}}|z_1,\mathrm{},z_{p_0})=`$
$`{\displaystyle \underset{\sigma _0S_{p_0}}{}}{\displaystyle \underset{\sigma _1S_{p_1}}{}}\mathrm{}{\displaystyle \underset{\sigma _{n3}S_{p_{n3}}}{}}{\displaystyle \underset{i=0}{\overset{n2}{}}}{\displaystyle \underset{\alpha _i=i+1}{\overset{n2}{}}}{\displaystyle \underset{k_{\alpha _i}=p_{\alpha _i+1}+1}{\overset{p_{\alpha _i}}{}}}{\displaystyle \underset{l_{\alpha _i}=p_{\alpha _i}+1}{\overset{p_i}{}}}{\displaystyle \frac{b(\lambda _{q_{i1}+\sigma _i(l_{\alpha _i})}\lambda _{\sigma _{i1}(k_{\alpha _i})})}{b(\lambda _{q_{i1}+\sigma _i(k_{\alpha _i})}\lambda _{q_{i1}+\sigma _i(l_{\alpha _i})})}}`$
$`{\displaystyle \underset{\gamma _i=i}{\overset{n2}{}}}B_{p_{\gamma _i}p_{\gamma _i+1}}^{(2)}(\lambda _{q_{i1}+\sigma _i(p_{\gamma _{i+1}+1})}\mathrm{}\lambda _{q_{i1}+\sigma _i(p_{\gamma _i})}|\lambda _{\sigma _{i1}(p_{\gamma _{i+1}+1})}\mathrm{}\lambda _{\sigma _{i1}(p_{\gamma _i})})\times `$
$`B_{p_{n2}}^{(2)}\left(\lambda _{q_{n3}+1}\mathrm{}\lambda _{q_{n3}+p_{n2}}|\lambda _{q_{n4}+\sigma _{n3}(1)}\mathrm{}\lambda _{q_{n4}+\sigma _{n3}(p_{n2})}\right)`$ (59)
where by definition
$$q_i=\underset{j=0}{\overset{i}{}}p_j;q_1=0$$
and
$$\lambda _{\sigma _1(k)}=z_k.$$
Eq’s (57) and (59) supply the explicit representation of the $`sl(n)`$ wave vectors in terms of $`sl(2)`$ vectors, that is, the resolution of the Bethe hierarchy.
## 6 Conclusions
The form of the factorizing $`F`$-matrix presented in section 3 is of an intriguing simplicity. We suspect that a representation theoretical aspect is lurking behind it which escapes our present knowledge. It should be noted that we arrived at this ansatz by guesswork immediately for the full $`F`$-matrix instead of taking the detour via partial $`F`$-matrices, as proposed in . It seems rather likely that we would have missed the simplicity of the ansatz if we had chosen the approach via partial $`F`$-matrices.
Our original hope was to find a structure for the Bethe wave vectors which is as suggestive as the one displayed for the case of Gaudin magnets in . This goal has not yet been achieved completely since we are not in possession of an entirely satisfactory representation of $`sl(2)`$ wave vectors, which are the building blocks for the final formula (59) of section 5. The representations (47), (48) both have the drawback that they do display the singularity structure of the wave vectors in a redundant manner. (The matter is further discussed in Appendix B.) We nevertheless nourish the hope that our findings will be of some help to bring effective large $`n`$ calculations of thermodynamical quantities into the range of the algebraic Bethe ansatz method.
Acknowledgement: We are indebted to J.–M. Maillet for a seminar and ensuing fruitful discussions, which initiated the present work. H.B. thanks the Alexander von Humboldt Foundation for support. R.F. was supported by the TMR network contract FMRX-CT96-0012 of the European Commission.
## Appendix A
In this appendix we verify the $`sl`$(n) algebra relations taking the formulas for the generators $`\stackrel{~}{E}_{\alpha ,\alpha \pm 1}`$ of section 4 as a starting point.
We use the following defining relations for a semisimple Lie algebra :
Fix a root system with a basis $`\{\alpha _1,\mathrm{},\alpha _l\}`$. Let L be the Lie algebra generated by $`3l`$ elements $`\{E_{+\alpha _i},E_{\alpha _i},H_i;\mathrm{\hspace{0.33em}1}il\}`$. $`L`$ is uniquely determined by the relations
* S1 $`[E_{+\alpha _i},E_{\alpha _j}]=\delta _{ij}H_{\alpha _i}`$
* S2 $`[H_{\alpha _i},E_{\pm \alpha _j}]=\pm A_{ji}E_{\pm \alpha _j}`$
* S3 $`[H_{\alpha _i},H_{\alpha _j}]=0`$
* S4 <sup>1</sup><sup>1</sup>1$`(ad_x)`$ is a shorthand notation for $`\underset{\mathrm{ntimes}}{\underset{}{ad_xad_x\mathrm{}ad_x}}`$, such that e.g. $`(ad_x)^2(y)=[x,[x,y]]`$ $`\left(ad_{E_\pm ^{\alpha _i}}\right)^{1A_{ji}}\left(E_\pm ^{\alpha _j}\right)=0i=1,\mathrm{},l;ij`$
with $`A_{ij}=2\frac{(\alpha _i,\alpha _j)}{(\alpha _j,\alpha _j)}`$ denoting the Cartan matrix.
We recall from section 4 the expressions for the generators of the algebra $`sl(n)`$ corresponding to simple roots:
$`\stackrel{~}{E}_{+\alpha }`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}E_{+\alpha }^{(i)}_{ji}(1\mathrm{I}_N+{\displaystyle \frac{\eta }{z_iz_j}}e_{\alpha \alpha })_{[j]}{\displaystyle \underset{i=1}{\overset{N}{}}}E_{+\alpha }^{(i)}_{ji}_{(i,j)}^\alpha `$
$`\stackrel{~}{E}_\alpha `$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{N}{}}}E_\alpha ^{(i)}_{ji}\left(1\mathrm{I}_N+{\displaystyle \frac{\eta }{z_jz_i}}e_{\alpha +1\alpha +1}\right)_{[j]}{\displaystyle \underset{i=1}{\overset{N}{}}}E_\alpha ^{(i)}_{ji}\stackrel{~}{}_{(j,i)}^\alpha `$ (60)
where $`\left(e_{ij}\right)_{kl}=\delta _{ik}\delta _{jl}`$ are the elementary matrices and $`\left(E_{+\alpha }^{(k)}\right)_{ij}=\delta _{\alpha i}\delta _{\alpha +1j},\left(E_\alpha ^{(k)}\right)_{ij}=\delta _{\alpha +1i}\delta _{\alpha j}`$ are the simple roots of $`sl(n)`$ acting in the k-th space.
Using their definitions one has <sup>2</sup><sup>2</sup>2$`_{i,j}^{}`$ means $`_{i,jij}`$.
$`[\stackrel{~}{E}_{+\alpha },\stackrel{~}{E}_\beta ]=`$
$`=`$ $`{\displaystyle \underset{i}{}}[E_{+\alpha }^{(i)},E_\beta ^{(i)}]_{ji}_{(i,j)}^\alpha \stackrel{~}{}_{(j,i)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}(E_\alpha ^{(i)}\stackrel{~}{}_{(i,j)}^\beta _{(i,j)}^\alpha E_\beta ^{(j)}\stackrel{~}{}_{(i,j)}^\beta E_\alpha ^{(i)}E_\beta ^{(j)}_{(i,j)}^\alpha )_{ki,j}_{(i,k)}^\alpha \stackrel{~}{}_{(j,k)}^\beta =`$
$`=`$ $`{\displaystyle \underset{i}{}}\delta _{\alpha \beta }H_\alpha ^{(i)}_{ji}_{(i,j)}^\alpha \stackrel{~}{}_{(j,i)}^\alpha `$
where we exploited the fact that the second sum vanishes term by term identically for all $`\{\alpha ,\beta \}`$.
The dressing can be written as $`_{(i,j)}^\alpha \stackrel{~}{}_{(j,i)}^\alpha =1\mathrm{I}_{[j]}+\frac{\eta }{z_iz_j}H_\alpha ^{(j)}`$, because $`\left(H_\alpha \right)_{ij}=\delta _{\alpha i}\delta _{\alpha j}\delta _{\alpha +1i}\delta _{\alpha +1j}`$.
Against first appearance the Cartan operators $`H_\alpha `$ remain without dressing. For this purpose we consider the expression
$`_{i=1}^N(1\mathrm{I}_{[i]}+{\displaystyle \frac{\eta }{\lambda z_i}}H_\alpha ^{(i)})=1\mathrm{I}_{[N]}+{\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{\eta }{\lambda z_i}}H_\alpha ^{(i)}_{ji}(1\mathrm{I}_{[j]}+{\displaystyle \frac{\eta }{z_iz_j}}H_\alpha ^{(j)}).`$ (62)
This identity can be proved by noting that both sides have the same limit for $`\lambda \mathrm{}`$ and that the residues at the simple poles $`\lambda =z_i`$ are identical. If we now consider the order $`1/\lambda `$ in the expansion of both sides we obtain
$`{\displaystyle \underset{i}{}}H_\alpha ^{(i)}_{ji}1\mathrm{I}_{[j]}={\displaystyle \underset{i}{}}H_\alpha ^{(i)}_{ji}(1\mathrm{I}_{[j]}+{\displaystyle \frac{\eta }{z_iz_j}}H_\alpha ^{(j)})={\displaystyle \underset{i}{}}H_\alpha ^{(i)}_{ji}_{(i,j)}^\alpha \stackrel{~}{}_{(j,i)}^\alpha `$ (63)
which finishes the proof that the Cartan operators associated with the simple roots aquire no dressing, which in turn renders the proof of the commutativity of the Cartan operators trivial.
To prove the Serre relation
$`\left(ad_{E_\pm ^{\alpha ^i}}\right)^{1A_{ji}}\left(E_\pm ^{\alpha ^j}\right)=0i=1,\mathrm{},N1;ij`$ (64)
we have to distinguish 2 cases:
$`1.`$ $`|ji|=1[E_\pm ^{\alpha ^i},[E_\pm ^{\alpha ^i},E_\pm ^{\alpha ^j}]]=0`$ (65)
$`2.`$ $`|ji|>1[E_\pm ^{\alpha ^i},E_\pm ^{\alpha ^j}]=0`$
as all other matrix-elements of the Cartan matrix are zero. (For $`sl(n)`$ we have $`A_{ii}=2,A_{i+1i}=A_{ii+1}=1,A_{ij}=0`$ otherwise.)
To proceed with the proof we list some useful relations:
$`\left(E_{+\alpha }_{(i,j)}^\beta \right)`$ $`=`$ $`d_{\alpha +1}^\beta (i,j)E_{+\alpha }`$
$`\left(_{(i,j)}^\beta E_{+\alpha }\right)`$ $`=`$ $`d_\alpha ^\beta (i,j)E_{+\alpha }`$
$`\left(E_\alpha _{(i,j)}^\beta \right)`$ $`=`$ $`d_\alpha ^\beta (i,j)E_\alpha `$
$`\left(_{(i,j)}^\beta E_\alpha \right)`$ $`=`$ $`d_{\alpha +1}^\beta (i,j)E_\alpha `$
where $`d_\alpha ^\beta (i,j)`$ means the $`\alpha ^{\mathrm{th}}`$ element on the diagonal of the matrix $`_{(i,j)}^\beta `$.
We now look at the first case of (65) and show the argument for the positive roots:
$`[\stackrel{~}{E}_\alpha ,\stackrel{~}{E}_\beta ]`$ $`=`$ $`{\displaystyle \underset{i}{}}[E_\alpha ^{(i)},E_\beta ^{(i)}]_{ji}_{(i,j)}^\alpha _{(i,j)}^\beta `$ (67)
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}{\displaystyle \frac{\eta }{z_jz_i}}E_\alpha ^{(i)}E_\beta ^{(j)}_{ki,j}_{(i,k)}^\alpha _{(j,k)}^\beta `$
and thus
$`[\stackrel{~}{E}_\alpha ,[\stackrel{~}{E}_\alpha ,\stackrel{~}{E}_\beta ]]`$
$`=`$ $`{\displaystyle \underset{i}{}}[E_\alpha ^{(i)},[E_\alpha ^{(i)},E_\beta ^{(i)}]]_{ji}_{(i,j)}^\alpha _{(i,j)}^\alpha _{(i,j)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}(E_\alpha ^{(i)}_{(j,i)}^\alpha _{(j,i)}^\beta _{(i,j)}^\alpha [E_\alpha ^{(j)},E_\beta ^{(j)}]_{(j,i)}^\alpha _{(j,i)}^\beta E_\alpha ^{(i)}[E_\alpha ^{(j)},E_\beta ^{(j)}]_{(i,j)}^\alpha )`$
$`_{ki,j}_{(i,k)}^\alpha _{(j,k)}^\alpha _{(j,k)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}{\displaystyle \frac{\eta }{z_jz_i}}(E_\alpha ^{(i)}E_\alpha ^{(i)}_{(i,j)}^\alpha E_\beta ^{(j)}E_\alpha ^{(i)}E_\alpha ^{(i)}E_\beta ^{(j)}_{(i,j)}^\alpha )_{ki,j}_{(i,k)}^\alpha _{(j,k)}^\alpha _{(j,k)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}{\displaystyle \frac{\eta }{z_jz_i}}(_{(j,i)}^\alpha E_\alpha ^{(i)}E_\alpha ^{(j)}E_\beta ^{(j)}E_\alpha ^{(i)}_{(j,i)}^\alpha E_\beta ^{(j)}E_\alpha ^{(j)})_{ki,j}_{(i,k)}^\alpha _{(j,k)}^\alpha _{(j,k)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j,k}{}}{}_{}{}^{}{\displaystyle \frac{\eta }{z_kz_j}}(E_\alpha ^{(i)}_{(j,i)}^\alpha _{(k,i)}^\beta _{(i,j)}^\alpha E_\alpha ^{(j)}_{(i,k)}^\alpha E_\beta ^{(k)}_{(j,i)}^\alpha _{(k,i)}^\beta E_\alpha ^{(i)}E_\alpha ^{(j)}_{(i,j)}^\alpha E_\beta ^{(k)}_{(i,k)}^\alpha )`$
$`_{li,j,k}_{(i,l)}^\alpha _{(j,l)}^\alpha _{(k,l)}^\beta `$
The first term in this sum vanishes due to the Serre relation for the undressed operators, the third because $`E_\alpha E_\alpha =0`$.
The second and the fourth term cancel each other, while the last term vanishes for fixed $`k`$, since the bracket yields
$`{\displaystyle \frac{\eta }{z_kz_j}}\left((1+{\displaystyle \frac{\eta }{z_kz_i}})(1+{\displaystyle \frac{\eta }{z_iz_j}})(1+{\displaystyle \frac{\eta }{z_jz_i}})\right)E_\alpha ^{(i)}E_\alpha ^{(j)}E_\beta ^{(k)}`$ (69)
which is antisymmetric under the exchange of $`i`$ and $`j`$.
The second case of (65) yields
$`[\stackrel{~}{E}_\alpha ,\stackrel{~}{E}_\beta ]`$ (70)
$`=`$ $`{\displaystyle \underset{i}{}}[E_\alpha ^{(i)},E_\beta ^{(i)}]_{ji}_{(i,j)}^\alpha _{(i,j)}^\beta `$
$`+`$ $`{\displaystyle \underset{i,j}{}}{}_{}{}^{}(E_\alpha ^{(i)}_{(j,i)}^\beta _{(i,j)}^\alpha E_\beta ^{(j)}_{(j,i)}^\beta E_\alpha ^{(i)}E_\beta ^{(j)}_{(i,j)}^\alpha )_{ki,j}_{(i,k)}^\alpha _{(j,k)}^\beta `$
where the first term in the sum vanishes due to the assumption for the undressed operators and the second term vanishes as the bracket is zero for $`|\alpha \beta |>1`$.
The proof for the $`\stackrel{~}{E}_{}^{\alpha ^i}`$ proceeds along the same lines.
We proceed to give the form of the non–simple roots which can be obtained as multiple commutators of simple roots (proof by induction on $`\alpha `$)
$`\stackrel{~}{E}_{i\alpha i}`$ $`=`$ $`[\stackrel{~}{E}_{i\alpha i\alpha +1},\mathrm{},[\stackrel{~}{E}_{i3i2},[\stackrel{~}{E}_{i2i1},\stackrel{~}{E}_{i1i}]]\mathrm{}]`$
$`=`$ $`{\displaystyle \underset{k=1}{\overset{\alpha }{}}}{\displaystyle \underset{i_1\mathrm{}i_k}{}}{\displaystyle \underset{\gamma =1}{\overset{k1}{}}}{\displaystyle \frac{\eta }{z_{i_\gamma }z_{i_{\gamma +1}}}}{\displaystyle \underset{\alpha =\beta _0>\beta _1>\mathrm{}>\beta _k=0}{}}_{l=1}^kE_{i\beta _{l1},i\beta _l}^{(i_l)}`$ (71)
$`_{ji_1\mathrm{}i_k}\mathrm{\Gamma }_{\genfrac{}{}{0pt}{}{\underset{}{i_k\mathrm{}i_k}}{\beta _0\beta _1}\genfrac{}{}{0pt}{}{\underset{}{i_{k1}\mathrm{}i_{k1}}}{\beta _1\beta _2}\mathrm{}\genfrac{}{}{0pt}{}{\underset{}{i_1\mathrm{}i_1}}{\beta _{k1}\beta _k};j;i}^{(j)}`$
with $`\mathrm{\Gamma }_{j;i_k\mathrm{}i_ki_{k1}\mathrm{}i_{k1}\mathrm{}i_1\mathrm{}i_1;i}^{(j)}=\text{diag}\{1,\mathrm{},1,b_{i_kj}^1,\mathrm{},b_{i_1j}^1,\underset{i}{\underset{}{1,\mathrm{},1}}\}_{[j]}`$.
A similar formula holds for the negative roots
$`\stackrel{~}{E}_{ii\alpha }`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{\alpha }{}}}{\displaystyle \underset{i_1\mathrm{}i_k}{}}{\displaystyle \underset{\gamma =1}{\overset{k1}{}}}{\displaystyle \frac{\eta }{z_{i_\gamma }z_{i_{\gamma +1}}}}{\displaystyle \underset{\alpha =\beta _0>\beta _1>\mathrm{}>\beta _k=0}{}}_{l=1}^kE_{i\beta _l,i\beta _{l1}}^{(i_l)}`$ (72)
$`_{ji_1\mathrm{}i_k}\mathrm{\Gamma }_{\genfrac{}{}{0pt}{}{\underset{}{i_k\mathrm{}i_k}}{\beta _0\beta _1}\genfrac{}{}{0pt}{}{\underset{}{i_{k1}\mathrm{}i_{k1}}}{\beta _1\beta _2}\mathrm{}\genfrac{}{}{0pt}{}{\underset{}{i_1\mathrm{}i_1}}{\beta _k1\beta _k};j;i}^{(j)}`$
with $`\mathrm{\Gamma }_{j;i_k\mathrm{}i_ki_{k1}\mathrm{}i_{k1}\mathrm{}i_1\mathrm{}i_1;i}^{(j)}=\text{diag}\{1,\mathrm{},1,b_{ji_k}^1,\mathrm{},b_{ji_1}^1,\underset{i1}{\underset{}{1,\mathrm{},1}}\}_{[j]}`$.
## Appendix B
In this appendix we discuss further details of the structure of the coefficients (cf. (48))
$`B_p^{(2)}(\lambda _1,\mathrm{},\lambda _p;z_{i_1},\mathrm{},z_{i_p})`$ $`=`$ $`{\displaystyle \frac{\underset{ij}{}(\lambda _iz_j)}{_{i>j}(\lambda _i\lambda _j)_{i>j}(z_jz_i)}}\text{det}X`$
$`X_{ij}`$ $`=`$ $`{\displaystyle \frac{1}{\lambda _iz_j}}{\displaystyle \frac{1}{\lambda _iz_j+\eta }}.`$ (73)
This representation, as concise as it is, has the disadvantage that it does not reflect the singularity structure in an economic way (the poles in the prefactor on the r.h.s. of (73) are cancelled by zeroes in the determinant).
One may cure the defect by appropriate manipulations on the determinant in (73). Subtracting for example the last row of $`X`$ from all the others, extracting a rational factor from the n-th row and proceeding in the same spirit with the (n-1)-th row and consecutively other rows one arrives at the equality
$`B_p^{(2)}(\lambda _1,\mathrm{},\lambda _p;z_{i_1},\mathrm{},z_{i_p})`$ $`=`$ $`{\displaystyle \frac{1}{_{i<j}(z_jz_i)}}{\displaystyle \frac{1}{_{ij}(\lambda _iz_j+\eta )}}\text{det}Y`$
$`Y_{\alpha ,x}`$ $`=`$ $`P_\alpha (\lambda ;z_{p_x})`$
$`P_\alpha (\lambda ;z_{p_x})`$ $`=`$ $`\left\{{\displaystyle \underset{i=0}{\overset{n\alpha }{}}}(\lambda _{ni}z_{p_x}+\eta ){\displaystyle \underset{i=0}{\overset{n\alpha }{}}}(\lambda _{ni}z_{p_x})\right\}{\displaystyle \underset{j=1}{\overset{\alpha 1}{}}}(\lambda _jz_{p_x}+\eta )(\lambda _jz_{p_x})`$
One should note that the polynomial $`P_\alpha `$ depends on all $`\lambda `$-variables but only on a single $`z`$-variable. It follows that one can continue to extract polynomial factors from the determinant in (LABEL:det1) by subtraction of columns from columns. The ensuing differences $`P_\alpha (\lambda ,z_{p_x})P_\alpha (\lambda ,z_{p_y})`$ supply the desired factors $`(z_{p_x}z_{p_y})`$ which compensate the pole factors $`\frac{1}{_{i>j}(z_iz_j)}`$ in (LABEL:det1).
Unfortunately we have not found a concise closed form for the polynomial multiplying the remaining prefactor $`\frac{1}{_{ij}(\lambda _iz_j+\eta )}`$ in the final expression.
|
warning/0002/hep-th0002253.html
|
ar5iv
|
text
|
# The Orbifolds of 𝑁=2 Superconformal Theories with 𝑐=3
## 1 Introduction
The complete understanding of the moduli space of $`N=2`$ superconformal field theories with central charge $`c=3`$ needs a description of all its orbifold theories. In a non-linear $`\sigma `$-model description, this concerns two dimensional tori and their orbifolds. For $`_3`$, $`_4`$ and $`_6`$ orbifolds, C. Vafa and N. Warner made predictions for (chiral, chiral) and (antichiral, antichiral) fields based on Landau-Ginzburg descriptions. Apparently, they never had been checked explicitely. The moduli spaces of those orbifold theories were obtained in . Landau-Ginzburg descriptions for the three orbifolds, we use the superpotentials $`\mathrm{\Phi }_1^3+\mathrm{\Phi }_2^3+\mathrm{\Phi }_3^3+6a\mathrm{\Phi }_1\mathrm{\Phi }_2\mathrm{\Phi }_3`$, $`\mathrm{\Phi }_1^4+\mathrm{\Phi }_2^4+a\mathrm{\Phi }_1^2\mathrm{\Phi }_2^2`$, and $`\mathrm{\Phi }_1^3+\mathrm{\Phi }_2^6+a\mathrm{\Phi }_1^2\mathrm{\Phi }_2^2`$, respectively. Note that we are interested in one dimensional moduli spaces, such that one needs superpotentials with one free parameter a or, in other words, singularities of modality one. Correlation functions for these potentials have been studied in . Here we calculate the $`_M`$ orbifold partition functions and check the predictions of C. Vafa and N. Warner. For $`c=6`$ similar calculations have been formulated by T. Eguchi et al . There, charges behave in a simpler way than for $`c=3`$. When fermions are omitted from the $`c=3`$ superconformal theories, one obtains $`c=2`$ bosonic theories. In this case the partition function for the $`_2`$ orbifold was given in .
The $`N=2`$ superconformal field theories with $`c=3`$ are described by a free chiral scalar superfield containing two real bosons or a single complex left (right) boson $`\phi ^\pm (z)=\phi ^1(z)\pm i\phi ^2(z)`$ $`(\overline{\phi }^\pm (\overline{z})=\overline{\phi }^1(\overline{z})\pm i\overline{\phi }^2(\overline{z}))`$ (each of $`c=1`$) and two Majorana-Weyl (MW) fermions or a free complex left(right) fermion $`\psi ^\pm (z)=\psi ^1(z)\pm i\psi ^2(z)`$ $`(\overline{\psi }^\pm (\overline{z})=\overline{\psi }^1(\overline{z})\pm i\overline{\psi }^2(\overline{z}))`$ (each of $`c=\frac{1}{2}`$). The action for this system may be written as
$$S=\frac{1}{2\pi }d^2z(G_{ij}\phi ^i\overline{}\phi ^j+B_{ij}\phi ^i\overline{}\phi ^j+\psi ^{}\overline{}\psi ^++\psi ^+\overline{}\psi ^{}).$$
(1)
In string theory language, this action corresponds to the superstring compactification on a two dimensional torus $`T^2=^2/\mathrm{\Lambda }`$. For the two dimensional lattice $`\mathrm{\Lambda }`$, we use a basis $`\{e_i\}`$ $`(i=1,2)`$. The action (1) depends on four real parameters or moduli, the constant symmetric metric $`G_{ij}=\frac{1}{2}e_ie_j`$ on $`T^2`$, and the antisymmetric tensor field $`B_{ij}=B_{ji}`$. It has $`N=2`$ superconformal symmetry. Directly from the action, we can determine the generators of the $`N=2`$ superconformal algebra, the stress-energy tensor $`T(z)`$, its super partners $`Q^i(z)=Q^1(z)\pm iQ^2(z)`$ ($`i=1,2`$), and the $`U(1)`$ current $`J(z)`$ with conformal dimensions $`h`$ equal to $`2`$, $`3/2`$, and $`1`$, respectively
$`T(z)`$ $`=`$ $`{\displaystyle \frac{1}{2}}\phi ^{}(z)\phi ^+(z){\displaystyle \frac{1}{4}}\psi ^{}\psi ^+(z){\displaystyle \frac{1}{4}}\psi ^+(z)\psi ^{}(z)`$
$`Q^\pm (z)`$ $`=`$ $`\psi ^{}(z)\phi ^\pm (z),J(z)={\displaystyle \frac{1}{2}}\psi ^{}(z)\psi ^+(z)={\displaystyle \frac{i}{2}}\epsilon ^{ij}\psi ^i(z)\psi ^j(z).`$ (2)
Similar relations hold for the antiholomorphic (right moving) generators of the $`N=2`$ superconformal algebra. They have the Laurent expansions
$$T(z)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}L_nz^{n2},Q^i(z)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}Q_rz^{r3/2},J(z)=\underset{n=\mathrm{}}{\overset{+\mathrm{}}{}}J_nz^{n1},$$
and satisfy $`N=2`$ superconformal algebra that can be found in . There are three different $`N=2`$ superconformal algebras, namely Ramond (R) (or periodic (P)), Neveu-Schwarz (NS) (or antiperiodic (A)) and twisted (T) algebras which correspond to different ways of choosing boundary conditions on the cylinder. Whatever boundary condition we choose the Virasoro generator $`L_n`$ is always integrally moded, because the bosonic stress-energy tensor is always periodic on the cylinder. For the Ramond (R) algebra, $`J_n`$ and $`Q_r^i`$ are integrally moded, i.e. n and r run over integral values. For the Neveu-Schwarz (NS) algebra, $`J_n`$ are integrally moded, $`Q_r^i`$ are half integrally moded, i.e. r run over half integral values. The twisted (T) algebra has integer modes for $`Q_r^1`$, half integer modes for $`J_n`$ and $`Q_r^2`$.
A field satisfying $`h=\pm \frac{q}{2}`$ is a left chiral or left antichiral primary field. (Similarly, a field satisfying $`\overline{h}=\pm \frac{\overline{q}}{2}`$ is a right chiral or right antichiral primary field). Note that the fermionic fields $`\{\psi ^\pm (z),\overline{\psi }^\pm (\overline{z})\}`$ all satisfy the above condition since they have charge $`\pm 1`$ and conformal dimension $`\frac{1}{2}`$ for both the left movers and right movers. The left primary chiral fields are $`\{1,\psi ^+(z)\}`$; the right chiral primary fields are $`\{1,\overline{\psi }^+(\overline{z})\}`$. The left and right antichiral primary fields are obtained from these by complex conjugation. Note that the conformal dimensions and $`U(1)`$ charges of an unique highest left-right chiral or antichiral primary field are $`(h,\overline{h})=(c/6,c/6)=(\frac{1}{2},\frac{1}{2})`$ and $`(q,\overline{q})=(\pm c/3,\pm c/3)=(\pm 1,\pm 1)`$, respectively (here $`c=3`$).
In general for $`N=2`$ superconformal theories, there are four types of rings arising from the various combinations of left-right chiral and left-right antichiral fields. We denote these rings by $`(c,c)`$, $`(a,a)`$, $`(a,c)`$, $`(c,a)`$. They are pairwise conjugate. For the $`_M`$, $`M\{3,4,6\}`$, orbifolds of $`N=2`$ superconformal theories with $`c=3`$, and for $`N=2`$ superconformal Landau-Ginzburg models, one obtains only $`(c,c)`$ and its conjugate $`(a,a)`$ rings. For such models, the $`(a,c)`$ and $`(c,a)`$ rings are trivial and consist only of the identity operator. We shall see this point explicitely in the discussion of $`_M`$ orbifolds and Landau-Ginzburg models.
The basic linearly independent elements of the $`(c,c)`$ ring of the $`N=2`$ superconformal field theory with $`c=3`$ is given by
$$_{(c,c)}=\{1,\psi ^+(z),\overline{\psi }^+(\overline{z}),\psi ^+(z)\overline{\psi }^+(\overline{z})\}.$$
(3)
Similarly, for the (a,c) ring one has
$$_{(a,c)}=\{1,\psi ^{}(z),\overline{\psi }^+(\overline{z}),\psi ^{}(z)\overline{\psi }^+(\overline{z})\}.$$
(4)
The elements of the two other rings $`_{(a,a)}`$ and $`_{(c,a)}`$ are obtained from $`_{(c,c)}`$ and $`_{(a,c)}`$ by complex conjugation.
The conformal dimensions and $`U(1)`$ charges of the ground states of Ramond sector are $`(h,\overline{h})=(c/24,c/24)=(1/8,1/8)`$ and $`(q,\overline{q})=(\pm 1/2,\pm 1/2)`$, which also contribute to the Witten index $`Tr(1)^F`$ . The operator $`(1)^F`$, where $`F=F_L+F_R`$, and $`F_L`$, $`F_R`$ are left-right moving fermion numbers, defined to anticommute with all the fermionic operators $`(1)^F\psi (z)=\psi (z)(1)^F`$, and to commute with all the bosonic operators $`(1)^F\phi (z)=\phi (z)(1)^F`$, as well as to satisfy $`((1)^F)^2`$. It can be defined in terms of zero mode $`U(1)`$ current as
$$(1)^F=e^{\pi i(J_0\overline{J}_o)}.$$
It is well known that one can connect the Neveu-Schwarz sector to the Ramond sector by spectral flow operation. It is the continous transformation and has the following form
$`L_n^\eta `$ $`=`$ $`L_n+\eta J_n+{\displaystyle \frac{c}{6}}\eta ^2\delta _{n,0}`$
$`J_n^\eta `$ $`=`$ $`J_n+{\displaystyle \frac{c}{3}}\eta \delta _{n,0}`$
$`Q_r^{\pm \eta }`$ $`=`$ $`Q_{r\pm \eta }^\pm .`$
The $`\eta `$ twisted operators $`L_n^\eta `$, $`Q_r^{\pm \eta }`$ and $`J_n^\eta `$ still satisfy the $`N=2`$ superconformal algebra for an arbitrary value of the parameter $`\eta `$. In particular, the zero mode eigenvalues $`h`$ of $`L_0`$ and $`q`$ of $`J_0`$ are changed by spectral flow as
$$h_\eta =h+\eta q+\eta ^2\frac{c}{6},q_\eta =q+\eta \frac{c}{3}.$$
(5)
By (5) with flow parameter $`\eta =\frac{1}{2}`$, the ground states of Ramond sector with conformal dimension $`(h,\overline{h})=(1/8,1/8)`$ and charge $`(q,\overline{q})=(\pm 1/2,\pm 1/2)`$ flow to the Neveu-Schwarz chiral primary fields with conformal dimension $`(h,\overline{h})=(1/2,1/2)`$ and charge $`(q,\overline{q})=(+1,+1)`$, or $`(h,\overline{h})=(q,\overline{q})=(0,0)`$. The flow between the NS and NS as well as R and R can be obtained by the flow parameter $`\eta =1`$. Besides, under the left-right symmetric spectral flow, $`q\overline{q}`$ does not change. Thus the Witten’s index is
$`Tr(1)^F`$ $`=`$ $`Tr_R\left[(1)^{J_0\overline{J}_0}q^{L_0\frac{c}{24}}\overline{q}^{\overline{L}_0\frac{c}{24}}\right]`$ (6)
$`=`$ $`Tr__\eta \left[(1)^{J_0^\eta \overline{J}_0^\eta }q^{L_0^\eta \frac{c}{24}}\overline{q}^{\overline{L}_0^\eta \frac{c}{24}}\right]`$
$`=`$ $`Tr_{NS}\left[(1)^{J_0\overline{J}_0}q^{L_0\frac{1}{2}J_0}\overline{q}^{\overline{L}_0\frac{1}{2}\overline{J}_0}\right]={\displaystyle \underset{}{}}e^{i\pi (q\overline{q})},`$
where the $`_\eta `$ in the second line is the Hilbert space of states which is twisted by the parameter $`\eta `$. The $``$ in the last line denotes the chiral ring. First line implies that the ground state of the Ramond sector gives nonvanishing contribution to the Witten index. The second line is obtained by applying the spectral flow to the first line. By setting $`\eta =\frac{1}{2}`$ one can flow from Ramond sector to the Neveu-Schwarz sector. (Note that $`J_0^\eta \overline{J}_0^\eta =J_0\overline{J}_0`$). Thus the Witten index receives contributions from either the ground states of Ramond sector or the chiral primary states of Neveu-Schwarz sector. The only difference between the charges of the NS chiral primary states and that of the Ramond ground states is $`\frac{c}{6}`$.
The Poincaré polynomial is
$$P(t,\overline{t})=Tr_{}t^{J_0\overline{J}_0},$$
(7)
which satisfies a duality relation $`P(t,\overline{t})=(t\overline{t})^{1/3}P(1/t,1/\overline{t})`$. Here $`t`$ and $`\overline{t}`$ can be regarded as an independent variables. By (6), (7) and (3), the Witten index and the Poincaré polynomial are
$$Tr(1)^F=0,P(t,\overline{t})_{(c,c)}=1+t+\overline{t}+t\overline{t}.$$
(8)
One notes that the Poincaré polynomial (8) and ring structure for $`(c,c)`$ and $`(a,c)`$ primary fields are isomorphic. However, this is not true in general.
The partition function for the $`N=2`$ superconformal theories with $`c=3`$ is constructed by tensoring the theory of a complex free boson defined on a 2-dimentional torus $`T^2`$ in the presence of constant background fields, with the theory of a single complex Dirac fermion, namely
$$Z(\tau ,\rho ,z):=Z(\tau ,\rho ,\sigma )Z_{Dirac}(\sigma ,z),$$
In the following we briefly discuss how the explicit expression of $`Z(\tau ,\rho ,z)`$ can be formulated. The $`Z(\tau ,\rho ,\sigma )`$ is the modular invariant partition function for two real boson compactified on the two dimensional torus
$$Z(\tau ,\rho ):=Z(\tau ,\rho ,\sigma )=trq^{L_0^b\frac{1}{12}}\overline{q}^{\overline{L}_0^b\frac{1}{12}}=\frac{1}{\left|\eta ^2(\sigma )\right|^2}\underset{\genfrac{}{}{0pt}{}{n_1,m_1}{n_2,m_2}}{}q^{\frac{p^2}{2}}\overline{q}^{\frac{\overline{p}^2}{2}},$$
(9)
where $`q=e^{2\pi i\sigma }`$ , $`\sigma =\sigma _1+i\sigma _2`$ parametrizes the world sheet torus, and $`\eta (\sigma )`$ is the Dedekind eta function defined as
$$\eta (\sigma )=q^{\frac{1}{24}}\underset{n=1}{\overset{\mathrm{}}{}}(1q^n).$$
The Virasoro zero mode operators for the bosons in (9) are given by
$$L_0^b=\underset{n>0}{}\alpha _n^i\alpha _n^i+\frac{1}{2}p^2,\overline{L}_0^b=\underset{n>0}{}\overline{\alpha }_n^i\overline{\alpha }_n^i+\frac{1}{2}\overline{p}^2.$$
(10)
The left-right moving zero mode momentum $`p`$ and $`\overline{p}`$ in (9) are defined as
$$(p,\overline{p}):=(n_ie^i+e^iB_{ji}m^j+\frac{1}{2}e_jm^j,n_ie^i+e^iB_{ji}m^j\frac{1}{2}e_jm^j),$$
(11)
where $`\{e_i^{}\}`$ are basis vectors for the dual lattice $`\mathrm{\Lambda }^{}`$ of $`\mathrm{\Lambda }`$, which satisfies $`e_ie_j^{}=\delta _{ij}`$ such that $`e^ie^j=\frac{1}{2}G^{ij}`$; the integers $`n_i`$ and $`m_i`$ are the momentum and winding numbers. The action of $`L_0^b`$ and $`\overline{L}_0^b`$ in (10) on the ground state $`|m_1,m_2,n_1,n_2`$, which is labeled by the momentum and winding numbers, is given by
$$L_0^b|m_1,m_2,n_1,n_2=\frac{1}{2}p^2|m_1,m_2,n_1,n_2,\overline{L}_0^b|m_1,m_2,n_1,n_2=\frac{1}{2}\overline{p}^2|m_1,m_2,n_1,n_2.$$
where we have used $`\alpha _n^i|m_1,m_2,n_1,n_2=0`$ and $`\overline{\alpha }_m^j|m_1,m_2,n_1,n_2=0`$ for $`n>0`$, $`m>0`$. It is well known that the momenta in (11) form four dimensional Lorentzian lattice with scalar product $`(p,\overline{p})(p^{},\overline{p}^{})=(pp^{}\overline{p}\overline{p}^{})`$, which is even (because $`p^2\overline{p}^2=2m^in_i2`$ ) and self-dual ( because $`\mathrm{\Lambda }=\mathrm{\Lambda }^{}`$). From (11), we easily write
$$p^2(\overline{p}^2)=\frac{1}{2}n_in_jG^{ij}+n_im_jB_{jl}G^{il}\pm n_im_i+\frac{1}{2}m_im_j(G_{ij}+B_{jk}B_{il}G^{kl}).$$
(12)
In the two dimensional case, it is convenient to group the four real parameteres ($`G_{11}`$, $`G_{12}`$, $`G_{22}`$, and $`B_{12}`$) in terms of two parameters $`\tau `$ and $`\rho `$ in the upper complex half plane as follows
$$\tau =\tau _1+i\tau _2=\frac{G_{12}}{G_{22}}+i\frac{\sqrt{G}}{G_{22}},\rho =\rho _1+i\rho _2=B_{12}+i\sqrt{G}.$$
Here $`\tau `$ represents the complex structure of the target space torus $`T^2`$, and $`\rho `$ is its complexified Kähler structure; both take values on the complex upper half plane; $`G=det(G_{ij})`$. Now we write $`(\text{12})`$ in terms of $`\tau `$ and $`\rho `$ in the following form
$`p^2`$ $`=`$ $`{\displaystyle \frac{1}{2\tau _2\rho _2}}|n_1\tau n_2\rho (m_2+\tau m_1)|^2`$
$`\overline{p}^2`$ $`=`$ $`{\displaystyle \frac{1}{2\tau _2\rho _2}}|n_1\tau n_2\overline{\rho }(m_2+\tau m_1)|^2.`$
Finally, torus partition function (9) takes the form
$$Z(\tau ,\rho )=\frac{1}{\left|\eta ^2(\sigma )\right|^2}\underset{\genfrac{}{}{0pt}{}{n_1,m_1}{n_2,m_2}}{}q^{\frac{1}{4\tau _2\rho _2}|n_1\tau n_2\rho (m_2+\tau m_1)|^2}\overline{q}^{\frac{1}{4\tau _2\rho _2}|n_1\tau n_2\overline{\rho }(m_2+\tau m_1)|^2}.$$
(13)
If $`\tau _1=\rho _1=0`$ (or $`G_{12}=B_{12}=0`$), then the torus partition function (13) is the product of two circle partition functions at $`c=1`$ with radius $`r_1=\sqrt{G_{22}}=\sqrt{\rho _2/\tau _2}`$ and $`r_2=\sqrt{G_{11}}=\sqrt{\tau _2\rho _2}`$
$$Z(\tau _2,\rho _2)=Z^{c=1}(\sqrt{\rho _2/\tau _2})Z^{c=1}(\sqrt{\tau _2\rho _2}).$$
The partition function for the Dirac fermion can be constructed by taking equal spin structures for the left and right fermions
$`Z_{Dirac}(\sigma ,z)`$ $`=`$ $`trq^{L_0^f\frac{1}{24}}\overline{q}^{\overline{L}_0^f\frac{1}{24}}y^{J_0}\overline{y}^{\overline{J}_0}`$ (14)
$`=`$ $`{\displaystyle \frac{1}{2}}\left(\left|{\displaystyle \frac{\vartheta _1(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _2(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _3(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _4(z,\sigma )}{\eta (\sigma )}}\right|^2\right),`$
where $`y=e^{2\pi iz}`$. Since the fermionic theory split into Neveu-Schwarz and Ramond sector the Virasoro zero mode generator for the Dirac fermions in (14) is given by
$$L_0^f=\underset{n>0}{}nd_n^id_n^in+\frac{1}{2}(NS),L_0^f=\underset{n>0}{}nd_n^id_n^i+\frac{1}{8}n(R).$$
Similar relation is true for the right moving component. The classical Jacobi theta functions $`\vartheta _i(z,\sigma )`$, $`i\{1,2,3,4\}`$ in (14) are defined in terms of sums and products as
$`\theta _1(z,\sigma )`$ $`=`$ $`i{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}(1)^nq^{\frac{1}{2}(n\frac{1}{2})^2}y^{n\frac{1}{2}}=iy^{\frac{1}{2}}q^{\frac{1}{8}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1q^n)(1yq^n)(1y^1q^{n1})`$
$`\theta _2(z,\sigma )`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}q^{\frac{1}{2}(n\frac{1}{2})^2}y^{n\frac{1}{2}}=y^{\frac{1}{2}}q^{\frac{1}{8}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1q^n)(1+yq^n)(1+y^1q^{n1})`$
$`\theta _3(z,\sigma )`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}q^{\frac{n^2}{2}}y^n={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1q^n)(1+yq^{n\frac{1}{2}})(1+y^1q^{n\frac{1}{2}})`$
$`\theta _4(z,\sigma )`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}(1)^nq^{\frac{n^2}{2}}y^n={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}(1q^n)(1yq^{n\frac{1}{2}})(1y^1q^{n\frac{1}{2}}).`$
Partition function for the $`N=2`$ superconformal theories with $`c=3`$ is thus given as
$`Z(\tau ,\rho ,z):`$ $`=`$ $`Z(\tau ,\rho )Z_{Dirac}(\sigma ,z)`$ (15)
$`=`$ $`{\displaystyle \frac{1}{\left|\eta ^2(\sigma )\right|^2}}q^{\frac{1}{4\tau _2\rho _2}|n_1\tau n_2\rho (m_2+\tau m_1)|^2}\overline{q}^{\frac{1}{4\tau _2\rho _2}|n_1\tau n_2\overline{\rho }(m_2+\tau m_1)|^2}\times `$
$`{\displaystyle \frac{1}{2}}\left(\left|{\displaystyle \frac{\vartheta _1(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _2(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _3(z,\sigma )}{\eta (\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _4(z,\sigma )}{\eta (\sigma )}}\right|^2\right).`$
## 2 General Prescription for $`_M`$ Orbifold Construction
In this section we will give the general procedure for the construction of the $`_M`$ orbifolds. In fact there are not many two dimensional $`_M`$ orbifolds, because the order M rotation must be an automorphism of some two dimensional lattice; therefore $`_M`$ must have order $`M=2,3,4`$, and $`6`$. The $`M=3`$ and $`M=6`$ require the hexagonal lattice $`(\tau =e^{2\pi i/3})`$; $`M=4`$ requires a square lattice $`(\tau =i)`$. Under the $`_M`$ symmetry bosonic fields and its modes $`\alpha _n^\pm `$ transform as
$$(g^k\phi )^\pm (z)=e^{\pm \frac{2\pi ik}{M}}\phi ^\pm (z),g^k\alpha _n^\pm g^k=e^{\pm \frac{2\pi ik}{M}}\alpha _n^\pm ,k=1,2,\mathrm{},M1.$$
(16)
Since we want to discuss superconformal orbifold theories, we should include the worldsheet fermion $`\psi `$’s as well. They transform as
$$(g^k\psi )^\pm (z)=e^{\pm \frac{2\pi ik}{M}}\psi ^\pm (z),g^kd_n^\pm g^k=e^{\pm \frac{2\pi ik}{M}}d_n^\pm ,k=1,2,\mathrm{}M1.$$
(17)
In fact this is also required by the $`N=2`$ superconformal invariance. The $`_M`$ rotations are the symmetries both the action $`(\text{1})`$ and $`N=2`$ world sheet supersymmetry generators $`(\text{1})`$. Thus the two dimensional $`N=2`$ superconformal orbifold models $`T^2/_M`$ may be constructed by identifying points of the two-dimensional torus $`T^2`$ under the symmetry group $`_M`$.
Let $`\stackrel{~}{}`$ be the Hilbert space of an orbifold theory. It has two sectors, namely untwisted and twisted sector, i.e, $`\stackrel{~}{}=\stackrel{~}{}_u\stackrel{~}{}_t`$. Let us consider first the untwisted sector of the orbifold theory. The untwisted Hilbert space will be a subspace of the Hilbert space for the $`N=2`$ theories with $`c=3`$. In the path integral for the partition function this means that the bosonic fields obey periodic boundary conditions along the space direction of the torus and twisted periodic boundary conditions in time. So on an orbifold, the untwisted sector boundary conditions on the bosonic field are given as
$`\phi ^+(1)`$ $`=`$ $`\phi ^+(0)+2\pi \mathrm{\Lambda }`$
$`\phi ^+(\sigma )`$ $`=`$ $`g\phi ^+(0)+2\pi \mathrm{\Lambda },`$ (18)
where $`g_M`$. For Ramond or Neveu-Schwarz fermion one has
$`\psi ^+(1)`$ $`=`$ $`\pm \psi ^+(0)`$
$`\psi ^+(\sigma )`$ $`=`$ $`\pm g\psi ^+(0).`$ (19)
Under the above boundary conditions, the bosonic field has expansion
$$\phi ^+(z)=q^+ip^+\mathrm{ln}z+i\underset{n0}{}\frac{1}{n}\alpha _n^+z^n,$$
(20)
for the fermionic field one has
$`\psi ^+(z)`$ $`=`$ $`{\displaystyle \underset{n}{}}d_n^+z^n\{\begin{array}{cc}& n(R)\hfill \\ & n+\frac{1}{2}(NS)\hfill \end{array}.`$ (23)
The untwisted Hilbert space $`\stackrel{~}{}_u`$ decomposes into $`_M`$ invariant and noninvariant space of states. In oder to construct consistent models, we must project out the group noninvariant space of states. In the Hamiltonian formalism, group invariant states are obtained by insertion of the projection operator $`P=\frac{1}{|_M|}\underset{g_M}{}g`$ into the trace over states. Here $`|_M|`$ is the number of elements in $`_M`$ and the sum $`g`$ runs over all elements in $`_M`$. Thus the untwisted sector partition function is
$$Z_u=tr_{\stackrel{~}{}_u}Pq^{L_0\frac{1}{8}}\overline{q}^{\overline{L}_0\frac{1}{8}}y^{J_0}\overline{y}^{\overline{J}_0}.$$
(24)
Here $`tr_{\stackrel{~}{}_u}`$ denote the trace in the untwisted Hilbert space sectors and $`L_0=L_0^b+L_0^f`$. In the path integral formalism, projection onto group invariant states in the untwisted sector is represented as
$$Z_u=\frac{1}{|_M|}\underset{g_M}{}g\underset{1}{\text{ }\text{ }\text{ }\text{ }},$$
where we sum over all possible twistings in the time direction of the torus. $`g\underset{1}{\text{ }\text{ }\text{ }\text{ }}`$ represents boundary conditions on any generic fields in the theory twisted by g in the time direction of the torus. The partition function of the original model is simply given by $`Z=1\underset{1}{\text{ }\text{ }\text{ }\text{ }}`$ .
The untwisted sector partition function is not modular invariant; one should take into account the contributions of twisted sector Hilbert space of states. For each element $`h_M`$ one can construct a twisted Hilbert space $`\stackrel{~}{}_h`$. In the path integral description the bosonic field obey the twisted boundary conditions
$`\phi ^+(1)`$ $`=`$ $`h\phi ^+(0)+2\pi \mathrm{\Lambda }`$
$`\phi ^+(\sigma )`$ $`=`$ $`g\phi ^+(0)+2\pi \mathrm{\Lambda }.`$ (25)
For Ramond or Neveu-Schwarz fermions one has
$`\psi ^+(1)`$ $`=`$ $`\pm h\psi ^+(0)`$
$`\psi ^+(\sigma )`$ $`=`$ $`\pm g\psi ^+(0),`$ (26)
where h and g are twists on the fields in the space and time direction of the torus. The mode expansion of the bosonic field which satisfies the boundary conditions $`(\text{2})`$ is
$$\phi ^+(z)=q_f^++i\underset{n+k/M}{}\frac{1}{n}\alpha _n^+z^n.$$
(27)
One can not have nonzero momentum or winding number here, since they are not consistent with the twisted boundary conditions. In this mode expansion $`q_f^+`$ denote the fixed points of $`T^2`$ under the $`_M`$ symmetry. The index f labels these fixed points. The mode expansion of the fermionic field which satisfies the boundary conditions $`(\text{2})`$ is
$$\psi ^+(z)=\underset{n+k/M+1/2s/2}{}d_n^+z^n,k=1,\mathrm{}M1,$$
(28)
where s is equal to zero in the Neveu-Schwarz sector, and to one in the Ramond sector. The twisted Hilbert space $`\stackrel{~}{}_t`$ decomposes into $`_M`$ invariant and noninvariant space of states. To construct consistent models, we again have to project onto group invariant states. In the Hamiltonian formalism, group invariant states are obtained by insertion of the projection operator $`P_h:=\frac{1}{|_M|}\underset{g_M:[g,h]=0}{}g`$ into the trace over states. In the path integral formalism, projection onto group invariant states in the twisted sector is representes as
$$Z_t=\frac{1}{|_M|}\underset{\genfrac{}{}{0pt}{}{g,h_M,}{h1,[g,h]=0}}{}g\underset{h}{\text{ }\text{ }\text{ }\text{ }},$$
where $`g\underset{h}{\text{ }\text{ }\text{ }\text{ }}`$ represents boundary conditions on the fields twisted by g and h in the time and space direction of the torus. Thus the twisted sector partition function has the form
$$Z_t=\underset{h_M,h1}{}tr_{\stackrel{~}{}_h}P_hq^{L_0\frac{1}{8}}\overline{q}^{\overline{L}_0\frac{1}{8}}y^{J_0}\overline{y}^{\overline{J}_0}=\frac{1}{|_M|}\underset{\genfrac{}{}{0pt}{}{g,h_M,}{h1,[g,h]=0}}{}g\underset{h}{\text{ }\text{ }\text{ }\text{ }}.$$
(29)
In fact, one may obtain the twisted sector partition function from (24) by modular transformations $`\sigma \sigma +1`$ and $`\sigma 1/\sigma `$. Thus, total modular invariant $`_M`$ orbifold partition function is a sum of (24) and (29)
$`Z_{_Morb}`$ $`=`$ $`{\displaystyle \frac{1}{|_M|}}{\displaystyle \underset{g_M}{}}g\underset{1}{\text{ }\text{ }\text{ }}+{\displaystyle \frac{1}{|_M|}}{\displaystyle \underset{g,h_M,h1}{}}g\underset{h}{\text{ }\text{ }\text{ }}`$ (30)
$`=`$ $`{\displaystyle \frac{1}{|_M|}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{g,h_M,}{[g,h]=0}}{}}g\underset{h}{\text{ }\text{ }\text{ }}={\displaystyle \underset{h_M}{}}tr_{\stackrel{~}{}_h}P_hq^{L_0\frac{1}{8}}\overline{q}^{\overline{L}_0\frac{1}{8}}y^{J_0}\overline{y}^{\overline{J}_0},`$
where we set $`\stackrel{~}{}_1:=\stackrel{~}{}_u`$ and $`P_1:=P`$. There is no discrete torsion for the $`_M`$ orbifolds, since all boxes $`g\underset{h}{\text{ }\text{ }\text{ }\text{ }}`$ are related by modular tranformations to a box of type $`g\underset{1}{\text{ }\text{ }\text{ }\text{ }}`$. Mathematically, the discrete torsion for a discrete group G is obtained from the cohomology $`H_2(G)`$, which vanishes for $`G=_M`$ .
In summary, in order to construct an orbifold model, one first formulates the Hilbert space of states on the torus, then one projects onto the group invariant states, finally one includes twisted sector contributions. For more details see ref. .
## 3 The $`_2`$ Orbifold
The two dimensional $`N=2`$ superconformal $`_2`$ orbifold model $`T^2/_2`$ can be constructed from (15) for arbitrary $`\tau `$ and $`\rho `$. Thus we may now produce another family of theories, i.e. $`_2`$ orbifold superconformal field theories with the same set of moduli as the $`N=2`$ theories with $`c=3`$ by following the general orbifold prescription introduced in section two. The action of $`g_2`$ on the bosonic Hilbert space sectors $`|m_1,m_2,n_1,n_2`$ is given by
$$g|m_1,m_2,n_1,n_2=|m_1,m_2,n_1,n_2.$$
(31)
In the following, we only discuss the bosonic part since the sum over the spin structures for the Dirac fermion is invariant under $`\psi ^\pm \psi ^\pm `$. Under the $`_2`$ symmetry the untwisted bosonic Hilbert spaces $`\stackrel{~}{}_u`$ decomposes into $`g=\pm 1`$ eigenspaces $`\stackrel{~}{}_u=\stackrel{~}{}_u^+\stackrel{~}{}_u^{}`$ as
$`\stackrel{~}{}_u^+`$ $`=`$ $`\{\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j}}^+(1+g)|m_1,m_2,n_1,n_2\}`$
$`+\{\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j+1}}^+(1g)|m_1,m_2,n_1,n_2\}`$
$`\stackrel{~}{}_u^{}`$ $`=`$ $`\{\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j+1}}^+(1+g)|m_1,m_2,n_1,n_2\}`$
$`+\{\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j}}^+(1g)|m_1,m_2,n_1,n_2\},`$
where $`k_i`$ takes positive integer values. By (24), untwisted $`_2`$ orbifold partition function is
$`Z_u=(q\overline{q})^{\frac{1}{8}}tr_{\stackrel{~}{}_u}{\displaystyle \frac{1}{2}}(1+g)q^{L_0}\overline{q}^{\overline{L}_0}y^{J_0}\overline{y}^{\overline{J}_0}.`$
The first term in the trace is equal to the partition function in $`(\text{15})`$ since there is no twist along the two cycles of the torus. The second term in the trace with $`g`$ inserted receives only contribution from the sector $`m_1=m_2=n_1=n_2=0`$ because each state obtained by acting on $`(1+g)|m_1,m_2,n_1,n_2`$ with creation operators has a counter part with the same $`L_0`$ eigenvalue obtained by acting on $`(1g)|m_1,m_2,n_1,n_2`$ with the same creation operators; however, these two states have opposite eigenvalues under $`g_2`$, and their contributions cancel in the trace. Thus, only the states obtained by acting creation operators $`\alpha _k^+`$ or $`\overline{\alpha }_k^+`$ on the vacuum $`|0,0,0,0`$ will contribute. Therefore the overall untwisted sector partition function is
$`Z_u`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{1}{|\eta ^2|^2}}{\displaystyle \underset{\genfrac{}{}{0pt}{}{n_1,m_1}{n_2,m_2}}{}}q^{\frac{p^2}{2}}\overline{q}^{\frac{\overline{p}^2}{2}}+{\displaystyle \frac{(q\overline{q})^{\frac{1}{12}}}{\underset{n=1}{\overset{\mathrm{}}{}}(1+q^n)^2(1+\overline{q}^n)^2}}\right)Z_{Dirac}`$
$`=`$ $`{\displaystyle \frac{1}{2}}\left(Z(\tau ,\rho )+\mathrm{\hspace{0.33em}4}\left|{\displaystyle \frac{\eta (\sigma )}{\vartheta _2(\sigma )}}\right|^2\right)Z_{Dirac}.`$
Under the symmetry action $`g`$ : $`\phi ^+\phi ^+`$ the torus has four fixed points. This implies that there are four twisted ground states with conformal dimension $`h=\overline{h}=1/8`$. So one may build four distinct Hilbert space sectors. However, these sectors lead to isomorphic physics, as they are related by translation symmetry of the torus. Denote the four twisted sector ground states by $`|\frac{1}{8},\frac{1}{8}_f`$, where $`f=1,2,3,4,`$. As untwisted bosonic Hilbert space sector, the twisted bosonic Hilbert space decomposes into $`g=\pm 1`$ eigenspaces $`\stackrel{~}{}_t=\stackrel{~}{}_t^+\stackrel{~}{}_t^{}`$ as
$`\stackrel{~}{}_t^+`$ $`=`$ $`\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j}}^+|{\displaystyle \frac{1}{8}},{\displaystyle \frac{1}{8}}_f`$
$`\stackrel{~}{}_t^{}`$ $`=`$ $`\alpha _{k_1}^+\mathrm{}\alpha _{k_l}^+\overline{\alpha }_{k_{l+1}}^+\mathrm{}\overline{\alpha }_{k_{2j+1}}^+|{\displaystyle \frac{1}{8}},{\displaystyle \frac{1}{8}}_f.`$
where $`k_i`$ takes half positive integer values. By (29), the twisted sector partition function is
$`Z_t`$ $`=`$ $`(q\overline{q})^{\frac{1}{12}}tr_{\stackrel{~}{}_t}{\displaystyle \frac{1}{2}}(1+g)q^{L_0}\overline{q}^{\overline{L}_0}Z_{Dirac}`$ (32)
$`=`$ $`4\times {\displaystyle \frac{1}{2}}\left(\left|{\displaystyle \frac{q^{\frac{1}{24}}}{_{n=1}^{\mathrm{}}(1q^{n1/2})^2}}\right|^2+\left|{\displaystyle \frac{q^{\frac{1}{24}}}{_{n=1}^{\mathrm{}}(1+q^{n1/2})^2}}\right|^2\right)Z_{Dirac}`$
$`=`$ $`4\times {\displaystyle \frac{1}{2}}\left(\left|{\displaystyle \frac{\eta (\sigma )}{\vartheta _4(\sigma )}}\right|^2+\left|{\displaystyle \frac{\eta (\sigma )}{\vartheta _3(\sigma )}}\right|^2\right)Z_{Dirac}.`$
Then the complete modular invariant $`_2`$ orbifold partition function has the form
$$Z_{_2orb}=\frac{1}{2}\left(Z(\tau ,\rho )+4\left|\frac{\eta (\sigma )}{\vartheta _2(\sigma )}\right|^2+\mathrm{\hspace{0.33em}4}\left|\frac{\eta (\sigma )}{\vartheta _3(\sigma )}\right|^2+\mathrm{\hspace{0.33em}4}\left|\frac{\eta (\sigma )}{\vartheta _4(\sigma )}\right|^2\right)Z_{Dirac}.$$
(33)
The (c, c), (a, c), and their complex conjugates, Ramond ground states as well as the Witten index for the $`_2`$ orbifold are the same as those for the $`N=2`$ theories with $`c=3`$.
## 4 The $`_3`$ Orbifold
By dividing the $`_3`$ symmetry from (15) for $`\tau =e^{2\pi i/3}`$ and arbitrary $`\rho `$, we may construct $`_3`$ orbifold model. The action of $`g_3`$ on the bosonic Hilbert space sectors is given by
$$g|m_1,m_2,n_1,n_2=|m_2,m_1m_2,n_2n_1,n_1.$$
(34)
By (24), the untwisted sector partition function is
$$Z_u=(q\overline{q})^{\frac{1}{8}}tr_{\stackrel{~}{}_u}\frac{1}{3}(1+g+g^2)q^{L_0}\overline{q}^{\overline{L}_0}y^{J_0}\overline{y}^{\overline{J}_0}.$$
By taking into account the equations (16), (17) (20), (23) and (34), the first term in the trace is equal to the original patition function (15), the second and third term receives only contribution from the Hilbert space sector built on $`|0,0,0,0`$. The untwisted sector partition function is therefore given by
$`Z_u={\displaystyle \frac{1}{3}}\left(Z(\tau =e^{2\pi i/3},\rho ,z)+{\displaystyle \frac{3}{2}}{\displaystyle \underset{i=1}{\overset{4}{}}}\left(\left|{\displaystyle \frac{\vartheta _i(z+\frac{1}{3},\sigma )}{\vartheta _1(\frac{1}{3},\sigma )}}\right|^2+\left|{\displaystyle \frac{\vartheta _i(z\frac{1}{3},\sigma )}{\vartheta _1(\frac{1}{3},\sigma )}}\right|^2\right)\right).`$
$`_3`$ does not act freely on the hexagonal torus. Thus one must consider new sectors, the twisted ones. In the $`T^2/_3`$ $`(\tau =e^{2\pi i/3})`$ manifold, there are three fixed points, and one can obtain three Hilbert space sectors corresponding to the expansion of the field about each of these fixed points. However these three sectors give the same physics. The conformal weight of the bosonic twisted ground state is $`(\frac{1}{9},\frac{1}{9})`$. For fermion, twisted sector conformal weight is $`(\frac{1}{18},\frac{1}{18})`$. Thus the total conformal weight of the twisted sector is then $`(\frac{1}{6},\frac{1}{6})`$. States in the twisted sector are generated by the action of creation operators on the twisted ground state.
By considering the equations (16), (17), (27), (28) and (34), the twisted sector partition function may be written as
$`Z_t`$ $`=`$ $`(q\overline{q})^{\frac{1}{8}}tr_{\stackrel{~}{}_t}{\displaystyle \frac{1}{3}}(1+g+g^2)q^{L_0}\overline{q}^{\overline{L}_0}y^{J_0}\overline{y}^{\overline{J}_0}`$ (35)
$`=`$ $`3\times {\displaystyle \frac{1}{2\times 3}}{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{l=1}{\overset{1}{}}}\left(\left|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}\frac{\sigma }{3},\sigma )}{\vartheta _1(\frac{l}{3}\frac{\sigma }{3},\sigma )}}\right|^2+\left|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}+\frac{\sigma }{3},\sigma )}{\vartheta _1(\frac{l}{3}+\frac{\sigma }{3},\sigma )}}\right|^2\right).`$
Then the complete modular invariant $`_3`$ orbifold partition function is
$`Z_{_3orb}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(Z(\tau =e^{\frac{2\pi i}{3}},\rho ,z)+{\displaystyle \frac{3}{2}}{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{s=1}{\overset{2}{}}}|{\displaystyle \frac{\vartheta _i(z+\frac{s}{3},\sigma )}{\vartheta _1(\frac{s}{3},\sigma )}}|^2`$ (36)
$`+{\displaystyle \frac{3}{2}}{\displaystyle \underset{l=1}{\overset{1}{}}}{\displaystyle \underset{i=1}{\overset{4}{}}}\left(\right|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}\frac{\sigma }{3},\sigma )}{\vartheta _1(\frac{l}{3}\frac{\sigma }{3},\sigma )}}|^2+\left|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}+\frac{\sigma }{3}\sigma )}{\vartheta _1(\frac{l}{3}+\frac{\sigma }{3},\sigma )}}|^2\right)).`$
We find eight Ramond ground states with conformal dimension $`(h,\overline{h})=(1/8,1/8)`$ and with charges $`(\pm 1/2,\pm 1/2)`$, $`3\times (\pm 1/6,\pm 1/6)`$, eight NS chiral primary states with conformal dimensions $`(0,0)`$, $`(1/2,1/2)`$, $`3\times (1/6,1/6)`$, $`3\times (1/3,1/3)`$ and with charges $`(0,0)`$, $`(1,1)`$, $`3\times (1/3,1/3)`$, $`3\times (2/3,2/3)`$, as well as eight NS antichiral primary states having the same conformal dimensions but the opposite charges as the NS chiral fields. By (5) with $`\eta =1/2`$, the ground states of the Ramond sector flow to the (c, c) primary states of the NS sector, namely
Ramond ground states $``$ NS chiral states
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`1`$
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`q^{1/2}\overline{q}^{1/2}y\overline{y}`$
$`3\times q^{1/8}\overline{q}^{1/8}y^{1/6}\overline{y}^{1/6}`$ $``$ $`3\times q^{1/6}\overline{q}^{1/6}y^{1/3}\overline{y}^{1/3}`$
$`3\times q^{1/8}\overline{q}^{1/8}y^{1/6}\overline{y}^{1/6}`$ $``$ $`3\times q^{1/3}\overline{q}^{1/3}y^{2/3}\overline{y}^{2/3}.`$ (37)
(Here $`q=e^{2\pi i\sigma }`$ and $`y=e^{2\pi iz}`$.) If we revers the direction of the spectral flow, we get an isomorphism between the (a, a) primary states and the ground states of the Ramond sector. By (6), (7) and (4) the Witten index and the Poincaré polynomial for the (c,c) states are
$$Tr(1)^F=8,P(t,\overline{t})_{(c,c)}=1+t\overline{t}+3t^{\frac{1}{3}}\overline{t}^{\frac{1}{3}}+3t^{\frac{2}{3}}\overline{t}^{\frac{2}{3}}.$$
(38)
The spectral flow from the NS sector to the NS sector can be obtained by flow parameter $`\eta =1`$. In the spectrum, there are no nontrivial (a, c) or its conjugate (c, a) states.
## 5 The $`_4`$ Orbifold
In this section, by dividing the $`_4`$ symmetry from (15) for $`\tau =i`$ and arbitrary $`\rho `$, we may formulate $`_4`$ orbifold model. The action of $`g_4`$ on the bosonic ground state sectors is given by
$$g|m_1,m_2,n_1,n_2=|m_2,m_1,n_2,n_1.$$
(39)
Under the rotation group $`_4`$ the square lattice has three fixed points. An analysis similar to the $`_3`$ orbifold shows there are twisted sectors associated with those fixed points, namely one fixed point corresponds to the $`_2`$ twist and two for the $`_4`$ twist. The weight of the bosonic and fermionic $`_4`$ twisted ground state is $`(\frac{3}{32},\frac{3}{32})`$ and $`(\frac{1}{32},\frac{1}{32})`$, respectively. Thus the total conformal weight of the $`_4`$ twisted sector is then $`(\frac{1}{8},\frac{1}{8})`$. The total $`_4`$ orbifold partition function can be obtained by summing over untwisted, $`_2`$, and $`_4`$ twisted sectors partition functions
$$Z_{_4orb}(\tau =i,\rho ,z)=Z_u+Z_{2t}+Z_{4t}.$$
By (16), (17), (20), (23), (39), and (24), we obtain the following untwisted sector partition function
$`Z_u`$ $`=`$ $`(q\overline{q})^{\frac{1}{8}}tr_{\stackrel{~}{}_u}{\displaystyle \frac{1}{4}}(1+g+g^2+g^3)q^{L_0}\overline{q}^{\overline{L}_0}y^{J_0}\overline{y}^{\overline{J}_0}`$
$`=`$ $`{\displaystyle \frac{1}{4}}\left(Z(\tau =i,\rho ,z)+{\displaystyle \underset{i=1}{\overset{4}{}}}\left|{\displaystyle \frac{\vartheta _i(z,\sigma )}{\vartheta _2(\sigma )}}\right|^2+{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{s=1}{\overset{3}{}}}\left|{\displaystyle \frac{\vartheta _i(z+\frac{s}{4},\sigma )}{\vartheta _1(\frac{s}{4},\sigma )}}\right|^2\right).`$
By (16), (17), (27), (28), (39), and (29), $`_4`$ twisted sector partition function may has the form
$$Z_{4t}=\frac{1}{4}\underset{i,l=1}{\overset{4}{}}\left(\left|y^{\frac{1}{4}}\frac{\vartheta _i(z+\frac{l}{4}\frac{\sigma }{4},\sigma )}{\vartheta _1(\frac{l}{4}\frac{\sigma }{4},\sigma )}\right|^2+\left|y^{\frac{1}{4}}\frac{\vartheta _i(z+\frac{l}{4}+\frac{\sigma }{4},\sigma )}{\vartheta _1(\frac{l}{4}+\frac{\sigma }{4},\sigma )}\right|^2+\left|\frac{\vartheta _i(z+\frac{l}{4},\sigma )}{\vartheta _4(\frac{l}{4},\sigma )}\right|^2\right).$$
The $`_2`$ twisted sector partition function can be read off from (32) by ommiting the factor of four. Thus, we may write the modular invariant $`_4`$ orbifold partition function in the following form
$`Z_{_4orb}`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{i,l=1}{\overset{4}{}}}(Z(\tau =i,\rho ,z)+{\displaystyle \underset{j=2}{\overset{4}{}}}|{\displaystyle \frac{\vartheta _i(z,\sigma )}{\vartheta _j(\sigma )}}|^2+{\displaystyle \underset{s=1}{\overset{3}{}}}|{\displaystyle \frac{\vartheta _i(z+\frac{s}{4},\sigma )}{\vartheta _1(\frac{s}{4},\sigma )}}|^2+`$ (40)
$`|{\displaystyle \frac{\vartheta _i(z+\frac{l}{4},\sigma )}{\vartheta _4(\frac{l}{4},\sigma )}}|^2+|y^{\frac{1}{4}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{4}\frac{\sigma }{4},\sigma )}{\vartheta _1(\frac{l}{4}\frac{\sigma }{4},\sigma )}}|^2+\left|y^{\frac{1}{4}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{4}+\frac{\sigma }{4},\sigma )}{\vartheta _1(\frac{l}{4}+\frac{\sigma }{4},\sigma )}}|^2\right).`$
In the spectrum there are nine Ramond ground states which flow to the NS chiral states under the spectral flow operation (5)with flow parameter $`\eta =1/2`$
Ramond ground states $``$ NS chiral states
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`1`$
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`q^{1/2}\overline{q}^{1/2}y\overline{y}`$
$`2\times q^{1/8}\overline{q}^{1/8}y^{1/4}\overline{y}^{1/4}`$ $``$ $`2\times q^{1/8}\overline{q}^{1/8}y^{1/4}\overline{y}^{1/4}`$
$`2\times q^{1/8}\overline{q}^{1/8}y^{1/4}\overline{y}^{1/4}`$ $``$ $`2\times q^{3/8}\overline{q}^{3/8}y^{3/4}\overline{y}^{3/4}`$
$`3\times q^{1/8}\overline{q}^{1/8}`$ $``$ $`3\times q^{1/4}\overline{q}^{1/4}y^{1/2}\overline{y}^{1/2}.`$ (41)
There are nine $`(a,a)`$ states which are given by the complex conjugation of $`(c,c)`$ states. As in the $`_3`$ orbifold case, one can get isomorphism between the $`(a,a)`$ primary states and the ground states of Ramond sector by reversing the direction of the spectral flow. By (6), (7) and (5), the Witten index and the Poincaré polynomial for the (c,c) states are
$`Tr(1)^F`$ $`=`$ $`9`$
$`P(t,\overline{t})_{(c,c)}`$ $`=`$ $`1+t\overline{t}+3t^{\frac{1}{2}}\overline{t}^{\frac{1}{2}}+2t^{\frac{1}{4}}\overline{t}^{\frac{1}{4}}+2t^{\frac{3}{4}}\overline{t}^{\frac{3}{4}}.`$
With the spectral flow parameter $`\eta =1`$, the NS sector comes back to the NS sector. One notes that the $`_4`$ orbifold model contains only (c, c) and their conjugate (a,a) states. For this model, the (a,c) and (c,a) states are trivial and consist only of the vacumm state.
## 6 The $`_6`$ Orbifold
We now construct a $`_6`$ orbifold model by dividing $`_6`$ symmetry from (15) for $`\tau =e^{2\pi i/3}`$ and arbitrary $`\rho `$. The bosonic ground state sectors transform as follows under the action of $`g_6`$
$$g|m_1,m_2,n_1,n_2=|m_1+m_2,m_1,n_2,n_1+n_2.$$
(43)
The hexagonal torus has three fixed points under the $`_6`$ rotation symmetry. There is a twisted sector associated with each of them. These are $`_2`$, $`_3`$ and $`_6`$ twisted sectors. The conformal dimension of the bosonic and fermionic $`_6`$ twisted ground state is $`(\frac{5}{72},\frac{5}{72})`$ and $`(\frac{1}{72},\frac{1}{72})`$, respectively. Thus the total conformal weight of the $`_6`$ twisted ground state is then $`(\frac{1}{12},\frac{1}{12})`$. The $`_6`$ orbifold partition function is the sum of partition functions of the untwisted, $`_2`$, $`_3`$, and $`_6`$ twisted sectors
$`Z_{_6orb}(\tau =e^{\frac{2\pi i}{3}},\rho ,z)=Z_u+Z_{2t}+Z_{3t}+Z_{6t}.`$
By applying the same method as for the construction of the $`_2`$, $`_3`$ and $`_4`$ orbifolds, we obtain the following untwisted $`_6`$ orbifold partition function
$`Z_u`$ $`=`$ $`(q\overline{q})^{\frac{1}{8}}tr_{\stackrel{~}{}_u}{\displaystyle \frac{1}{6}}(1+g+\mathrm{}+g^5)q^{L_0}\overline{q}^{\overline{L}_0}y^{J_0}\overline{y}^{\overline{J}_0}`$
$`=`$ $`{\displaystyle \frac{1}{6}}(Z(\tau =e^{\frac{2\pi i}{3}},\rho ,z)+{\displaystyle \frac{3}{2}}{\displaystyle \underset{i=1}{\overset{4}{}}}|{\displaystyle \frac{\vartheta _i(z,\sigma )}{\vartheta _2(\sigma )}}|^2`$
$`+{\displaystyle \frac{3}{2}}{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{s=1}{\overset{2}{}}}|{\displaystyle \frac{\vartheta _i(z+\frac{s}{3},\sigma )}{\vartheta _1(\frac{s}{3},\sigma )}}|^2+{\displaystyle \frac{1}{2}}{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{l=1}{\overset{1}{}}}|{\displaystyle \frac{\vartheta _i(z+\frac{l}{3},\sigma )}{\vartheta _2(\frac{l}{3},\sigma )}}|^2).`$
The $`_2`$ and $`_3`$ twisted sector partition functions can be read off from (32) and (35) by ommiting the factor of four and three, respectively. The $`_6`$ twisted sector partition function may have the form
$`Z_{6t}`$ $`=`$ $`{\displaystyle \frac{1}{12}}{\displaystyle \underset{i,k=1}{\overset{4}{}}}{\displaystyle \underset{l=1}{\overset{1}{}}}\left(\right|{\displaystyle \frac{\vartheta _i(z+\frac{l}{3},\sigma )}{\vartheta _3(\frac{l}{3},\sigma )}}|^2+|{\displaystyle \frac{\vartheta _i(z+\frac{l}{3},\sigma )}{\vartheta _4(\frac{l}{3},\sigma )}}|^2`$
$`+|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}+\frac{\sigma }{3},\sigma )}{\vartheta _k(\frac{l}{3}+\frac{\sigma }{3},\sigma )}}|^2+|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}\frac{\sigma }{3},\sigma )}{\vartheta _k(\frac{l}{3}\frac{\sigma }{3},\sigma )}}|^2).`$
All in all we obtain the following modular invariant partition function
$`Z_{_6orb}`$ $`=`$ $`{\displaystyle \frac{1}{6}}{\displaystyle \underset{i=1}{\overset{4}{}}}{\displaystyle \underset{j=2}{\overset{4}{}}}(Z(\tau =e^{\frac{2\pi i}{3}},\rho ,z)+{\displaystyle \frac{3}{2}}|{\displaystyle \frac{\vartheta _i(z,\sigma )}{\vartheta _j(\sigma )}}|^2+{\displaystyle \frac{3}{2}}{\displaystyle \underset{s=1}{\overset{2}{}}}|{\displaystyle \frac{\vartheta _i(z+\frac{s}{3},\sigma )}{\vartheta _1(\frac{s}{3},\sigma )}}|^2`$
$`+{\displaystyle \frac{1}{2}}{\displaystyle \underset{l=1}{\overset{1}{}}}\left(2\right|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}\frac{\sigma }{3},\sigma )}{\vartheta _1(\frac{l}{3}\frac{\sigma }{3},\sigma )}}|^2+\mathrm{\hspace{0.33em}2}|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}+\frac{\sigma }{3},\sigma )}{\vartheta _1(\frac{l}{3}+\frac{\sigma }{3},\sigma )}}|^2`$
$`+|{\displaystyle \frac{\vartheta _i(z+\frac{l}{3},\sigma )}{\vartheta _j(\frac{l}{3},\sigma )}}|^2+|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}\frac{\sigma }{3},\sigma )}{\vartheta _j(\frac{l}{3}\frac{\sigma }{3},\sigma )}}|^2+\left|y^{\frac{1}{3}}{\displaystyle \frac{\vartheta _i(z+\frac{l}{3}+\frac{\sigma }{3},\sigma )}{\vartheta _j(\frac{l}{3}+\frac{\sigma }{3},\sigma )}}|^2\right)).`$
In this model there are ten Ramond ground states. Again we connect the ground states of Ramond sector with NS chiral primary states using eq. (5) with $`\eta =1/2`$
Ramond ground states $``$ NS chiral states
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`1`$
$`q^{1/8}\overline{q}^{1/8}y^{1/2}\overline{y}^{1/2}`$ $``$ $`q^{1/2}\overline{q}^{1/2}y\overline{y}`$
$`2\times q^{1/8}\overline{q}^{1/8}`$ $``$ $`2\times q^{1/4}\overline{q}^{1/4}y^{1/2}\overline{y}^{1/2}`$
$`q^{1/8}\overline{q}^{1/8}y^{1/3}\overline{y}^{1/3}`$ $``$ $`q^{5/12}\overline{q}^{5/12}y^{5/6}\overline{y}^{5/6}`$
$`q^{1/8}\overline{q}^{1/8}y^{1/3}\overline{y}^{1/3}`$ $``$ $`q^{1/12}\overline{q}^{1/12}y^{1/6}\overline{y}^{1/6}`$
$`2\times q^{1/8}\overline{q}^{1/8}y^{1/6}\overline{y}^{1/6}`$ $``$ $`2\times q^{1/3}\overline{q}^{1/3}y^{2/3}\overline{y}^{2/3}`$
$`2\times q^{1/8}\overline{q}^{1/8}y^{1/6}\overline{y}^{1/6}`$ $``$ $`2\times q^{1/6}\overline{q}^{1/6}y^{1/3}\overline{y}^{1/3}.`$ (45)
By (6), (7) and (6), The Witten index and the Poincaré polynomials for the (c, c) states are
$`Tr(1)^F`$ $`=`$ $`10`$
$`P(t,\overline{t})_{(c,c)}`$ $`=`$ $`1+t\overline{t}+2t^{\frac{1}{2}}\overline{t}^{\frac{1}{2}}+t^{\frac{5}{6}}\overline{t}^{\frac{5}{6}}+t^{\frac{1}{6}}\overline{t}^{\frac{1}{6}}+2t^{\frac{2}{3}}\overline{t}^{\frac{2}{3}}+2t^{\frac{1}{3}}\overline{t}^{\frac{1}{3}}.`$ (46)
The (a,a) states are given by the complex conjugation of (c,c) states. We found no (a, c) and (c,a) states in this model.
## 7 $`N=2`$ Landau-Ginzburg Model
In this section, we first review some of the facts of the $`N=2`$ superconformal Landau-Ginzburg theories by following the articles , then we check the spectrum of the (c,c) fields and the Witten index. The $`N=2`$ superconformal Landau-Ginzburg action takes the following form
$$S=d^2zd^4\theta K(\mathrm{\Phi }_i,\overline{\mathrm{\Phi }}_i)+(d^2zd^2\theta W(\mathrm{\Phi }_i)+h.c).$$
(47)
$`\mathrm{\Phi }_i`$ $`(i=1,2,\mathrm{}n)`$ are the $`N=2`$ $`n`$ chiral scalar superfields which satisfy the condition $`\overline{D}_\pm \mathrm{\Phi }_i=D_\pm \overline{\mathrm{\Phi }}_i=0`$, where the superderivative defined as $`D_\pm =\frac{}{\theta ^\pm }+\theta ^{}\frac{}{z}`$. The first term (K) is called Kähler potential. It includes derivatives of the superfields. The conformal dimension of those fields is greater than $`(1,1)`$. Such fields are called irrelevant. The second term (W) is called superpotential which is a holomorphic function of the superfields. It contains only relevant fields, i.e. fields with conformal dimension $`(1,1)`$ or less than $`(1,1)`$. The holomorphic superpotential $`W(\mathrm{\Phi }_i)`$ is a quasi-homogeneous function with isolated singularities at $`\mathrm{\Phi }_i=0`$. In other words $`W(\mathrm{\Phi }_i)`$ is called quasi-homogeneous if it satisfies
$$W(\lambda ^{w_i}\mathrm{\Phi }_i)=\lambda ^dW(\mathrm{\Phi }_i),for\mathrm{\Phi }_i\lambda ^{w_i}\mathrm{\Phi }_i,$$
(48)
where $`w^i`$ and d are integers with no commen factors. It has isolated singularity at $`\mathrm{\Phi }_i=0`$ if it satisfies
$$W(\mathrm{\Phi }_i)|_0=0,_iW(\mathrm{\Phi }_j)|_0=0.$$
For every isolated quasi-homegeneous superpotential, there exists an $`N=2`$ superconformal field theory. One can read off the $`U(1)`$ charge of the lowest component of the chiral superfields $`\mathrm{\Phi }_i`$ from the action $`(\text{47})`$. The $`\theta `$ integrals in the first term have $`(left,right)`$ charges $`(1,1)`$. Because of neutrality of the action $`W(\mathrm{\Phi }_i)`$ has charge $`(1,1)`$. Thus, the chiral superfield $`\mathrm{\Phi }_i`$ must have charge $`q_i=\frac{w_i}{d}`$ for both its left-right moving components. Now one notes that for any state in the Landau-Ginzburg theory $`q_Lq_R`$ is always an integer. This is true for the chiral superfield $`\mathrm{\Phi }_i`$, as it has equal left-right charges. Moreover, it is also true for the most general fields because they are obtained by taking products of $`\mathrm{\Phi }_i`$ and $`\overline{\mathrm{\Phi }}_i`$, as well as products of their super derivatives. This implies that one can apply spectral flow to the Landau-Ginzburg models.
The local ring $``$ of the superpotential $`W(\mathrm{\Phi }_i)`$ of the Landau-Ginzburg model is obtained by taking into account all monomials of chiral superfields $`\mathrm{\Phi }_i`$ and setting $`_iW(\mathrm{\Phi }_j)|_0=0`$. The number of elements of the ring is denoted by $`\mu =dim`$. It is called multiplicity of $`W(\mathrm{\Phi }_i)`$. It is also equal to the Witten index $`Tr(1)^F`$.
The modality (or moduli is the number of free parameters in the theory.) m of a quasi-homogeneous superpotentials with isolated singularities is given by the number of chiral primary states with charge greater than or equal to one.
The Poincaré Polynomial for the Landau-Ginzburg theories is
$$P(t)=Tr_{}t^{dJ_0}=\underset{i=1}{\overset{n}{}}\frac{1t^{dw_i}}{1t^{w_i}},\text{ or }P(t,\overline{t})=Tr_{}t^{J_0}\overline{t}^{\overline{J}_0}.$$
(49)
This polynomial is only function of $`t\overline{t}`$. (Because Landau-Ginzburg primary chiral fields have equal left-right charges.) For convenience, $`t\overline{t}`$ is replaced by the variable $`t^d`$, where $`d`$ is defined in $`(\text{48})`$. The Witten index is
$$Tr(1)^F=P(t=1)=\mu =\underset{i=1}{\overset{n}{}}\frac{dw_i}{w_i}=\underset{i=1}{\overset{n}{}}\left(\frac{1}{q_i}1\right).$$
(50)
The highest charge and conformal dimension of chiral primary state $`|\chi `$ are given as
$$q_\chi =\underset{i=1}{\overset{\mathrm{}}{}}\frac{d2w_i}{d}=\underset{i}{}(12q_i),h_\chi =\frac{q_\chi }{2}=\underset{i=1}{\overset{n}{}}\left(\frac{1}{2}q_i\right).$$
By using $`h_\chi =\frac{c}{6}`$, the central charge of the Landau-Ginzburg theory is given as
$$c=6h_\chi =6\underset{i=1}{\overset{n}{}}\left(\frac{1}{2}q_i\right).$$
It is well known that the quasi-homogeneous superpotentials with isolated singularities for modality $`m=1`$ of the Landau-Ginzburg theories at $`c=3`$ are equivalent to the $`_M`$, $`M\{3,4,6\}`$, orbifolds of the $`N=2`$ theories at $`c=3`$. The corresponding superpotentials are given as
$`W_3(\mathrm{\Phi }_1,\mathrm{\Phi }_2,\mathrm{\Phi }_3)`$ $`=`$ $`\mathrm{\Phi }_1^3+\mathrm{\Phi }_2^3+\mathrm{\Phi }_3^3+6a\mathrm{\Phi }_1\mathrm{\Phi }_2\mathrm{\Phi }_3,a^3+270`$ (51)
$`W_4(\mathrm{\Phi }_1,\mathrm{\Phi }_2)`$ $`=`$ $`\mathrm{\Phi }_1^4+\mathrm{\Phi }_2^4+a\mathrm{\Phi }_1^2\mathrm{\Phi }_2^2,a^24`$ (52)
$`W_6(\mathrm{\Phi }_1,\mathrm{\Phi }_2)`$ $`=`$ $`\mathrm{\Phi }_1^3+\mathrm{\Phi }_2^6+a\mathrm{\Phi }_1^2\mathrm{\Phi }_2^2,4a^3+270.`$ (53)
With the knowledge in this section, we may write the basic linearly independent elements of the (c,c) ring of superpotential (51) in the following form
| chiral fields | 1 | $`\mathrm{\Phi }_1`$ | $`\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_3`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_3`$ | $`\mathrm{\Phi }_2\mathrm{\Phi }_3`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2\mathrm{\Phi }_3`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| charges | 0 | 1/3 | 1/3 | 1/3 | 2/3 | 2/3 | 2/3 | 1 |
| dimensions | 0 | 1/6 | 1/6 | 1/6 | 1/3 | 1/3 | 1/3 | 1/2. |
(54)
By (50), (49) and (54), the Witten index and Poincaré polynomial are
$$Tr(1)^F=8,P(t,\overline{t})_{(c,c)}=Tr_{}t^{J_0}\overline{t}^{\overline{J}_0}=1+t\overline{t}+3t^{\frac{1}{3}}\overline{t}^{\frac{1}{3}}+3t^{\frac{2}{3}}\overline{t}^{\frac{2}{3}}.$$
(55)
For the superpotential (52) we have
| chiral fields | 1 | $`\mathrm{\Phi }_1`$ | $`\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_1^2`$ | $`\mathrm{\Phi }_2^2`$ | $`\mathrm{\Phi }_1^2\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2^2`$ | $`\mathrm{\Phi }_1^2\mathrm{\Phi }_2^2`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| charges | 0 | 1/4 | 1/4 | 1/2 | 1/2 | 1/2 | 3/4 | 3/4 | 1 |
| dimensions | 0 | 1/8 | 1/8 | 1/4 | 1/4 | 1/4 | 3/8 | 3/8 | 1/2 . |
(56)
By (50), (49) and (56), the Witten index and Poincaré polynomial are
$$Tr(1)^F=9,P(t,\overline{t})_{(c,c)}=1+t\overline{t}+3t^{\frac{1}{2}}\overline{t}^{\frac{1}{2}}+2t^{\frac{1}{4}}\overline{t}^{\frac{1}{4}}+2t^{\frac{3}{4}}\overline{t}^{\frac{3}{4}}$$
(57)
Similarly, for the superpotential (53) we may get
| chiral fields | 1 | $`\mathrm{\Phi }_1`$ | $`\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2`$ | $`\mathrm{\Phi }_2^2`$ | $`\mathrm{\Phi }_2^3`$ | $`\mathrm{\Phi }_2^4`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2^2`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2^3`$ | $`\mathrm{\Phi }_1\mathrm{\Phi }_2^4`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| charges | 0 | 1/6 | 1/3 | 1/2 | 2/3 | 1/3 | 1/2 | 2/3 | 5/6 | 1 |
| dimensions | 0 | 1/12 | 1/6 | 1/4 | 1/3 | 1/6 | 1/4 | 1/3 | 5/12 | 1/2 . |
(58)
By (50), (49) and (58), the Witten index and Poincaré polynomial are
$$Tr(1)^F=10,P(t,\overline{t})_{(c,c)}=1+t\overline{t}+2t^{\frac{1}{2}}\overline{t}^{\frac{1}{2}}+t^{\frac{5}{6}}\overline{t}^{\frac{5}{6}}+t^{\frac{1}{6}}\overline{t}^{\frac{1}{6}}+2t^{\frac{2}{3}}\overline{t}^{\frac{2}{3}}+2t^{\frac{1}{3}}\overline{t}^{\frac{1}{3}}.$$
(59)
### Conclusion
The partition functions for $`_M`$ orbifolds have been calculated. The Witten indexes, the spectrum of (chiral, chiral) fields for the $`_M`$, $`M\{3,4,6\}`$, orbifolds and for the Landau-Ginzburg superpotentials (5153) are given in equations (38), (5), (6), (4), (5), (6) and (54), (56), (58), (55), (57), (59), respectively. The results are in in agreement with the Landau-Ginzburg predictions of C. Vafa and N. Warner.
### Acknowledgements
It is great pleasure to thank my supervisor Professor Werner Nahm for countless very helpful and very encouraging discussions. I would like to thank K. Wendland for numerous helpful discussions. I also would like to thank D. Brungs for his help with Mathematica. I am grateful to M. Soika for his help with Latex, and for proof reading, as well as for his constant hospitality. I am also grateful to H. Eberle for proof reading.
This work was supported by Deutscher Akademischer Austauschdienst (DAAD) and in part by TMR.
|
warning/0002/cond-mat0002321.html
|
ar5iv
|
text
|
# Single hole dynamics in the 𝑡-𝐽 model on a square lattice
## I Introduction
Since the pioneering work by Brinkman and Rice (BR) the dynamics of a hole in an antiferromagnet remained as a recurring open problem in condensed matter physics. After the discovery of high temperature superconductors and the suggestions by Anderson on the possibility of a non-Fermi liquid state in those materials, the question whether the quasiparticle weight of a hole vanishes due to the interaction with an antiferromagnetic background became central in the field of strongly correlated fermions.
The BR treatment led to a fully incoherent spectrum in the so-called retraceable path approximation, for an antiferromagnetic Ising-like background, in the limit $`J_z0`$. The retraceable path approximation is exact in one and in infinite dimensions but not in two dimensions since contributions of loops (Trugman paths ) may lead to a coherent propagation of the hole. Furthermore, for an Ising-like background it was shown within a Lanczos scheme , that a finite quasiparticle weight is obtained. For the case of physical interest, namely with a Heisenberg spin-background, a large number of numerical methods led to conflicting results. Whereas exact diagonalizations found large quasiparticle peaks at the lower edge of the spectrum , quantum Monte Carlo (QMC) results were interpreted as leading to a vanishing quasiparticle weight . Since exact diagonalizations are possible only on very small lattices, finite size scaling cannot be performed reliably. On the other hand, QMC simulations suffered from the minus-sign problem, such that scaling was not possible with reasonable confidence. Further studies based on the self-consistent Born approximation (SCBA) gave a finite quasiparticle weight. However, since there fluctuations of the spin-background are only taken into account in the frame of a spin-wave approximation, the results obtained are not conclusive. Exact results for the supersymmetric point $`J=2t`$ were obtained by Sorella , that give important benchmarks for any analytical or numerical method (see Sec. III B), but unfortunately, they cannot be rigorously extended to the physical relevant parameter range $`J0.4t`$.
Quite recently, the dynamics of a single hole in an antiferromagnetic background became experimentally accessible by angle resolved photoemission spectroscopy (ARPES) in undoped materials like Sr<sub>2</sub>CuO<sub>2</sub>Cl<sub>2</sub> and Ca<sub>2</sub>CuO<sub>2</sub>Cl<sub>2</sub> . The main features observed there are a minimum of the dispersion at $`\stackrel{}{k}=(\pi /2,\pi /2)`$ together with a vanishing of spectral weight beyond this point along the (1,1) direction. The obtained spectra show that the very flat portion around $`(\pi ,0)`$, that in optimally doped materials is almost degenerate with the bottom of the spectrum at $`(\pi /2,\pi /2)`$ , is shifted upwards (in a hole representation) by approximately 300 meV. This contradicts the single hole spectra found theoretically so far, where essentially the lower edge of the spectrum at $`\stackrel{}{k}=(\pi /2,\pi /2)`$ and $`(\pi ,0)`$ are almost degenerate, such that additional second and third nearest neighbor hopping terms were suggested , that lead to an agreement of the exact diagonalization results with experiments. Such terms were made recently responsible also for the vanishing of spectral weight close to $`(\pi /2,\pi /2)`$ by reducing the quasiparticle weight .
In the following we present dynamical properties of a single hole in a two-dimensional $`t`$-$`J`$ model on lattices with up to $`24\times 24`$ sites in the parameter range $`0.4J/t4`$. Results were obtained with a new QMC algorithm, where the spin background is simulated with a loop-algorithm and the hole is exactly propagated for a given configuration of the spin background. The lower edge of the spectrum is obtained directly from the asymptotic form of the imaginary time Green’s function. The resulting dispersion agrees with previous results obtained within SCBA and series expansions for $`J/t<1`$, whereas for $`J/t>1`$ only agreement with series expansions is found. In particular, a flat dispersion is obtained around $`\stackrel{}{k}=(\pi ,0)`$ very close in value to the bottom of the band at $`\stackrel{}{k}=(\pi /2,\pi /2)`$, in contrast to the experiments . The asymptotics of the imaginary time Green’s function delivers also the quasiparticle weight for that band. Finite size scaling is presented showing that $`Z(\stackrel{}{k})`$ is finite for the parameter range considered, such that the lower edge of the spectrum corresponds to a coherent quasiparticle. Furthermore, our data are consistent with another exact prediction , namely that at the supersymmetric point and in the thermodynamic limit, $`Z(\stackrel{}{Q})/Z(0)=(2m)^2`$, where $`\stackrel{}{Q}=(\pi ,\pi )`$ is the antiferromagnetic wave vector and $`m`$ is the staggered magnetization. The spectral function $`A(\stackrel{}{k},\omega )`$ is calculated by analytic continuation with maximum entropy (MaxEnt) . Overall agreement is found with exact diagonalizations. At the supersymmetric point, the delta function predicted by Sorella for the wave vector $`\stackrel{}{k}=(0,0)`$ is exactly reproduced. By extracting the contribution of the quasiparticle from the imaginary time Green’s function, a resonance above the quasiparticle band is made evident, that together with the lower edge of the spectrum scales as $`J^{2/3}`$, in agreement with the string picture used to described the excitations for a hole in an antiferromagnetic Ising-background. Remarkably, also the prefactors of the corresponding Airy functions are needed in order to properly describe the distance between the resonance and the quasiparticle band.
The paper is organized as follows. Section II describes the model, a canonical transformation that leads to a bilinear form in spinless fermions interacting with $`S=\frac{1}{2}`$ pseudospins, and the algorithm. Since the Hamiltonian for the transformed $`t`$-$`J`$ model is bilinear in fermions (the holes), their propagation can be calculated exactly given a pseudospin configuration. In Sec. III the results are discussed. Section III A describes the lower edge of the spectrum and how it is obtained. In Sec. III B the results for the quasiparticle weight are shown. Section III C describes the spectral function $`A(\stackrel{}{k},\omega )`$ and the string excitations. Finally, the conclusions are given in Sec. IV.
## II The model and the algorithm
The $`t`$-$`J`$ model is a suitable one to simulate the dynamics of a single hole in an antiferromagnet. On the one side, it can be obtained from the Hubbard model in the large coupling limit, which at half-filling leads to the Heisenberg antiferromagnet. On the other side, it is the relevant one to simulate the cuprates, as shown by Zhang and Rice , and hence, to compare with experiments . Its Hamiltonian is
$$H_{tJ}=t\underset{<i,j>,\sigma }{}\stackrel{~}{c}_{i,\sigma }^{}\stackrel{~}{c}_{j,\sigma }+J\underset{<i,j>}{}\left(\stackrel{}{S}_i\stackrel{}{S}_j\frac{1}{4}\stackrel{~}{n}_i\stackrel{~}{n}_j\right),$$
(1)
where $`\stackrel{~}{c}_{i,\sigma }^{}`$ are projected fermion operators $`\stackrel{~}{c}_{i,\sigma }^{}=(1c_{i,\sigma }^{}c_{i,\sigma })c_{i,\sigma }^{}`$ , $`\stackrel{~}{n}_i=\underset{\alpha }{}\stackrel{~}{c}_{i,\alpha }^{}\stackrel{~}{c}_{i,\alpha }`$, $`\stackrel{}{S}_i=(1/2)\underset{\alpha ,\beta }{}c_{i,\alpha }^{}\stackrel{}{\sigma }_{\alpha ,\beta }c_{i,\beta }`$, and the sum runs over nearest neighbours only. In order to render Eq. (1) a bilinear form in fermionic operators, we perform a canonical transformation
$$c_i^{}=\gamma _{i,+}f_i\gamma _{i,}f_i^{},c_i^{}=\sigma _{i,}(f_i+f_i^{}),$$
(2)
where $`\gamma _{i,\pm }=(1\pm \sigma _{i,z})/2`$ and $`\sigma _{i,\pm }=(\sigma _{i,x}\pm i\sigma _{i,y})/2`$. The spinless fermion operators fulfill the canonical anticommutation relations $`\{f_i^{},f_j\}=\delta _{i,j}`$, and $`\sigma _{i,a},a=x,y,`$ or $`z`$, are the Pauli matrices. The Hamiltonian becomes
$$\stackrel{~}{H}_{tJ}=+t\underset{<i,j>}{}P_{ij}f_i^{}f_j+\frac{J}{2}\underset{<i,j>}{}\mathrm{\Delta }_{ij}(P_{ij}1),$$
(3)
where $`P_{ij}=(1+\stackrel{}{\sigma }_i\stackrel{}{\sigma }_j)/2`$, $`\mathrm{\Delta }_{ij}=(1n_in_j)`$ and $`n_i=f_i^{}f_i`$. The constraint to avoid doubly occupied states transforms to the conserved and holonomic constraint $`_i\gamma _{i,}f_i^{}f_i=0`$. This constraint simply means, that a spinless fermion and a pseudospin $``$ are not allowed to sit on the same site.
In order to obtain the dynamics of the hole, we calculate the one-particle Green’s function for spin up,
$$G(ij,\tau )=T\stackrel{~}{c}_{i,}(\tau )\stackrel{~}{c}_{j,}^{}=Tf_i^{}(\tau )f_j$$
(4)
where $`T`$ corresponds to the time ordering operator. Inserting complete sets of spin states the quantity above transforms as
$`G(ij,\tau )`$ $`=`$ $`{\displaystyle \frac{\underset{\sigma _1}{}v|\sigma _1|e^{(\beta \tau )\stackrel{~}{H}_{tJ}}f_je^{\tau \stackrel{~}{H}_{tJ}}f_i^{}|\sigma _1|v}{\underset{\sigma _1}{}\sigma _1|e^{\beta \stackrel{~}{H}_{tJ}}|\sigma _1}}`$ (5)
$`=`$ $`{\displaystyle \underset{\stackrel{}{\sigma }}{}}P(\stackrel{}{\sigma })\times {\displaystyle \frac{v|f_je^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _n,\sigma _{n1})}\mathrm{}e^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _2,\sigma _1)}f_i^{}|v}{\sigma _n|e^{\mathrm{\Delta }\tau \stackrel{~}{H}_{tJ}}|\sigma _{n1}\mathrm{}\sigma _2|e^{\mathrm{\Delta }\tau \stackrel{~}{H}_{tJ}}|\sigma _1}}+𝒪(\mathrm{\Delta }\tau ^2)`$ (6)
$`=`$ $`{\displaystyle \underset{\stackrel{}{\sigma }}{}}P(\stackrel{}{\sigma })G(i,j,\tau ,\stackrel{}{\sigma })+𝒪(\mathrm{\Delta }\tau ^2)`$ (7)
Here $`m\mathrm{\Delta }\tau =\beta `$, $`n\mathrm{\Delta }\tau =\tau `$, $`\mathrm{\Delta }\tau t1`$ and $`\mathrm{exp}\left(\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _1,\sigma _2)\right)`$ is the evolution operator for the holes, given the spin configuration $`(\sigma _1,\sigma _2)`$. In the case of single hole dynamics, $`|v`$ is the vacuum state for holes, and $`P(\stackrel{}{\sigma })`$ is the probability distribution of a Heisenberg antiferromagnet for the configuration $`\stackrel{}{\sigma }`$, where $`\stackrel{}{\sigma }`$ is a vector containing all intermediate states $`(\sigma _1,\mathrm{}\sigma _n,\mathrm{},\sigma _1)`$. The sum over spins is performed in a very efficient way by using a world-line loop-algorithm for a Heisenberg antiferromagnet with discretized imaginary time. In general we have $`\mathrm{\Delta }\tau =0.05`$, such that the extrapolation to $`\mathrm{\Delta }\tau =0`$ leads to values of the observables within the statistical error bars. The inverse temperature $`\beta `$ is taken such that the energy is well converged ($`\beta J20`$ for $`16\times 16`$ and $`\beta J=30`$ for $`24\times 24`$ sites), and therefore, the data correspond to the ground-state. As the evolution operator for the holes is a bilinear form in the fermion operators, $`G(i,j,\tau ,\stackrel{}{\sigma })`$ can be calculated exactly, in contrast to a direct implementation in the loop algorithm , where fermion paths are sampled stochastically. $`G(i,j,\tau ,\stackrel{}{\sigma })`$ contains a sum over all possible fermion paths between $`(i,0)`$ and $`(j,\tau )`$. The numerical effort to calculate $`G(i,j,\tau ,\stackrel{}{\sigma })i,\tau `$ scales as $`N\tau `$, where N is the number of lattice points in space. Therefore, the present method is more efficient for large systems than e.g. projector algorithms for the Hubbard model, that scale with the system size cubed.
With the representation of Eq. (3), the propagation of down spin electrons cannot be easily considered, since the operators $`\sigma _{i,\pm }`$ cut world-lines. This is certainly not a problem for finite-size systems, where SU(2) symmetry is conserved. Since $`P(\stackrel{}{\sigma })`$ is the probability distribution for the quantum antiferromagnet, the algorithm does not suffer from sign problems on bipartite lattices and non-frustrating magnetic interactions in any dimension.
We now address the explicit calculation of $`G(i,j,\tau ,\sigma )`$. In a first step, we introduce additional complete sets of single fermion states in the numerator of Eq.(7), such that $`G(i,j,\tau ,\stackrel{}{\sigma })`$ becomes
$`v|f_j\left({\displaystyle \underset{\stackrel{}{l}}{}}\right|f_{l_n}{\displaystyle \frac{f_{l_n}\left|e^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _n,\sigma _{n1})}\right|f_{l_{n1}}}{\sigma _n\left|e^{\mathrm{\Delta }\tau H}\right|\sigma _{n1}}}{\displaystyle \frac{f_{l_{n1}}\left|e^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _{n1},\sigma _{n2})}\right|f_{l_{n2}}}{\sigma _{n1}\left|e^{\mathrm{\Delta }\tau H}\right|\sigma _{n2}}}\mathrm{}`$ (8)
$`\times \mathrm{}{\displaystyle \frac{f_{l_2}\left|e^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _2,\sigma _1)}\right|f_{l_1}}{\sigma _2\left|e^{\mathrm{\Delta }\tau H}\right|\sigma _1}}f_{l_1}\left|\right)f_i^{}|v`$ (9)
$`={\displaystyle \underset{\stackrel{}{l}}{}}\left(\delta _j^{l_n}U(\sigma _n,\sigma _{n1})_{l_n}^{l_{n1}}\mathrm{}U(\sigma _2,\sigma _1)_{l_2}^{l_1}\delta _{l_1}^i\right),`$ (10)
where the sum $`_\stackrel{}{l}`$ runs over all possible intermediate one particle states in the fermionic Hilbert space $`\{|f_l\}`$. The propagators $`f_{l_p}\left|e^{\mathrm{\Delta }\tau \stackrel{~}{H}(\sigma _p,\sigma _{p1})}\right|f_{l_{p1}}/\sigma _p\left|e^{\mathrm{\Delta }\tau H}\right|\sigma _{p1}`$ are only nonzero, when $`l_p`$ and $`l_{p1}`$ belong to the same plaquette. Therefore the entries of the matrices $`U(\sigma _p,\sigma _{p1})`$ are nonzero only at the positions which correspond to a plaquette in the checkerboard breakup. These entries are given in Table I. As we are only interested in the Hilbert space of no double occupancy, we have to enforce the constraint at one single position of the propagation by projecting out the fermionic states which do not respect the constraint. We do so at $`\tau =0`$ corresponding to the first propagation.
The one-particle spectral function
$$A(\stackrel{}{k},\omega )=\underset{f,\sigma }{}\left|f,N1\left|c_{\stackrel{}{k},\sigma }\right|\mathrm{\hspace{0.17em}0},N\right|^2\delta \left(\omega E_0^N+E_f^{N1}\right),$$
(11)
is connected with the Green’s function in imaginary time at $`T=0`$, by the spectral theorem
$$G(\stackrel{}{k},\tau )=\underset{\mathrm{}}{\overset{\mathrm{}}{}}𝑑\omega \frac{\mathrm{exp}(\tau \omega )}{\pi }A(\stackrel{}{k},\omega ).$$
(12)
Here $`|\mathrm{\hspace{0.17em}0},N`$ is the ground-state at half filling with energy $`E_0^N`$ and $`|f,N1`$ are states in the $`N1`$ particle Hilbert space with energy $`E_f^{N1}`$. We perform the inversion of Eq. (12), that due to the statistical errors of $`G(\stackrel{}{k},\tau )`$ is an extremely ill-posed problem, by means of MaxEnt, where the $`A(\stackrel{}{k},\omega )`$ obtained is the one that maximizes the probability $`P(A|G)`$, given the Green’s function $`G(\stackrel{}{k},\tau )`$. Correlations in the imaginary time data were taken into account by considering the covariance matrix. Details about MaxEnt can be found in the comprehensive review article by J.E. Gubernatis and M. Jarrell .
We would like to stress finally, that part of the dynamical data presented below were obtained without use of MaxEnt but directly extracted from the imaginary time Green’s function. This is possible due to the high statistics and stability attainable with the present algorithm. The slowest decaying exponential, that corresponds to the excitation with lowest energy can be extracted simply by fitting the tail of the Green’s function at large values of $`\tau `$. This leads to the value of the excitation and its corresponding weight, as shown in Secs. III A and III B. Furthermore, in connection with MaxEnt, the next higher excitation can be obtained by subtracting the contribution from the lowest one from the Green’s function. This procedure is discussed in Sec. III C.
## III Results
We concentrate in the following on three aspects of the dynamics of a single hole in a Heisenberg antiferromagnet. First we consider in Sec. III A the lower edge of the spectrum. This is a quantity that can be obtained by several other methods, including various Monte Carlo algorithms, such that the relative accuracy of each one and the region in parameter space, where each method gives best results, can be assessed. In our case, this quantity is obtained from the asymptotic behavior of the one-particle Green’s function in imaginary time. However, not only the energy but also the weight of such an excitation can be extracted from the asymptotics, leading to the quasiparticle weight, as discussed in Sec. III B. The present algorithm is up to now the only one capable of extracting this information for the $`t`$-$`J`$ model free of approximations on large lattices (in general up to $`16\times 16`$ and for $`J/t=2`$ up to $`24\times 24`$ sites). For small lattice sizes, the results can be compared with exact diagonalizations, whereas for large systems only comparisons with approximate methods like SCBA can be made. Finally, the whole spectrum is considered in Sec. III C, where the spectral function $`A(\stackrel{}{k},\omega )`$ is discussed. Using the information from the lower edge of the spectrum, a resonance above the quasiparticle band is identified, that is very well described as a string excitation.
### A The lower edge of the spectrum
The accuracy and stability of the data allow in our case to obtain the lower edge of the spectrum directly from the slope of the one-particle Green’s function as a function of imaginary time $`\tau `$, for large values of $`\tau `$. Figure 1 shows the asymptotics in imaginary time for two values of the coupling constant, showing that the most accurate results are obtained, when $`J/t=2`$. $`J/t=0.4`$ is the smallest coupling, where such a procedure can be applied. In order to check the results obtained at the smallest coupling, we made additional calculations at $`\mathrm{\Delta }\tau t=0.2`$ (all other calculations are done at $`\mathrm{\Delta }\tau t=0.05`$), where larger values of $`\tau t`$ can be reached. The resulting Green’s functions are the same within the error bars, indicating a small $`\mathrm{\Delta }\tau `$ effect.
Figure 2 shows the lower edge of the spectrum for $`J/t=0.4`$ and $`J/t=2`$ in a $`16\times 16`$ sites system. The energies are displayed with respect to the ground-state energy of the Heisenberg antiferromagnet. The results are compared with variational Monte Carlo (VMC) , Green’s function Monte Carlo (GFMC) , and series expansions , whenever data is available. At $`J/t=0.4`$ (Fig. 2a), where our results are most affected by fluctuations, we observe good agreement with GFMC. The behavior of the statistical error is similar in both methods, with larger fluctuations around $`\stackrel{}{k}=(0,0)`$ and $`(\pi ,\pi )`$. Around $`\stackrel{}{k}=(\pi ,0)`$ our results show somewhat larger fluctuations. For $`J/t=0.4`$ VMC , also appears to be very accurate concerning the lower edge. When its energies are compared to our calculations and the GFMC technique, we find that their energies are within the error bars of the exact QMC calculations. At $`\stackrel{}{k}=(0,0)`$, the variational result is at the lower edge of the error bars of our calculation, and have the smallest statistical error of all three approaches. At this specific $`k`$-point both GFMC and our approach have large fluctuations before the state with lowest energy is clearly reached. As mentioned above, additional calculations with $`\mathrm{\Delta }\tau t=0.2`$ were performed, in order to check the results obtained, without observing significant changes.
Figure 2b shows that at $`J/t=2`$, where our algorithm leads to much more accurate results, the variational results are too high in energy, but still close to our numerically exact ones. For values of $`J/t1`$, additional results from series expansions are available. At $`J/t=2`$ we observe in general a very good agreement. Only around $`(\pi ,0)`$ we see that series expansions slightly underestimate the energy of the hole. The general features of the lower edge are not substantially modified when going from $`J/t=0.4`$ to $`J/t=4`$. This is shown in Fig. 3, where the only changes observed are an overall shift in energy with respect to the Heisenberg antiferromagnet and a change in the bandwidth. The shift in energy can be followed by considering the dependence of $`ϵ(\pi /2,\pi /2)`$ on $`J/t`$. This dependence is rather accurately described by $`ϵ(J/t)/t=3.28+a_1(J/t)^{2/3}`$, where $`a_1`$ is the first eigenvalue of the dimensionless Airy equation (see Fig. 13 in Sec. III C). Such a scaling of the hole energies is found in the $`t`$-$`J_z`$ model in the continuum limit for small values of $`J_z`$ , when loops along the path of the hole are disregarded. In that case, the constant is $`2\sqrt{z1}`$, where $`z`$ is the coordination number. The resulting string picture gives an accurate description of the lowest excitations close to $`\stackrel{}{k}=(\pi /2,\pi /2)`$. As will be shown in Sec. III C, also the next higher excitation can be described by the string picture.
Figure 4 shows the bandwidth obtained in our simulations compared with exact diagonalizations , GFMC , SCBA , VMC and series expansions . For $`J/t<0.8`$ good agreement is found among all methods, whereas for larger values of $`J`$, only series expansions and VMC agree with our data. This, and the fact that the string picture gives a good representation of the lowest lying states, suggest that a perturbation expansion as performed in series expansions can be used to interpret the distinctive features of the lower edge. In particular the flat band observed around $`\stackrel{}{k}=(\pi ,0)`$ and the fact that the degeneracy between this point and $`\stackrel{}{k}=(\pi /2,\pi /2)`$ suggested by some approaches is clearly lifted, as shown by our simulation, are very well reproduced by series expansions. The flat bands can be well observed for all considered values of $`J/t`$, when considering the lower edge (Fig. 2 and 3) and the complete spectral function (Fig. 11). Our data clearly show for $`J/t0.6`$, that the neighboring points of $`\stackrel{}{k}=(\pi ,0)`$ are generally slightly higher in energy. The band in this area does not seem to be completely flat, but it changes its curvature with local minima of the dispersion at the points $`(\pi ,\delta )`$ and $`(\pi \delta ,0)`$, when going along the (1,0) and (0,1) directions respectively, with the caveat that they are well defined beyond the error bars only for $`J/t>1`$. In all the cases we find $`\delta 0.3\pi `$. This region with a very flat band spans an extremely large area in the Brillouin zone.
A flat band on a similarly wide region in the Brillouin zone around $`\stackrel{}{k}=(\pi ,0)`$ is also observed in photoemission spectroscopy of cuprates close to the Fermi-energy in the optimally doped compounds. As doping is reduced, that portion of the spectrum opens a pseudogap and weight is transferred to higher energies , until in the undoped materials, this portion is about $`2J`$ ($`300`$meV) above the minimum at $`\stackrel{}{k}=(\pi /2,\pi /2)`$ . The energy difference between the points $`\stackrel{}{k}=(\pi /2,\pi /2)`$ and $`\stackrel{}{k}=(\pi ,0)`$ is in our simulation about $`\mathrm{\Delta }=(0.25\pm 0.10)t`$ ($`J/2`$ for $`J=0.4t`$). The rather large error corresponds mainly to $`J/t<1`$. No significant dependence on $`J/t`$ can be observed in the whole range under consideration, in contrast to the results from SCBA and series expansions. However, it could be that the $`J`$-dependence is masked in our case by large fluctuations, taking into account that the variations observed for this quantity by SCBA and series expansions are much smaller than the one observed for the bandwidth. SCBA gives values ranging from $`0.17t`$ ($`J/t=1`$) to $`0.12t`$ ($`J/t=4`$), that are smaller than the values we obtain. On the other hand, series expansions obtain values between $`0.15t`$ at $`J/t=1`$ and $`0.25t`$ at $`J/t=2.5`$. The values obtained by series expansions are consistent with our results for large values of $`J/t`$.
### B The quasiparticle weight
The quasiparticle weight is the weight of the exponential with the slowest decay, that is the exponential that determines the lower edge of the spectrum. This weight is
$`Z(\stackrel{}{k})=\underset{\tau \mathrm{}}{lim}G(\tau ,\stackrel{}{k})\mathrm{exp}\left[\left(ϵ_\stackrel{}{k}ϵ_0\right)\tau \right]`$ (13)
In the following we focus on the thermodynamic limit of $`Z(\stackrel{}{k})`$ for the wave vectors $`\stackrel{}{k}=(\pi ,0)`$ and $`\stackrel{}{k}=(\pi /2,\pi /2)`$. Figures 5 and 6 show the finite-size scaling on these two points for $`J/t=2`$ and $`J/t=0.6`$ respectively. For both $`k`$\- and $`J`$-values, an appreciable quasiparticle weight is obtained, demonstrating that the lower edge of the spectrum describes the band of a coherent quasiparticle. The determination of the quasiparticle weight is only accurate for $`J/t0.6`$. Below that value, the quality of the data is less satisfactory and, for $`J/t=0.4`$ the value presented can be taken only as an upper bound. The size dependence of $`Z(\pi /2,\pi /2)`$ and $`Z(\pi ,0)`$ is not very large and scales linearly with the inverse linear size of the system for $`J/t0.6`$, in agreement with SCBA . The size dependence at $`(\pi /2,\pi /2)`$ is systematically larger than at $`(\pi ,0)`$. The sizes considered are $`L\times L`$, with $`L=16,12,8`$, and 4. At $`J/t=2`$ we use additionally a $`24\times 24`$ lattice. Values from exact diagonalization were included when available.
Figure 7 shows that the extrapolated quasiparticle weight increases with $`J/t`$ both for $`\stackrel{}{k}=(\pi ,0)`$ and $`\stackrel{}{k}=(\pi /2,\pi /2)`$. At $`J/t=4`$ the quasiparticle reaches about $`80\%`$ of its maximal value. The changes of the quasiparticle weight with $`J/t`$ are small when $`J/t1`$ and the slope becomes steeper for smaller values. Estimates of the quasiparticle weight were given both by VMC and SCBA , the difference being rather small. The general trend is that VMC overestimates it at small $`J`$ whereas SCBA overestimates it at large $`J`$. For definiteness we compare our results with SCBA for a $`16\times 16`$ system in Fig. 8. We find a rather good agreement between both methods. As in our case $`Z(\pi ,0)>Z(\pi /2,\pi /2)`$ for all considered values of $`J/t`$. At small values of $`J`$ ($`0.01J/t0.5`$) SCBA finds a scaling of $`Z(\pi /2,\pi /2)=0.31J^{2/3}`$ and $`Z(\pi ,0)=0.35J^{0.7}`$. For $`J/t1`$, the results from SCBA overestimates the quasiparticle weight at the two considered $`k`$-points, with an increasing deviation for larger values of $`J/t`$. Based on the quantitative agreement of SCBA with our results for small $`J`$, we can confidently conclude that the quasiparticle at $`\stackrel{}{k}=(0,\pi )`$ and $`(\pi /2,\pi /2)`$ should be finite for all values of $`J`$ in the physically relevant region (i.e. $`J/t>0.1`$).
As mentioned in the introduction, there are exact results for the quasiparticle weight at the supersymmetric point in two dimensions . On the one hand, $`Z(\stackrel{}{k}=0)=1/2`$, a requirement that is fulfilled by our simulation, where the Green’s function at that particular $`k`$-point consists of a single exponential. In contrast to this, the estimate of SCBA is approximately $`0.45`$ and that of VMC $`0.32`$. Furthermore, Sorella showed that $`Z(\stackrel{}{Q})/Z(0)(2m)^2`$, where $`m^2=S(\stackrel{}{Q})/N`$, $`S(\stackrel{}{Q})`$ being the magnetic structure factor at the antiferromagnetic wave vector. The equality is reached in the thermodynamic limit. Figure 9 shows the evolution with system size of $`Z(\stackrel{}{Q})`$ together with results from exact diagonalization for a $`4\times 4`$ system and $`(2m)^20.37`$ for $`L\mathrm{}`$. Although large error bars show that the determination of $`Z(\stackrel{}{k})`$ is less satisfactory for $`\stackrel{}{k}=\stackrel{}{Q}`$ than at $`\stackrel{}{k}=(\pi /2,\pi /2)`$, the data are consistent with the exact result. It was further suggested that if $`Z(\stackrel{}{k}+\stackrel{}{Q})/Z(\stackrel{}{k})=(2m)^2`$ is satisfied for $`\stackrel{}{k}0`$, a jump in the quasiparticle weight should be observed on crossing the border of the magnetic zone. Figure 10 shows $`Z(\stackrel{}{k})`$ along the symmetry directions in the Brillouin zone for a $`24\times 24`$ system and $`J/t=2`$. Our data do not show any sizable jump. Unfortunately, it is not possible to consider arbitrarily long imaginary times since as Eq. (13) shows, the errors are amplified exponentially. Therefore, our results cannot be considered as a proof of continuity. However, in view of the good agreement with the above mentioned exact results, we consider them as a convincing evidence.
### C Spectral function and string excitations
The results discussed in Sec. III A for the lower edge of the spectrum and in Sec. III B for the quasiparticle weight can be recognized in the spectral function (Fig. 11) obtained by using MaxEnt. For clarity, the maximum of each curve is normalized to $`1`$ in the plots. The small numbers on the right hand side of the figures correspond to the maximal value of $`A(\stackrel{}{k},\omega )`$ when the integral $`\underset{\mathrm{}}{\overset{\mathrm{}}{}}𝑑\omega A(\stackrel{}{k},\omega )`$ is properly normalized to $`\pi /2`$. The lower edge of the spectrum remains like in the previous section, but the accuracy of its location in $`A(\stackrel{}{k},\omega )`$ is reduced by MaxEnt. The peaks around $`(0,\pi )`$ and $`(\pi /2,\pi /2)`$ are generally very sharp, in agreement with the fact that a finite quasiparticle weight was found in Sec. III B. A transfer of weight from high to low energies can be observed, when $`J/t`$ is increased, consistent with the increase in the quasiparticle weight (Fig. 7 in Sec. III B).
When compared to the 1D case , it is seen that the high energy excitations in the 2D case are extremely broad. The total bandwidth remains essentially constant as a function of $`J`$ in contrast to the 1D case, where it scales as $`4t+J`$.
For values of the coupling in the range $`J/t2`$ we observe satellite peaks in the region around $`\stackrel{}{k}=(\pi /2,\pi /2)`$ (Fig. 11) next to the lowest energy peak which is extremely sharp and corresponds to a quasiparticle. The $`\delta `$-peak can not be handled satisfactorily by MaxEnt. As can be seen by comparison of Figs. 2 and 11, MaxEnt gives some weight at energies lower than the band edge. This additional weight has to be balanced in some way, such that this error propagates to the other side of the $`\delta `$-peak. Small peaks in the vicinity of the $`\delta `$-peak can therefore not be resolved. In order to resolve structures close to the quasiparticle peak, we subtract the exponential corresponding to the lowest energy (see Fig. 12). The thus modified Green’s function can now be used as input of MaxEnt.
Before proceeding to the results, let us remark that, when the MaxEnt results obtained with the modified Green’s function, i.e. after the subtraction of the lowest exponential, are viewed closely, on occasions, an additional peak appears at the bottom of the spectrum (this effect can be seen e.g. for $`(\pi /2,\pi /2)`$ in Fig. 13). To exclude, that this peak corresponds to a real physical effect, we take several modified Green’s functions, that are consistent with the exponential of the lowest peak, within the statistical error. Therefore, we take the lowest and the highest exponential, that are consistent with the results obtained in Sec. III A, and use them as input of MaxEnt. As can be seen in Fig. 14, the new peak that appears below the low energy peak of the original function, and hence is artificial, is only observed in two cases with varying position, whereas the two other peaks can be always observed, no matter which exponential is subtracted (always within the statistical errors). The position of these high energy peaks is not changed by the different subtractions, only the width is affected. In all cases discussed, a small shift of these structures can be observed with respect to the ones in the spectrum without the subtraction. However, the positions assumed by these structures after the subtraction is not affected by the different subtractions within the values allowed by the statistical errors. We conclude that the initial small shift is due to the inability of MaxEnt to concentrate the weight of the delta-function of the quasiparticle peak to a single energy value.
The result of the procedure described above is shown in Fig. 15. For $`(\pi /2,\pi /2)`$ there are only little changes of the position of the maxima of the existing peaks at small $`J/t`$ compared to the full spectral function (except the low energy peak, that disappeared). We can further observe, that the satellite peak next to the low energy peak can now be seen for all values of $`J/t2`$. One should notice, that no additional weight has been produced at high energies, but the normalization has changed (again the maximal value is normalized to 1, not the area of the spectral function).
At $`(\pi /2,\pi /2)`$ (Fig. 15(a)) the resolution of the second-lowest excitation is quite clear, when applying the above method, whereas at $`(\pi ,0)`$ (Fig. 15(b)) the results are either not accurate enough, or the corresponding excitation is weaker. For $`J/t=1.2`$ the resolution is not good enough to separate the two resonances at $`(\pi ,0)`$. Generally the excitations at higher energies at $`(\pi ,0)`$ are broader than at $`(\pi /2,\pi /2)`$, so that the positions of the maxima are not as well defined. Similar structures were observed in exact diagonalization and in SCBA , and were ascribed to string excitations.
When the string picture is valid, as it is expected in the $`t`$-$`J_z`$ model the hole is confined by a linear potential, leading to (k-independent) eigenvalues of the energy given by
$$E_n/t=2\sqrt{3}+a_n(J_z/t)^{2/3},$$
(14)
where $`a_n`$ are the eigenvalues of a dimensionless Airy equation . The first three eigenvalues are given by $`a_n=2.33,4.08,5.52`$. In Fig. 16 the results for the first three excitations are given for $`\stackrel{}{k}=(\pi /2,\pi /2)`$, and are compared to the predictions from SCBA. The error bars on the second and third peak are obtained as the width of the MaxEnt peak at half intensity, the error bars of the first peak are taken as in Sec. III A. We find, that for $`J/t2`$ the lowest peak can be accurately described by $`ϵ_0(\pi /2,\pi /2)=E_H3.28t+2.33(J/t)^{2/3}t`$, where $`E_H`$ is the Heisenberg energy per site, and the second peak by $`ϵ_1(\pi /2,\pi /2)=E_H3.28t+4.08(J/t)^{2/3}t`$. The value of $`3.28t+E_H`$ is the result obtained from SCBA , whereas the prefactors of $`(J/t)^{2/3}`$ are exactly the values of the dimensionless Airy function, implying that the first two peaks behave (within our error bars) exactly as it is expected by the string picture. In contrast to this, a fit from SCBA for the first three excitations in the $`t`$-$`J`$ model for values of $`J/t0.4`$ results in $`a_n=2.16,5.46,7.81`$, also with the exponent $`2/3`$ , leading to a clear disagreement with our data. The third peak that can be resolved cannot be explained by the string picture, since its distance to the lower band edge is independent of $`J`$ and has a value of about $`4t`$. The existence of a string excitation is not restricted to $`(\pi /2,\pi /2)`$, but it can also be observed between $`(\pi /2,\pi /2)`$ and $`(\pi /2,3\pi /4)`$. This is demonstrated for the value $`J/t=0.6`$ (see Fig. 13).
The results above lead to the conclusion that the lowest excitations can be well described by the string picture. However, it should be kept in mind that the string picture originates in the Ising limit for $`J/t1`$, and that it is based on the continuum limit, that seems far away from our case with strings of lengths between two and a maximum of five lattice points, that correspond to the first two string excitations. Moreover, the string picture predicts a band without dispersion, that is clearly not the case in our simulations. A way to reconcile this paradoxical situation is given by the very good quantitative agreement between QMC and series expansions for the dispersion of the quasiparticle and its bandwidth for a fairly large range in $`J`$. As shown by the expansion around the Ising limit, a coherent motion of the hole is made possible after the creation of strings due to hopping processes, by appropriate spin-flips, the shortest string being of length two. The lowest order contribution appears in third order, where the points $`(\pi /2,\pi /2)`$ and $`(\pi ,0)`$ are degenerate. Fourth and higher order processes remove this degeneracy, giving rise to a band that agrees qualitatively very well with the one obtained in QMC. Therefore on top of the coherent motion determined by $`J_{}`$, string like excitations are possible and related to $`J_z`$ and $`t`$. Such a possibility was already proposed by Béran, Poilblanc and Laughlin (BPL) on the basis of exact diagonalizations on small systems and is confirmed unambiguously by our simulations on large systems.
At $`J/t>2`$ the excitations fall below the values predicted by the string picture. In those regions the string picture is no longer valid, as the relaxation of the disturbed spin bonds is faster than the motion of the hole.
## IV Conclusions
A new QMC algorithm was presented that allows a rather accurate determination of the single hole dynamics in a two dimensional Heisenberg $`S1/2`$ antiferromagnet. The main advantages of this algorithm are the combination of the loop algorithm for the update of the spins and the exact evolution of the hole for a given spin configuration. Due to the diverging correlation length at zero temperature, large autocorrelation times should be expected for algorithms with local updates, a problem that is avoided here by the global update of the spins. On the other hand, the exact evolution of the hole for a given spin background avoids further statistical errors that would be introduced if the hole is updated stochastically, as in recently proposed approaches . In fact, the accuracy achieved allows for a determination of several dynamical quantities on large lattice sizes, leading to the possibility of a finite size scaling of e.g. the quasiparticle weight.
First (Sec. III A) we discussed the lower edge of the spectrum that is obtained directly from the asymptotics in imaginary time of the Green’s function. This quantity is accessible to different techniques, that are however, with the exception of GFMC, either restricted to small lattice sizes or approximate. The comparison shows that very accurate results are given by series expansions over a large range of parameters, supporting thus the interpretation of the relevant physical processes for the coherent motion of the hole in the frame of a pertubative expansion around the Ising limit. This picture is further enforced by our study of the quasiparticle weight (Sec. III B) and the spectral function (Sec. III C). In Sec. III B it was shown, that indeed the lower edge of the spectrum describes the coherent propagation of a hole with finite quasiparticle weight. This is the case for all the parameter range studied, and due to the good agreement with SCBA especially for small values of $`J`$, one can conclude that this coherent propagation takes place for essentially $`J>0`$. Furthermore, by considering structures next to the lowest peak in the spectral function (Sec. III C), it is seen that the lowest excitations around the wave vector $`\stackrel{}{k}=(\pi /2,\pi /2)`$ are very well described by the levels of strings usually discussed for the $`t`$-$`J_z`$ model, giving further support to the pertubative picture, where the hole creates strings during its motion through the lattice, that are healed by exchange processes, leading thus to coherence. In fact, the strings for the first two levels, that agree quantitatively with our simulations correspond to a length of two and five lattice sites. Strings of length two are the dominant contributions in series expansions for the dispersion of the quasiparticle. Moreover, our findings showing the existence of string resonances above the quasiparticle pole lend support to a picture developed by BPL , where the composite nature of the quasiparticle is advanced. In previous exact diagonalization studies, the existence of such resonances, that were first observed in $`4\times 4`$ lattices were not clearly identified on larger lattices . We have shown in Sec. III C that they can be quantitatively identified with string excitations. However, they are visible only in a rather narrow region along the line $`k_x\pi /2`$, $`\pi /2<k_y<3\pi /4`$, such that in small lattices with up to 26 sites, these features can be very much affected by boundary effects. Following BPL, the quasiparticle can be viewed as a light holon attached to a spinon by a confining potential, the one that gives rise to the spectrum of string excitations.
A comparison with experiments fails due to the small gap between the lowest peak at $`\stackrel{}{k}=(\pi /2,\pi /2)`$ and the flat band around $`\stackrel{}{k}=(\pi ,0)`$. It was suggested by several authors that this shift might be obtained introducing hopping terms to second and third nearest neighbors. Furthermore, it was found in exact diagonalizations that such extra terms lead to a noticeable reduction of the quasiparticle weight. Since exact diagonalizations with second and third nearest neighbors in lattices with 18 and 26 sites suffer considerably under finite size effects, a discussion of the influence of longer range hopping on the quasiparticle weight must be carried out in much larger lattices. Such studies are presently under way.
This work was supported by Sonderforschungsbereich 382. The numerical calculations were performed at HLRS Stuttgart. We thank the above institutions for their support. We are grateful to P. Horsch, E. Manousakis, D. Poilblanc, P. Prevlošek, and S. Sorella for helpful and instructive discussions, and to CECAM, where part of these discussions took place, for its hospitality.
|
warning/0002/math0002225.html
|
ar5iv
|
text
|
# Null-geodesics in complex conformal manifolds and the LeBrun correspondence
## 1. Introduction
On a complex manifold, the existence of a complex-Riemannian metric implies, in general, strong topological assumptions, especially if the manifold is compact (e.g. the — square of the — canonical bundle has to be trivial). However, any analytic (pseudo-) Riemannian (or conformal) manifold can be complexified, and a natural question is to see to what extent the global properties of the real manifold (e.g. existence of closed (null-) geodesics) hold for the complexified spaces. This complexification procedure naturally occurs in twistor theory (see below), which has been intensively studied for Riemannian space-times (see, e.g., , , , ); the complex-Riemannian setting, in which historically the twistor theory was first introduced , can provide a link to the Lorentzian geometry.
In complex conformal geometry (which implies weaker assumptions on the topology of the manifold), the conformal structure is determined by the set of null-geodesics, which can be organized as a complex manifold under some topological conditions , . A natural question is which complex conformal manifolds admit compact null-geodesics; for example, if a self-dual manifold admits a globally-defined twistor space, then application of a twistorial interpretation of the Weyl tensor , implies that it is conformally flat, and the compact null-geodesic is simply-connected.
Our main result (section 4, Theorem 4) states that, if a conformal complex $`n`$-manifold admits a rational curve as a null-geodesic, then it is conformally flat (see also for the case of a complex projective manifold). The proof uses the properties of Jacobi fields along the considered compact, simply-connected, null-geodesic : namely, we compute the normal bundle of a compact, simply-connected, null-geodesic, and we show that the small deformations of the latter as a compact curve, or as a null-geodesic, coincide (section 4, Proposition 5). In addition to that, we use, for the (more difficult) case of dimension 3, a criterion for conformal flatness from , and we apply it to a locally defined, by the LeBrun correspondence (see below), self-dual ambient.
The other topic of this paper uses implicitly another application of twistor theory: It has been shown by LeBrun , , that any conformal 3-manifold can be locally realized as the conformal infinity of a self-dual Einstein (with non-zero scalar curvature) 4-manifold. We have, thus, a local correspondence assigning to a conformal structure in dimension 3 a self-dual Einstein metric in dimension 4, which we call the LeBrun correspondence.
As conformal structures of both manifolds are encoded in the complex, resp. $`CR`$, structure of their twistor spaces, they are implicitly related, for example if the 3-manifold $`M`$ is conformally flat, its ambient $`N`$ equally is. It is, however, difficult to obtain an explicit relation between the conformal invariants of $`M`$ and those of $`N`$ by twistorial methods, as there is no simple expression of the Cotton-York tensor of $`M^3`$ in twistorial terms, and the twistorial interpretation of the Weyl tensor of $`N^4`$ is highly non-linear .
In this paper we find a relation between these two conformal invariants of the manifolds involved in the LeBrun correspondence, or, more generally, of an umbilic submanifold $`M^3`$ and of its self-dual ambient $`N^4`$. It appears that the Weyl tensor of $`N^4`$ identically vanishes along $`M^3`$, and thus the Cotton-York tensor of $`N^4`$, restricted to $`M^3`$, is conformally invariant and can be identified with the Cotton-York tensor of $`M^3`$; in this case, it is also equal to the normal derivative of the Weyl tensor of $`N^4`$ (section 3, Theorem 1). This gives conditions for an open self-dual 4-manifold to admit a conformal infinity.
The paper is organized as follows : in section 2 we recall a few basic facts about complex- Riemannian and -conformal geometry, in section 3 we relate the conformal invariants of a 3-dimensional conformal infinity to those of its self-dual ambient (arising from the LeBrun correspondence), and in section 4 we state our results about conformal complex manifolds containing compact null-geodesics.
Throughout the paper we use the following conventions: in complex-Riemannian (or -conformal) geometry we use the same terminology as in the real framework (metric, Levi-Civita connection, curvature), and the holomorphic bundles are denoted like the corresponding bundles in real geometry (for example, the holomorphic tangent bundle of M is denoted simply by $`T𝐌`$, rather than the more precise $`T^{1,0}𝐌`$) ; manifolds with holomorphic conformal structures are denoted by bold-face letters (except in section 3, where the results hold also in the real framework).
## 2. Holomorphic conformal geometry
###### Definition 1.
Let $`𝐌`$ be a complex manifold, let $`n`$ be its complex dimension. A complex-Riemannian metric $`g`$ on it is a holomorphic section of $`S^2T^{}𝐌`$ which is non-degenerate at any point. A holomorphic conformal structure on $`𝐌`$ is a holomorphic line subbundle $`C`$ in $`S^2T^{}𝐌`$ such that any non-vanishing local section of $`C`$ is a local complex-Riemannian metric.
During the rest of this section, and of the whole fourth section of this paper, we shall simply denote these structures as metric and conformal structure (therefore omitting any reference to the complex framework).
A metric on $`T𝐌`$ induces metrics (i.e. non-degenerate symmetric bilinear forms) on all tensor bundles, in particular the square of the canonical bundle $`\kappa :=\mathrm{\Lambda }^nT^{}𝐌`$ is trivialized. If we can choose an orientation (defined as follows), then the canonical bundle itself can be trivialized by a volume form of norm 1. There are exactly 2 such forms in each fiber of $`\mathrm{\Lambda }^nT^{}𝐌`$, and an orientation is the choice of one of them, depending continously (thus holomorphically) on the base point.
Remark.
The notion of orientation is generally related to the reduction (when possible) of the structure group of the frame bundle from $`G`$ to the connected component of the identity $`G_0`$; in absence of any structure, the group $`G`$ is simply the connected $`GL(n,)`$, so the notion of orientation has no meaning in “raw” complex geometry. But a Riemannian metric on $`𝐌`$ is equivalent to the reduction of its frame bundle to a $`O(n,)`$-bundle, where $`O(n,):=\{AGL(n,)|A^tA=\mathrm{𝟏}\}`$; a further choice of an orientation reduces the structure group of the frame bundle to the connected component of this group, containing the identity: $`SO(n,):=O(n,)SL(n,)`$.
Unlike in the real framework, these reductions, always possible on (small) contractible open sets, are submitted to some topological constraints if we want to define them globally on $`𝐌`$.
Weaker constraints are implied by the existence of a conformal structure : the square of the canonical bundle needs to admit a $`n`$th order root $`L^2C`$, where $`L:=\kappa ^{1/n}`$ is the<sup>1</sup><sup>1</sup>1the dual of $`L`$ is a square root of $`C`$; the choice of such a square root is implied, if $`n=2m+1`$, by the conformal structure $`C`$ as $`\kappa =C^mL^1`$; if $`n`$ is even, neither $`C`$ nor an orientation — see below — imply the choice of $`L`$. (respectively, one of the) weight 1–density bundle of the manifold $`𝐌`$. We can study the conformal structure $`C`$ using the formalism of density bundles and of Weyl derivatives . From now on, we shall not make use of the weight 1–density bundle, but only of $`C=L^2`$, which is enough to define the conformal structure.
Remark.
A conformal structure is equivalent to the reduction of the structure group of the frame bundle to $`CO(n,):=O(n,)\times ^{}/\{\pm \mathrm{𝟏}\}`$ (the quotient is due to the fact that $`\mathrm{𝟏}O(n,)`$). This group is disconnected if $`n`$ is even, and the connected component of 1 is $`CO_0(n,)=SO(n,)\times ^{}/\{\pm \mathrm{𝟏}\}`$, but it is connected if $`n`$ is odd (and in this case the right hand side of the previous equality coincides with $`CO(n,)`$). Therefore, although a complex-Riemannian 3-manifold admits, locally, 2 possible orientations, they are conformally equivalent, fact that makes impossible a canonical way to associate an orientation to a metric in the conformal class.
Remark.
In complex-, as in real-Riemannian geometry, the orientation determines (and is determined by) a family of compatible oriented orthonormal basis in $`T𝐌`$; if $`dim𝐌`$ is even, by multiplying all the vectors of such a basis by a non-zero complex number we obtain an oriented orthonormal basis compatible with another metric in the conformal class (if $`dim𝐌`$ is odd, multiplication by $`1`$ yields a basis compatible with the same metric, but with the opposite orientation).
The notion of orientation, in four-dimensional conformal geometry, is important for the definition of anti–, resp. self-duality (see below). More generally, the Hodge $``$ operator (defined as in real Riemannian geometry) is conformally invariant, and gives an explicit expression for the splitting of the bundle of 2-forms and of the curvature tensor , see also next section.
Maybe the simplest way to view a conformal structure is as an equivalence class $`c`$ of local metrics, two such local representants $`g`$ and $`h`$ satisfying, on the open set where both are defined, to $`g=fh`$, with $`f`$ a non-vanishing holomorphic function. Unlike in the real framework, global representants may not exist in general. From now on we shall consider a conformal structure on $`𝐌`$ as being given by the conformal class $`c`$ rather than by the line bundle $`C`$.
Geometrically, a conformal structure is given by its isotropy cone $`𝒞T𝐌`$ of vectors of norm 0. Because of the non-degeneracy of any local metric in $`c`$, the projective isotropy cone $`(𝒞)`$ is a non-degenerate hyperquadric in $`(T𝐌)`$. In dimension 3, $`(𝒞)`$ is a conic (curve) in $`^2`$, and in dimension 4 it is a ruled surface in $`^3`$ ; therefore, in this latter case, there are 2 families — each of which can be characterized with respect to a given orientation — of isotropic planes in $`T𝐌`$ called $`\alpha `$-, resp. $`\beta `$-planes.
For any local metric we have a Levi-Civita connection, whose curvature has the same components as in the real Riemannian geometry (see next section). In particular, the Weyl tensor is independent of the metric in the conformal class.
The geodesics for which the tangent direction at a point (and thus, at any point) is isotropic are called null-geodesics, and they are locally independent (up to a reparametrization) of the metric in the conformal class. The same is true for higher-dimensional totally geodesic and isotropic submanifolds — if they exist —, called null-submanifolds. In dimension 4 they are $`\alpha `$-, resp. $`\beta `$-surfaces (tangent to $`\alpha `$-, resp. $`\beta `$-planes, see above), and they exist if and only if the (oriented) conformal structure is anti-, resp. self-dual, i.e. the component $`W^+`$, resp. $`W^{}`$, of the Weyl tensor $`W`$ of $`(𝐌,c)`$ vanishes identically ,,.
If the manifold is self-dual, one considers locally the twistor space $`Z`$ of $`(𝐌^4,c)`$ as the set of $`\beta `$-surfaces<sup>2</sup><sup>2</sup>2see , , and section 4 for an explanation of the difficulties of a global definition of the twistor space.. It is a 3-dimensional complex manifold , containing rational curves whose normal bundle is isomorphic to $`𝒪(1)𝒪(1)`$ (called twistor lines) (where $`𝒪(1)`$ is the dual of the tautological bundle $`𝒪(1)`$ of $`^1`$). If, in addition, we can choose an Einstein metric $`g`$ in the conformal class $`c`$, we get an extra structure on $`Z`$, namely a distribution of 2-planes, which is totally integrable (and yields a foliation) if the scalar curvature of $`g`$ vanishes, otherwise it is a contact structure ,,. Conversely, from a manifold $`Z`$ containing twistor lines as above (called a twistor space), plus — possibly — the additional 2-planes distribution, one can recover — at least locally —, via the reverse Penrose construction , the self-dual manifold $`(𝐌^4,c)`$.
In all generality, one can always consider locally (on a geodesically convex open set, for example, see section 4) the space of null-geodesics of a conformal manifold $`(𝐌^n,c)`$, and the key point in the LeBrun correspondence (defined below) is that the space of null-geodesics of a 3-dimensional conformal manifold $`(𝐌^3,c)`$ (also called the twistor space of M) is a twistor space endowed with a contact structure, therefore we get (locally again) a self-dual manifold $`(𝐍^4,c)`$, in which $`(𝐌^3,c)`$ is umbilic, and it is the conformal infinity of an Einstein metric $`g`$ on $`𝐍`$ with non-zero scalar curvature , .
###### Definition 2.
Let $`(𝐌,c)`$ be a conformal 3-manifold, that we shall suppose civilized (e.g. geodesically connected for some metric in the conformal class). The LeBrun correspondence associates to $`𝐌`$ the (germ-unique) self-dual Einstein 4-manifold $`𝐍`$ such that the twistor spaces of $`𝐌`$ and $`𝐍`$ coincide.
###### Proposition 1.
, In the LeBrun correspondence, $`(𝐌^3,c)`$ is an umbilic hypersurface of $`(𝐍^4,c)`$ (and has the induced conformal structure) and the Einstein metric of $`𝐍^4`$ has a second order pole at $`𝐌^3`$ (conformal infinity). Conversely, in such a geometric setting, the twistor spaces of the manifolds $`𝐌`$ and $`𝐍`$ coincide.
Remark.
There is no a priori definition of a conformal infinity of an open (real- or complex-) Riemannian manifold $`X`$, even if the metric is complete (in the real framework). Here we consider uniquely the case when this infinity is an (umbilic) submanifold (or boundary) of a conformal extension of $`X`$, $`\overline{X}X`$; the conformal structure extends smoothly beyond the infinity. In other cases, in which the notion of conformal infinity can still be defined, the conformal structure is singular at infinity, which, in these cases, is no longer conformal, but admits instead a $`CR`$ , or a quaternionic contact structure .
## 3. Conformal infinity of a self-dual manifold
The object of this section is to find a relationship between the conformal invariants of a conformal infinity and of its self-dual ambient arising from the LeBrun correspondence. The results are local, and they hold in the complex as well as in the real Riemannian or in the signature (2,2) pseudo-Riemannian framework. We begin by recalling a few facts about the conformally invariant tensors in Riemannian geometry.
For a $`n`$-dimensional Riemannian manifold $`(M^n,g)`$, the curvature has the following expression:
(1) $`R^M(X,Y)`$ $`=`$ $`(h\text{I})(X,Y)+W(X,Y),\text{ where}`$
(2) $`(h\text{I})(X,Y)`$ $`:=`$ $`h(X,)Yh(Y,)X,X,YTM,`$
is the suspension by the identity I of the normalized Ricci tensor
(3)
$$h=\frac{1}{2n(n1)}\text{Scal}^gg+\frac{1}{n2}\text{Ric}_0;$$
$`Scal`$ and $`Ric_0`$ are the scalar curvature, resp. the trace-free Ricci tensor, and, together with the Weyl tensor $`W`$, they are the irreducible components of the curvature under the orthogonal group if $`n5`$. If $`n=3`$, $`W`$ vanishes identically, and if $`n=4`$ it further decomposes in two irreducible components $`W^+`$, resp. $`W^{}`$, called the self-dual, resp. anti-self-dual (or positive, resp. negative) Weyl tensor.
The Weyl tensor, viewed as a section in $`\text{Hom}(\mathrm{\Lambda }^2TMTM,TM)`$ (as a (3,1)–tensor), is conformally invariant, and, if $`n4`$, it completely determines, locally, the conformal structure of $`(M,[g])`$ (for a self-dual manifold, $`W^{}0`$, thus the Weyl tensor actually coincides with $`W^+`$). In dimension $`n=3`$, this function is fulfilled by the Cotton-York tensor, which can be defined in all dimensions by
(4)
$$C(X,Y)(Z):=(_Xh)(Y,Z)(_Yh)(X,Z),X,Y,ZTM,$$
and it can be shown that, for another metric $`g^{}:=e^{2\phi }g`$ in the same conformal class, the corresponding Cotton-York tensor $`C^{}`$ is related to $`C`$ by the formula :
(5)
$$C^{}(X,Y)(Z)=C(X,Y)(Z)\text{d}\phi (W(X,Y)Z).$$
In particular $`C`$ is conformally invariant along the zero set of $`W`$, thus everywhere if $`dimM=3`$.
Remark.
The Cotton-York tensor $`C`$ of $`M`$ is a 2-form with values in $`T^{}M`$, and it satisfies a first Bianchi identity, as $`h`$ is a symmetric tensor, and also a contracted (second) Bianchi identity, coming from the second Bianchi identity in Riemannian geometry, :
(6) $`{\displaystyle C(X,Y)(Z)}`$ $`=`$ $`0\text{ circular sum};`$
(7) $`{\displaystyle C(X,e_i)(e_i)}`$ $`=`$ $`0\text{ trace over an orthonormal basis}.`$
This means that $`C`$ is an irreducible tensor if $`n=3`$ or $`n>4`$, and, if $`n=4`$, $`C`$ has two irreducible components, the self-dual, resp. the anti-self-dual Cotton-York tensor
$$C^+\mathrm{\Lambda }^+M\mathrm{\Lambda }^1M,\text{resp.}C^{}\mathrm{\Lambda }^{}M\mathrm{\Lambda }^1M.$$
They both satisfy (6) and (7) (note that these two relations are equivalent in their case).
The Cotton-York tensor is related to the Weyl tensor of $`M`$ by the formula :
(8)
$$\delta W=C,$$
where $`\delta :\mathrm{\Lambda }^2M\mathrm{\Lambda }^2M\mathrm{\Lambda }^2M\mathrm{\Lambda }^1M`$ is induced by the codifferential on the second factor, and by the Levi-Civita connection $``$. Then, again if $`n=4`$, $`C^+`$ has to be the component of $`\delta W`$ in $`\mathrm{\Lambda }^+M\mathrm{\Lambda }^1M`$, and we know that the restriction of $`W^{}`$ to $`\mathrm{\Lambda }^+M\mathrm{\Lambda }^2M`$ is identically zero. This means that
(9) $`\delta W^+`$ $`=`$ $`C^+,\text{and also}`$
(10) $`\delta W^{}`$ $`=`$ $`C^{}.`$
We have thus:
###### Lemma 1.
On a self-dual manifold, $`C^{}`$ vanishes identically.
We consider now the situation in the LeBrun correspondence : Let $`(M,c)`$ be a 3-dimensional conformal manifold, and we suppose, without any local loss of generality, that it is the conformal infinity of the self-dual manifold $`(N,c)`$ (no use will be made of the Einstein metric on it); $`MN`$ is, thus, an umbilic hypersurface, such that the restriction of the conformal structure $`c`$ of $`N`$ to $`M`$ is non-degenerate (equivalently, $`TM`$ is nowhere tangent to an isotropic cone) and coincides with the conformal structure, still denoted by $`c`$, on $`M`$.
If we introduce the Hodge operator $`^M:\mathrm{\Lambda }^2M\mathrm{\Lambda }^1M`$, then the curvature tensor $`R^M`$ is equivalent to the symmetric 2-tensor $`^MR^M^M`$. A straightforward application of the above formula yields
(11)
$$R^M:=^MR^M^M=h+(\mathrm{tr}h)I.$$
For the 4-dimensional manifold $`N`$, the components of the Riemannian curvature can also be expressed as eigenspaces of $``$-type operators. Namely, considering $`R:=R^N`$ as a symmetric endomorphism of $`\mathrm{\Lambda }^2N=\mathrm{\Lambda }^+N\mathrm{\Lambda }^{}N`$, $`W^+`$ is the trace-free component of $`R`$ in $`\mathrm{End}(\mathrm{\Lambda }^+N)`$, and $`W^{}`$ is the trace-free component of $`R`$ in $`\mathrm{End}(\mathrm{\Lambda }^{}N)`$ . (The scalar curvature is four times the trace of $`R|_{\mathrm{\Lambda }^+}`$ or of $`R|_\mathrm{\Lambda }^{}`$, and the trace-free Ricci tensor is identified to the component of $`R`$ sending $`\mathrm{\Lambda }^+`$ into $`\mathrm{\Lambda }^{}`$ .)
We can canonically identify $`\mathrm{\Lambda }^+N`$ and $`\mathrm{\Lambda }^{}N`$, restricted to $`MN`$, to $`\mathrm{\Lambda }^2M`$, by:
(12)
$$\begin{array}{ccc}\mathrm{\Lambda }^2M\alpha \hfill & & \hfill \alpha +^N\alpha \mathrm{\Lambda }^+N\\ \mathrm{\Lambda }^2M\alpha \hfill & & \hfill \alpha ^N\alpha \mathrm{\Lambda }^{}N.\end{array}$$
Our first result is:
###### Theorem 1.
Let $`M`$ be an umbilic hypersurface of a self-dual manifold $`N`$. Then:
(i) The Weyl tensor of $`N`$ vanishes along $`M`$ :
$$W^+|_M0;$$
(ii) The Cotton-York tensor of $`M`$ is related to the self-dual Weyl tensor of $`N`$ by the formula:
$$g(_\nu W^+(A),B)_x=C(A)(^MB)_x,xM$$
where $`A,B\mathrm{\Lambda }^2T_xM`$, $`\nu T_xM`$ is unitary for the metric $`g`$, and the Hodge operator $`^M`$ is induced by $`g`$ and the orientation on $`M`$ admitting $`\nu `$ as an exterior normal vector.
(iii) The restriction to $`M`$ of the (self-dual) Cotton-York tensor of $`N`$ is equal to the Cotton-York tensor of $`M`$ :
$$C^+(X,Y)(Z)=C^M(X,Y)(Z),X,Y,ZTM.$$
###### Proof.
The claimed identities are conformally invariant : for (i) it is obvious, and the conformal invariance of (iii) follows from (i) and (5); to see that for (ii), let $`X,Y,Z,\nu `$ be a $`g`$-orthonormal oriented basis of $`N`$, such that $`X,Y,Z`$ is a $`g`$-orthonormal basis on $`M`$ giving the orientation as above. Then $`^M(ZX)=Y`$, and, if we take $`A:=XY,B:=ZX`$, the identity (ii) becomes
(13)
$$_\nu W^+(X,Y)Z,X=C(X,Y)(Y),$$
where angle brackets denote the scalar product induced by $`g`$.
The tensors $`W^+,C`$, in the above form, are independent of the chosen metric $`g`$ , which depends on the normal vector $`\nu `$, supposed to be $`g`$-unitary. If $`\nu ^{}:=\lambda \nu `$, for $`\lambda ^{}`$, then the corresponding metric $`g^{}=\lambda ^2g`$, and also $`_{}^{M}{}_{}{}^{}=\lambda ^1^M`$, thus the identity (13) for $`\nu ^{},g^{}`$ is equivalent to the one for $`\nu ,g`$.
Remark.
As $`W^+`$ is the trace-free component of the Riemannian curvature contained in $`\text{End}(\mathrm{\Lambda }^+N)`$, and is symmetric, it is enough to evaluate it on pairs $`A,B\mathrm{\Lambda }^2M\mathrm{\Lambda }^+N`$ which are unitary and orthogonal for the metric $`g`$, therefore the check of the equation (13) will prove the Theorem.
As $`W^\pm `$ are $`^N`$-eigenvectors in $`\text{End}_0(\mathrm{\Lambda }^2N)`$ (the space of trace-free endomorphisms of $`\mathrm{\Lambda }^2N`$), they are determined by the following formulas, where $`X,Y,Z`$ is any oriented orthonormal basis of $`TM`$:
$`W^+(X,Y)Z,X`$ $`=`$ $`{\displaystyle \frac{1}{4}}(R(X,Y)Z,X+R(Z,\nu )Y,\nu +`$
$`+R(X,Y)Y,\nu +R(Z,\nu )Z,X)`$
$`W^{}(X,Y)Z,X`$ $`=`$ $`{\displaystyle \frac{1}{4}}(R(X,Y)Z,X+R(Z,\nu )Y,\nu `$
$`R(X,Y)Y,\nu R(Z,\nu )Z,X),`$
where $`X,Y,Z,\nu `$ is supposed to be a local extension, around a region of $`M`$, of the $`g`$-orthonormal frame used in (13). As $`N`$ is self-dual, $`W^{}`$ is identically zero, thus, in the points $`xM`$, we have
(16)
$$W^+(X,Y)Z,X_x=\frac{1}{2}(R(X,Y)Y,\nu +R(Z,X)Z,\nu )_x.$$
It is a standard fact that, if $`M`$ is umbilic, there is a local metric $`g`$ in the conformal class $`c`$ of $`N`$, such that, for $`g`$, $`M`$ is totally geodesic. Without any loss of generality, because of the conformal invariance of the claimed identities (see above), we fix such a metric. Then we have
(17)
$$R(X^{},Y^{})Z^{}=R^M(X^{},Y^{})Z^{},X^{},Y^{},Z^{}TM,$$
which, together with (16), implies that $`W^+|_M0`$, and thus proves the point (i) in the Theorem.
On the other hand, (17), together with (16) and (3), yield
(18)
$$R(X,Y)Z,X_x+R(Z,\nu )Y,\nu _x=0,xM.$$
Let us compute now the normal derivative of $`W^+`$ in a point $`xM`$; we suppose that $`X,Y,Z,\nu `$ are locally extended by an orthonormal frame, and that they are parallel at $`x`$ (we omit, for simplicity of notation, the point $`x`$ in the following lines:
$$_\nu W^+(X,Y)Z,X=\frac{1}{2}(_\nu R(X,Y)Y,\nu +_\nu R(Z,X)Z,\nu ),$$
from (16). This is then equal to:
$$\begin{array}{ccc}\hfill _\nu W^+(X,Y)Z,X& =& \frac{1}{2}(_XR(Y,\nu )Y,\nu +_YR(\nu ,X)Y,\nu +\hfill \\ & & +_ZR(X,\nu )Z,\nu +_XR(\nu ,Z)Z,\nu ),\hfill \end{array}$$
from the second Bianchi identity. Then we have
$$\begin{array}{ccc}\hfill _\nu W^+(X,Y)Z,X& =& \frac{1}{2}(_XR(Z,X)Z,X+_YR(Z,Y)Z,X+\hfill \\ & & +_ZR(Y,Z)X,Y+_XR(Y,X)X,Y)\hfill \end{array}$$
from analogs of (18). Then
$$\begin{array}{ccc}\hfill _\nu W^+(X,Y)Z,X& =& \frac{1}{2}(_XR^M(Z,X)Z,X+_YR^M(Z,Y)Z,X+\hfill \\ & & +_ZR^M(Y,Z)X,Y+_XR^M(Y,X)X,Y)\hfill \end{array}$$
from (17)
$$\begin{array}{ccc}\hfill _\nu W^+(X,Y)Z,X& =& \frac{1}{2}(_Xh(Z,Z)+_Xh(X,X)+_Yh(Y,X)\hfill \\ & & _Zh(Z,X)_Xh(X,X)_Xh(Y,Y)),\hfill \end{array}$$
from (1). Finally, from (7), we get
$$_\nu W^+(X,Y)Z,X=\frac{1}{2}(C(X,Z)(Z)C(X,Y)(Y))=C(X,Y)(Y)$$
This proves equation (13) and the point (ii) in the Theorem.
For the point (iii), we use (13) and (9); from the codifferential of $`W^+`$, only the derivative along the normal vector, $`\nu `$, can be non-zero, as $`W^+`$ vanishes along $`M`$. This proves the Theorem. ∎
Remark.
The point (i) gives a condition for a self-dual manifold to admit an umbilic hypersurface : $`W^+`$ has to vanish along it, generically at order 0 (following (ii)), thus such a hypersurface, if it exists, is locally defined as the zero set of $`W^+`$.
Considering the umbilic hypersurfaces which arise from the LeBrun correspondence, the point (i) gives a condition for an open self-dual manifold to admit such a conformal infinity, namely it has to be asymptotically conformally flat.
## 4. Null-geodesics of complex conformal manifolds
### 4.1. Properties of the twistor space of a 3-dimensional conformal manifold.
Consider (M,$`c`$) a complex 3-dimensional conformal manifold. In some topological conditions (M has to be civilized ; as a geodesically convex set is always of this type , any point has a basis of civilized neighbourhoods), the twistor space of M is defined as the space $`Z`$ of null-geodesics of M, and it is a complex 3-manifold, containing twistor lines (i.e. rational curves with normal bundle isomorphic to $`𝒪(1)𝒪(1)`$), and endowed with a distribution of 2-planes $`F_{\overline{\gamma }}T_{\overline{\gamma }}Z`$ which is a contact structure .
We denote by $`\overline{\gamma }`$ the point of $`Z`$ corresponding, in the following way, to the null-geodesic $`\gamma `$ : some twistor lines tangent to $`F_{\overline{\gamma }}`$ (actually a non-empty open set in the space of curves of $`Z`$, tangent to $`F`$) correspond to the points of the null-geodesic $`\gamma `$<sup>3</sup><sup>3</sup>3deformations of these twistor lines are, in general, not tangent to the distribution $`F`$; they correspond to points in the self-dual ambient $`N`$ (and outside of $`M`$) arising from the LeBrun correspondence..
The first question we raise is whether there exist twistor lines tangent to any direction of a given 2-plane $`F_{\overline{\gamma }}`$; the answer is:
###### Theorem 2.
Let $`Z`$ be a twistor space of a conformal civilized 3-manifold M; let $`F_{\overline{\gamma }}T_{\overline{\gamma }}Z`$ be its contact structure. Suppose there is a point $`\overline{\gamma }Z`$ such that there are twistor lines tangent to any direction in $`F_{\overline{\gamma }}`$. Then $`Z`$ is projectively flat, and M is conformally flat.
This follows directly from , Theorem 3, which has a similar statement referring to the twistor space of a self-dual manifold; we simply apply it to the ambient N from the LeBrun correspondence; its conformal flatness implies the flatness of M (Theorem 1).
Remark.
The key point in the above cited Theorem is a twistorial interpretation of the Weyl tensor of a self-dual manifold N , Theorem 2, together with the remark that, for a given 2-plane $`F`$ in $`T_{\overline{\beta }}Z`$, the union of all twistor lines tangent to it (supposing there exists one pointing in any direction of $`F`$) is a complex surface which is smooth at $`\overline{\beta }Z`$. The above cited Theorem and the following one (Theorem $`3^{}`$ from ) show that this situation implies the vanishing of the Weyl tensor of N, $`W^+`$, in certain directions:
###### Theorem 3.
Let $`Z`$ be the twistor space of the (civilized) self-dual manifold N, and let $`\overline{\beta }Z`$ be a point in $`Z`$, corresponding to the $`\beta `$-surface $`\beta `$N; let $`F^\gamma T_{\overline{\beta }}Z`$ be a 2-plane, corresponding to the null-geodesic $`\gamma \beta `$ . Suppose that, for each direction $`\sigma (F^\gamma )(T_{\overline{\beta }}Z)`$, there is a smooth (non-necessarily compact) curve $`Z_\sigma `$ tangent to $`\sigma `$, such that:
(i) if $`\sigma `$ is tangent to a twistor line $`Z_x`$ at $`\overline{\beta }`$, then $`Z_\sigma =Z_x`$;
(ii) $`Z_\sigma `$ varies smoothly with $`\sigma (F^\gamma )`$.
Then
$$\overline{Z}_\beta ^\gamma :=\underset{\sigma (F^\gamma )}{}Z_\sigma $$
is a smooth surface around $`\beta `$, and $`W^+(F_x^\gamma )=0,x\gamma `$, where $`F_x^\gamma T_x𝐌`$ is the $`\alpha `$-plane containing $`\dot{\gamma }`$.
In other words, if the integral $`\alpha `$-cone corresponding to the 2-plane $`FT_{\overline{\gamma }}Z`$ — defined as the union of all twistor lines tangent $`F`$ — can be completed to a surface, smooth around its “vertex” $`\overline{\gamma }`$, then $`W^+`$ vanishes along the null-geodesic $`\gamma `$ (whose points correspond to the twistor lines that constitute the $`\alpha `$-cone ).
### 4.2. Compact null-geodesics and conformal flatness.
Our main result is:
###### Theorem 4.
Let M be a conformal $`n`$-manifold containing an immersed rational curve as null-geodesic. Then M is conformally flat.
This fact has been proven by Ye for complex projective manifolds — we are grateful to B. Klingler for having brought this paper into our attention. In the above Theorem, we do not make any assumption on the topology of the manifold, but only of one null-geodesic contained in it.
###### Proof.
The proof is different in the cases $`n>3`$ and $`n=3`$; one of the reasons is that conformal flatness reduces, in higher dimensions, to the vanishing of the Weyl tensor, while in dimension 3 it is a higher-order condition more difficult to handle. The first step (common to all cases) is to prove that a small deformation (seen just as a compact submanifold of M, ) of such a compact null-geodesic $`\gamma `$ is still a compact null-geodesic, and to characterize the global sections of the normal bundle of $`\gamma `$ as locally determined by Jacobi fields.
###### Lemma 2.
Let $`\gamma `$ be a (immersed) null-geodesic in $`(𝐌,c)`$. Let $`J`$ be a vector field along $`\gamma `$. Then the condition $`\dot{J}\dot{\gamma }`$ (where $`\dot{J}:=_{\dot{\gamma }}^gJ`$) is independent of the metric $`g`$ with respect to which we take the derivative $`^g`$.
###### Proof.
The relation between two Levi-Civita connections (or, more generally, Weyl structures) of metrics in the same conformal class, is given by :
(19)
$$_X^{}Y_XY=\theta (X)Y+\theta (Y)X\theta ^{\mathrm{}}g(X,Y),$$
where $`\theta `$ is a 1-form, and the rising of indices in $`\theta ^{\mathrm{}}`$ is made using the same (arbitrary) metric $`gc`$ as in the scalar product $`g(X,Y)`$. The Lemma immediately follows. ∎
Denote by $`N(\gamma )`$ the normal bundle of $`\gamma `$ in M, and by $`N^{}(\gamma )`$ its subbundle represented by vectors orthogonal to $`\dot{\gamma }`$. Fix a metric $`g`$ in the conformal class $`c`$. Let $`J`$ be a Jacobi field along $`\gamma `$, satisfying to the Jacobi equation
$$\ddot{J}=R^g(\dot{\gamma },J)\dot{\gamma }.$$
It represents an infinitesimal deformation of $`\gamma `$ through null-geodesics if and only if $`\dot{J}\dot{\gamma }`$. $`J`$ induces a section in the normal bundle $`N(\gamma )`$, or $`N^{}(\gamma )`$ if $`J\dot{\gamma }`$ in a point, hence everywhere. We want to prove that this section is independent of the connection $`^g`$:
###### Proposition 2.
The Jacobi equations on a null-geodesic $`\gamma `$M induce a second order linear differential operator $`P`$ on $`N(\gamma )`$, which, restricted to the sections $`J`$ such that $`\dot{J}\dot{\gamma }`$, depends only on the conformal structure $`c`$ of M. In particular, $`P`$ restricted to $`N^{}(\gamma )`$ is conformally invariant.
###### Proof.
For a Levi-Civita connection $``$ of a local metric on M, we locally define the following differential operator on the sections of $`T𝐌|_\gamma `$:
$$P:\mathrm{\Gamma }(T𝐌|_\gamma S^2(T\gamma ))\mathrm{\Gamma }(T𝐌|_\gamma ),$$
by $`P(Y;X,X):=_X_XY_{_XX}YR(X,Y)X`$. Because $`\gamma `$ is a null-geodesic, $`P`$ induces a (local) differential operator on $`N(\gamma )`$, and we need to relate $`P`$ to the corresponding operator $`P^{}`$ induced by another connection $`^{}`$. First we write
$$P(Y,X,X)=_X[X,Y]+_{[X,Y]}X[_XX,Y],$$
then we recall that another Levi-Civita connection $`^{}`$ is related to $``$ by the formula (19), such that we get
$$P^{}(Y;X,X)P(Y;X,X)=2[Y.(\theta (X))\theta ([X,Y])]Xg(_XY,X)\theta ^{\mathrm{}},$$
which is identically zero modulo $`T\gamma `$, provided that $`_XYX`$ (the latter condition being independent of the Levi-Civita connection, according to the previous Lemma). ∎
Using the fact that $`^1`$ is the union of two contractible sets $`U_1U_2`$ (on each of which the Jacobi equation, with any initial condition — the same for $`U_1`$ and for $`U_2`$ — in $`x_0U_1U_2`$, has a unique solution — and these solutions necessarily coincide on the connected intersection $`U_1U_2`$), we immediately get:
###### Proposition 3.
Let $`\gamma `$ be an immersed null-geodesic, diffeomorphic to a projective line $`^1`$. Then any local Jacobi field $`J`$ with $`\dot{J}\dot{\gamma }`$ induces a global normal field $`\nu ^J`$ on $`\gamma `$.
This has important consequences about the normal bundle of $`\gamma `$ in M, as Jacobi fields provide it with global sections; in particular, $`N(\gamma )/N^{}(\gamma )`$ is a line bundle admitting nowhere-vanishing sections, hence it is trivial; on the other hand, $`N^{}(\gamma )`$ is a $`(n2)`$–rank bundle over $`^1`$, admitting sections with any prescribed 1-jet (induced, again, by some appropriate Jacobi fields), hence
(20)
$$N^{}(\gamma )\underset{k=1}{\overset{n2}{}}𝒪(a_k);N(\gamma )\underset{k=1}{\overset{n2}{}}𝒪(a_k)𝒪(0);a_k^{}.$$
For $`a_k`$, this is the general form of a vector bundle over $`^1`$, according to a theorem of Grothendieck; the condition $`a_k1`$ arises from the existence of sections of $`N^{}(\gamma )`$ with prescribed 1-jet. We are going to show later that all $`a_k`$ are equal to 1. First we prove:
###### Proposition 4.
Null-geodesics close to a compact, simply-connected one are also compact and simply-connected, and they are generically embedded.
###### Proof.
We consider the projectivized bundle $`(𝒞)`$ of the isotropic cone $`𝒞T𝐌`$. It is a standard fact that any null-geodesic $`\gamma 𝐌`$ has a canonical horizontal lift $`\stackrel{~}{\gamma }(𝒞)`$ (depending only on the conformal structure), such that $`\pi _{}(T_s\stackrel{~}{\gamma })=T_x\gamma `$, where $`\pi :(𝒞)𝐌`$ is the projection, and $`s\pi ^1(x)`$.
Note that $`\stackrel{~}{\gamma }`$ is always embedded, even if $`\gamma `$ may have self-intersections (it is always immersed).
The lifts of the null-geodesics of M consist in a foliation of $`(𝒞)`$, which has a compact, simply-connected leaf, namely the lift $`\stackrel{~}{\gamma }`$ of our compact, simply-connected null-geodesic $`\gamma `$. By Reeb’s stability Theorem , then there is a saturated tubular neighbourhood of $`\stackrel{~}{\gamma }`$, diffeomorphic to $`\stackrel{~}{\gamma }\times D`$ (where $`D^{n2}`$ is a polydisc), such that the leaves close to $`\stackrel{~}{\gamma }`$ are identified, via the above diffeomorphism, to the (compact and simply-connected) curves $`\stackrel{~}{\gamma }\times \{\text{z}\},zD`$.
So all null-geodesics close to $`\gamma `$ are compact and simply connected. If $`\gamma `$ has self-intersections at the points $`x_1,\mathrm{},x_k`$, we blow-up M at those points, and the lift of $`\gamma `$ is now embedded. So must be then the lifts of the null-geodesics close to $`\gamma `$, as they are now deformations of the lifted (hence, embedded) curve. But, generically, such curves avoid the finite set of points $`x_1,\mathrm{},x_k`$; the corresponding null-geodesics must have been embedded from the beginning. ∎
From now on, according to the previous Proposition, we may suppose that $`\gamma `$ is a compact, simply-connected, embedded null-geodesic.
We compute the normal bundle of $`\gamma `$ in M, using the relation (20) and the projection $`\pi :(𝒞)𝐌`$, as follows: We have the following exact sequence of bundles:
(21)
$$0N^\pi (\stackrel{~}{\gamma })N(\stackrel{~}{\gamma })\pi ^{}N(\gamma )0,$$
where $`N^\pi (\stackrel{~}{\gamma })`$ is the normal subbundle of $`\stackrel{~}{\gamma }`$ represented by vectors tangent to the fibers of $`\pi `$ and $`N(\stackrel{~}{\gamma })`$ is the normal bundle of $`\stackrel{~}{\gamma }`$ in $`(𝒞)`$. In a point $`T_x\gamma \stackrel{~}{\gamma }(𝒞)`$, the fiber of $`\pi `$ is equal to $`(𝒞)_x`$, so the tangent space to it is isomorphic to $`\text{Hom}(T_x\gamma ,N_x^{}(\gamma ))`$, for the projective variety $`(𝒞)_x(T_x𝐌)`$. Thus
$$N^\pi (\stackrel{~}{\gamma })\text{Hom}(T\gamma ,N^{}(\gamma ))𝒪(2)N^{}(\gamma ),$$
as $`T\gamma T^1𝒪(2)`$.
The central bundle in the exact sequence (21) is trivial, because $`\stackrel{~}{\gamma }`$ is a leaf of a foliation. (20) and (21) imply that the Chern numbers $`a_1,\mathrm{},a_{n2}`$ are subject to the following constraint:
$$\underset{k=1}{\overset{n2}{}}(2a_k2)=0,$$
thus, as $`a_k1`$, we have $`a_k=1,k=\overline{1,n2}`$. We have then:
###### Proposition 5.
The normal bundle of a compact, simply-connected, null-geodesic $`\gamma `$ in M is isomorphic to
$$N(\gamma )\left(^{n2}𝒪(1)\right)𝒪(0),$$
and all its global sections are induced by Jacobi fields $`J`$ such that $`\dot{J}\dot{\gamma }`$. Moreover, the deformations of $`\gamma `$ as a compact curve coincide with the null-geodesics close to $`\gamma `$.
The last statement follows from the expression of the normal bundle, and a Theorem of Kodaira : the normal bundle satisfies $`H^1(N(\gamma ))=0`$, thus the space of deformations of $`\gamma `$ as a compact curve has dimension equal to $`dimH^0(N(\gamma ))=dim\mathrm{\Gamma }(N(\gamma ))=2n3`$, which is precisely the dimension of the space of null-geodesics, defined locally, over a geodesically convex open set, as the space of the leaves of the horizontal foliation of $`(𝒞)`$ . We conclude using the fact that all null-geodesics close to $`\gamma `$ are deformations of this one (as a compact, and simply-connected, curve).
We return to the proof of Theorem 4. Consider first the case when the dimension of M, $`n>3`$. We are going to show that the Weyl tensor of M, $`W`$, is identically zero (a special sub-case is $`n=4`$, when $`W=W^++W^{}`$). For simplicity, suppose first that $`n>4`$, and consider the fiber of $`N^{}(\gamma )`$ at an arbitrary point $`x\gamma `$: it has a non-degenerate conformal structure, induced from M, and the isotropy cone spans the whole fiber $`N^{}(\gamma )_x`$ (this still holds for $`n=4`$, but not for $`n=3`$). Let $`LN^{}(\gamma )`$ be an isotropic line. We have:
###### Lemma 3.
Let $`\gamma ^U`$ be an open set of the null-geodesic $`\gamma `$, on which local metrics $`g,g^{}c`$ are well defined. If a (locally defined, over $`\gamma ^U`$) line subbundle $`LN^{}(\gamma )`$ is parallel (or stable) for $`^g`$, then it is parallel for $`^g^{}`$ as well.
The proof is a straightforward application of (19).
Let $`(L_1)_x,\mathrm{},(L_{n2})_x`$ be linearly independent isotropic lines in $`N^{}(\gamma )_x`$. According to the previous Lemma, and to the fact that $`\gamma `$ is simply-connected, their parallel transport over $`\gamma `$ does not depend on any Levi-Civita connection of a metric in the conformal class. We get, thus, a global splitting
(22)
$$N^{}(\gamma )=L_1\mathrm{}L_{n2},$$
where the line bundles $`L_i,i=\overline{1,n2}`$ are all isotropic and parallel.
All these bundles are isomorphic to $`𝒪(b_i),b_i`$. As their sections are also sections of $`N^{}(\gamma )^{n2}𝒪(1)`$, they cannot vanish at more that 1 point, for each $`L_i`$, thus $`b_i1,i=\overline{1,n2}`$. On the other hand, the sum of all $`b_i`$’s has to be $`n2`$, thus $`b_i=1,i=\overline{1,n2}`$.
Let $`\varphi _i`$ be a section of $`L_i`$; from Proposition 5, it is locally represented by a Jacobi field $`J_i`$, for the metric $`gc`$. From the Jacobi equation, by taking the scalar product with $`J`$, we get:
(23)
$$g(R(\dot{\gamma },J)\dot{\gamma },J)=0,$$
and it is easy to see that, because of the fact that all scalar products involving $`\dot{\gamma }`$ and $`J`$ are 0, the term $`h\text{I}`$ of the curvature (1) satisfies the above relation identically. This equation holds, at $`x`$, for any isotropic vector $`J_x\dot{\gamma }_x`$, but we may consider also other compact, simply-connected, null-geodesics $`\gamma ^{}`$, containing $`x`$, and close to $`\gamma `$ (namely, small deformations of the compact curve $`\gamma `$).
For any 2-plane $`FT_x`$M, we denote by $`R^F`$ the sectional curvature of $`F`$:
$$R^F:S^2(\mathrm{\Lambda }^2F),R^F(XY,XY):=R(XY),XY,XY\mathrm{\Lambda }^2F,$$
and we have seen that, if $`F`$ is totally isotropic, $`R^F`$ depends only on $`W`$ (and on the metric $`g`$ only via the scalar product $`.,.`$).
###### Lemma 4.
If $`dim𝐌>4`$, the Weyl tensor at $`x𝐌`$, $`W_x`$, is determined by the sectional curvatures $`\{R^F,FU(F_0)\}`$, for $`U(F_0)`$ a small neighbourhood of the totally isotropic arbitrary 2-plane $`F_0T_x𝐌`$ in the Grassmanian of totally isotropic 2-planes at $`x`$.
Remark.
A similar statement holds in dimension 4, but in that case, the Grassmanian of totally isotropic 2-planes has 2 connected components; as a consequence, $`W^+`$ is determined by the sectional curvatures of $`\alpha `$-planes, and $`W^{}`$ by the sectional curvatures of the $`\beta `$-planes , Proposition 2.
###### Proof.
This reduces to the claim that $`W_x=0`$ if and only if
(24)
$$R^F=0,FU(F_0),$$
which is a problem of linear algebra. If we consider the space $`𝒦`$ of all curvature tensors $`R^{}`$ satisfying $`(R^{})^F=0,FU(F_0)`$, then this is a vector space, which is invariant to the action of $`𝔰𝔬(n,)`$ (which is the Lie algebra infinitesimal action corresponding to the action of $`SO(n,)`$ — note that the Grassmanian of totally isotropic planes is preserved by this action). But there are only 3 $`𝔰𝔬(n,)`$-irreducible components of the space of curvature tensors, and we have seen that for the Ricci-like tensors $`h\text{I}`$, the totally isotropic planes always have zero sectional curvature. Then either any Weyl tensor satisfying (24) is zero at $`x`$, or $`𝒦`$ contains the whole space of curvature tensors. The latter possibility can easily be excluded by an example of a curvature tensor $`K`$ satisfying:
$$K(X_0,Y_0)X_0=A_0,\text{ where }Y_0,A_0=1,\text{ and }$$
$$X_0,X_0=X_0,Y_0=Y_0,Y_0=X_0,A_0=0.$$
From (23) and the previous Lemma we conclude that $`W_x=0`$ for any $`x`$ contained in a compact, simply-connected, null-geodesic; but we know from Proposition 5 that the set of such points contains a neighbourhood of $`\gamma `$, thus, by analyticity of $`W`$, it vanishes identically.
The proof is similar in dimension 4 (note that, in the self-dual case, we can retrieve the result by applying Theorem 3; this is how we shall proceed for the case of dimension 3, using the LeBrun correspondence); the difference with the higher-dimensional case is that the splitting (22) is canonical, $`L_1`$ corresponding, say, to the $`\alpha `$-plane $`F^\alpha `$ containing $`\dot{\gamma }`$, and $`L_2`$ to the $`\beta `$-plane $`F^\beta `$ containing $`\dot{\gamma }`$. It is important now that each of $`L_1,L_2`$ is isomorphic to $`𝒪(1)`$, because the vanishing of $`R^{F^\alpha }`$ implies $`W^+0`$, and the vanishing of $`R^{F^\beta }`$ implies $`W^{}0`$ , Proposition 2. Thus the manifold $`(𝐌^4,c)`$ is conformally flat.
Consider now the particular case where $`n=3`$. We are going to use the LeBrun correspondence, then Theorem 3, to prove that M is then conformally flat. Note that we cannot use directly Theorem 1 and the above proven result for self-dual manifolds, as the ambient self-dual manifold $`N`$ can only be defined for a civilized (e.g. geodesically convex) 3-manifold.
We cover $`\gamma `$ with geodesically convex open sets $`U_i,i=\overline{1,n}`$, such that:
(25)
$$ij\text{ such that }U_iU_j\gamma \mathrm{},U_{ij}(U_iU_j),$$
where $`U_{ij}`$ is still geodesically convex (with respect to some particular Levi-Civita connection). This is possible by choosing $`U_i,i=\overline{1,n}`$, small enough . Then we choose a relatively compact tubular neighbourhood $`N(r_0)`$ of $`\gamma `$, such that its closure is covered by the $`U_i`$’s. We may choose this tubular neighbourhood small enough to be contained in the projection $`U`$, from $`(𝒞)`$, of a saturated neighbourhood (see Proposition 4) of the lift $`\stackrel{~}{\gamma }`$.
We consider then the twistor spaces $`Z_i`$, the spaces of null-geodesics of $`U_i`$. The compact, simply-connected, null-geodesics close to $`\gamma `$ identify (diffeomorphically) the neighbourhoods of $`\overline{\gamma }^iZ_i`$ with the space $`Z`$ of the deformations (contained in $`U`$) of $`\gamma `$ as a compact curve. We can see then (a small open set of) $`Z`$ as an open set common to all the $`Z_i`$’s:
Following LeBrun, we define the self-dual manifolds $`𝐍_i`$ as the spaces of projective lines in $`Z_i`$, with normal bundle $`𝒪(1)𝒪(1)`$. Then $`U_i`$ is an umbilic hypersurface in $`𝐍_i`$.
The local twistor spaces $`Z_i`$ admit contact structures, which coincide on $`Z`$, and contain projective lines $`Z_x^i`$ corresponding to points $`x\gamma U_i`$. If we denote by $`Z_{ij}`$ the twistor space of $`U_{ij}`$, then $`Z_{ij}`$ is identified to an open set in $`Z_i`$ and, at the same time, to an open set in $`Z_j`$, in particular the twistor lines $`Z_x^iZ^i`$ and $`Z_x^jZ^j`$ (corresponding to the same point $`xU_{ij}`$) are identified. Thus $`Z_x^iZ`$ and $`Z_x^jZ`$ coincide and we denote by $`Z_x`$ this (non-compact) curve in $`Z`$, and by $`F`$ the canonical contact structure of $`Z`$ (restricted from the ones of $`Z_i`$).
Remark.
We already have obtained that the integral $`\alpha `$-cone (i.e. the union of twistor lines passing through $`\overline{\gamma }`$ and tangent to $`F_{\overline{\gamma }}`$, see the comment after Theorem 3) corresponding to $`F_{\overline{\gamma }}`$ is a part of a smooth surface: the union of the lines $`Z_x`$, $`x\gamma `$. Thus, from Theorem 3, the Weyl tensor $`W_i^+`$ of the self-dual manifold $`𝐍_i`$ vanishes on the $`\alpha `$-planes generated by $`T\gamma `$. But this is nothing new: we know, from Theorem 1, that $`W_i^+`$ vanishes on $`U_i`$.
We intend to apply Theorem 3 to prove that $`W_i^+`$ vanishes on points close to $`U_i`$, but generically in $`𝐍_iU_i`$. We do that by showing that the integral $`\alpha `$-cones corresponding to planes $`F^{}T_{\overline{\gamma }}Z`$ close to $`F`$ are parts of smooth surfaces, then we conclude using Theorem 3.
First we choose Hermitian metrics $`h_i`$ on $`Z_i`$, such that they coincide (with $`h`$) on $`Z`$. We have a diffeomorphism between $`\gamma `$ and $`(F_{\overline{\gamma }})`$, so we choose relatively compact open sets in $`(T_{\overline{\gamma }}Z)`$, covering $`(F_{\overline{\gamma }})`$, with the following properties: As the metrics $`h_i`$ induce metrics on $`𝐍_i`$, we first choose a small enough distance $`r_1>0`$ such that
1. $`i`$, there is a sub-covering $`V_iU_i`$ of $`\gamma `$ such that the “tubular neighbourhoods” $`Q_i:=\{y𝐍_i|\mathrm{d}(y,\overline{V}_i)r_1\pi _i(y)\overline{V}_i\gamma \}`$ are compact ($`\mathrm{d}(y,\overline{V}_i)`$ is the distance between $`y`$ and $`\overline{V}_i`$, and $`\pi _i`$ is the “orthogonal projection” — for the Hermitian metric — from $`𝐍_i`$ to $`\gamma U_i`$; it is well defined because of the condition below);
2. $`r_1`$ is less than the bijectivity radius of the (Hermitian) exponentials for the points of $`\overline{V}_i`$ in $`𝐍_i`$, and for the points of $`\overline{V_iV_j}`$ in $`𝐍_{ij}`$ (if $`U_iU_j\gamma \mathrm{}.`$).
We have then
###### Lemma 5.
For any $`y_iQ_i𝐍_i`$, $`y_jQ_j𝐍_j`$ such that the curves $`Z_{y_i}:=Z_{y_i}^iZ`$, $`Z_{y_j}:=Z_{y_j}^jZ`$ are tangent to the same direction in $`\overline{\gamma }Z`$, the respective curves $`Z_{y_i},Z_{y_j}`$ coincide.
###### Proof.
We first note that the projection $`\pi _i`$ from $`𝐍_i`$ to $`\gamma U_i`$ is equivalent to the $`h`$–orthogonal projection of the direction of $`T_{\overline{\gamma }}Z_{y_i}`$ to a direction in $`F_{\overline{\gamma }}`$, so $`\pi _i(y_i)=\pi _j(y_j)=:y\gamma `$; thus $`y`$ belongs to both $`U_i`$ and $`U_j`$, and we use again the twistor space $`Z_{ij}`$ to conclude that $`Z_{y_i}`$ and $`Z_{y_j}`$ are “restrictions” to $`Z`$ of the same projective line (as they both have the same tangent space at $`\overline{\gamma }`$) $`Z_{y_{ij}}^{ij}`$, for a point $`y_{ij}𝐍_{ij}`$. ∎
Now we have a tubular neighbourhood $`S(T_{\overline{\gamma }}Z)`$ of $`(F_{\overline{\gamma }})`$, of radius $`r_1/2`$, such that, for any 2-plane $`F^{}S`$, the conditions in Theorem 3 are satisfied (considering any of the local twistor spaces $`Z_i`$).
We recall that, via the LeBrun correspondence, a point $`\overline{\gamma }_0`$ in the twistor space of $`𝐌_0`$, $`Z_0`$, is identified to the point $`\overline{\beta }_0`$ in the twistor space of $`𝐍_0`$, still denoted by $`Z_0`$. They correspond to the null-geodesic $`\gamma _0𝐌_0`$, resp. to the $`\beta `$-surface $`\beta _0𝐍_0`$, such that $`\gamma _0\beta _0`$ , . The planes $`F^{}`$ above are included in $`T_{\overline{\gamma }}ZT_{\overline{\beta ^i}}Z_i`$, and they correspond to null-geodesics in $`𝐍_i`$ contained in $`\beta ^i`$ , .
By Theorem 3, we conclude that the Weyl tensor $`W_i^+`$ of $`𝐍_i`$ vanishes along all null-geodesics of $`𝐍_i`$, close (in $`Q_i`$) to $`\gamma `$ and included in the $`\beta `$-surface $`\beta ^i`$, determined by $`\gamma `$. This means that $`W^+`$ vanishes everywhere on $`\beta ^i`$. By deforming $`\gamma `$, we obtain that $`W_i^+`$ vanishes on a neighbourhood of $`U_i`$ in $`𝐍_i`$, hence $`𝐍_i`$ is conformally flat.
It follows from Theorem 1 that $`U_i`$, hence M, is conformally flat. ∎
Mathematisches Institut
Humboldt Universität zu Berlin
Unter den Linden 6, 10099 Berlin
Germany
e-mail: belgun@mathematik.hu-berlin.de
|
warning/0002/cond-mat0002038.html
|
ar5iv
|
text
|
# SU-4240-714 The Statistical Mechanics of Membranes
## 1 Introduction
The statistical mechanics of one-dimensional structures (polymers) is fascinating and has proved to be fruitful from the fundamental and applied points of view . The key reasons for this success lie in the notion of universality and the relative simplicity of one-dimensional geometry. Many features of the long-wavelength behavior of polymers are independent of the detailed physical and chemical nature of the monomers that constitute the polymer building blocks and their bonding into macromolecules. These microscopic details simply wash out in the thermodynamic limit of large systems and allow predictions of critical exponents that should apply to a wide class of microscopically distinct polymeric systems. Polymers are also sufficiently simple that considerable analytic and numerical progress has been possible. Their statistical mechanics is essentially that of ensembles of various classes of random walks in some $`d`$-dimensional bulk or embedding space.
A natural extension of these systems is to intrinsic two-dimensional structures which we may call generically call membranes. The statistical mechanics of these random surfaces is far more complex than that of polymers because two-dimensional geometry is far richer than the very restricted geometry of lines. Even planar two-dimensional — monolayers — are complex, as evidenced by the KTNHY theory of defect-mediated melting of monolayers with two distinct continuous phase transitions separating an intermediate hexatic phase, characterized by quasi-long-range bond orientational order, from both a low-temperature crystalline phase and a high-temperature fluid phase. But full-fledged membranes are subject also to shape fluctuations and their macroscopic behavior is determined by a subtle interplay between their particular microscopic order and the entropy of shape and elastic deformations. For membranes, unlike polymers, distinct types of microscopic order (crystalline, hexatic, fluid) will lead to distinct long-wavelength behavior and consequently a rich set of universality classes.
Flexible membranes are an important member of the enormous class of soft condensed matter systems , those which respond easily to external forces. Their physical properties are to a considerable extent dominated by the entropy of thermal fluctuations.
In this review we will describe some of the presently understood behavior of crystalline (fixed-connectivity), hexatic and fluid membranes, including the relevance of self-avoidance, intrinsic anisotropy and topological defects. Emphasis will be given to the role of the renormalization group in elucidating the critical behavior of membranes. The polymer pastures may be lovely but a dazzling world awaits those who wander into the membrane meadows.
The outline of the review is the following. In sec. 2 we describe a variety of important physical examples of membranes, with representatives from the key universality classes. In sec. 3 we introduce basic notions from the renormalization group and some formalism that we will use in the rest of the review. In sec. 4 we review the phase structure of crystalline membranes for both phantom and self-avoiding membranes, including a thorough discussion of the fixed-point structure, RG flows and critical exponents of each global phase. In sec. 5 we turn to the same issues for intrinsically anisotropic membranes, with the new feature of the tubular phase. In sec. 6 we address the consequences of allowing for membrane defects, leading to a discussion of the hexatic membrane universality class. We end with a brief discussion of fluid membranes in sec. 7 and conclusions.
## 2 Physical examples of membranes
There are many concrete realizations of membranes in nature, which greatly enhances the significance of their study. Crystalline membranes, sometimes termed tethered or polymerized membranes, are the natural generalization of linear polymer chains to intrinsically two-dimensional structures. They possess in-plane elastic moduli as well as bending rigidity and are characterized by broken translational invariance in the plane and fixed connectivity resulting from relatively strong bonding. Geometrically speaking they have a preferred two-dimensional metric. Let’s look at some of the examples. One can polymerize suitable chiral oligomeric precursors to form molecular sheets . This approach is based directly on the idea of creating an intrinsically two-dimensional polymer. Alternatively one can permanently cross-link fluid-like Langmuir-Blodgett films or amphiphilic bilayers by adding certain functional groups to the hydrocarbon tails and/or the polar heads as shown schematically in Fig. 1.
The cytoskeletons of cell membranes are beautiful and naturally occurring crystalline membranes that are essential to cell membrane stability and functionality. The simplest and most thoroughly studied example is the cytoskeleton of mammalian erythrocytes (red blood cells). The human body has roughly $`5\times 10^{13}`$ red blood cells. The red blood cell cytoskeleton is a fishnet-like network of triangular plaquettes formed primarily by the proteins spectrin and actin. The links of the mesh are spectrin tetramers (of length approximately 200 nm) and the nodes are short actin filaments (of length 37 nm and typically 13 actin monomers long) , as seen in Fig.3 and Fig.3. There are roughly 70,000 triangular plaquettes in the mesh altogether and the cytoskeleton as a whole is bound by ankyrin and other proteins to the cytoplasmic side of the fluid phospholipid bilayer which constitutes the other key component of the red blood cell membrane.
There are also inorganic realizations of crystalline membranes. Graphitic oxide (GO) membranes are micron size sheets of solid carbon with thicknesses on the order of 10Å, formed by exfoliating carbon with a strong oxidizing agent. Their structure in an aqueous suspension has been examined by several groups . Metal dichalcogenides such as MoS<sub>2</sub> have also been observed to form rag-like sheets . Finally similar structures occur in the large sheet molecules, shown in Fig.4, believed to be an ingredient in glassy $`B_2O_3`$.
In contrast to crystalline membranes, fluid membranes are characterized by vanishing shear modulus and dynamical connectivity. They exhibit significant shape fluctuations controlled by an effective bending rigidity parameter.
A rich source of physical realizations of fluid membranes is found in amphiphilic systems . Amphiphiles are molecules with a two-fold character – one part is hydrophobic and another part hydrophilic. The classic examples are lipid molecules, such as phospholipids, which have polar or ionic head groups (the hydrophilic component) and hydrocarbon tails (the hydrophobic component). Such systems are observed to self-assemble into a bewildering array of ordered structures, such as monolayers, planar (see Fig.5) and spherical bilayers (vesicles or liposomes) (see Fig. 6) as well as lamellar, hexagonal and bicontinuous phases . In each case the basic ingredients are thin and highly flexible surfaces of amphiphiles. The lipid bilayer of cell membranes may itself be viewed as a fluid membrane with considerable disorder in the form of membrane proteins (both peripheral and integral) and with, generally, an attached crystalline cytoskeleton, such as the spectrin/actin mesh discussed above.
A complete understanding of these biological membranes will require a thorough understanding of each of its components (fluid and crystalline) followed by the challenging problem of the coupled system with thermal fluctuations, self-avoidance, potential anisotropy and disorder. The full system is currently beyond the scope of analytic and numerical methods but there has been considerable progress in the last fifteen years.
Related examples of fluid membranes arise when the surface tension between two normally immiscible substances, such as oil and water, is significantly lowered by the surface action of amphiphiles (surfactants), which preferentially orient with their polar heads in water and their hydrocarbon tails in oil. For some range of amphiphile concentration both phases can span the system, leading to a bicontinuous complex fluid known as a microemulsion. The oil-water interface of a microemulsion is a rather unruly fluid surface with strong thermal fluctuations (see Fig.8).
The structures formed by membrane/polymer complexes are of considerable current theoretical, experimental and medical interest. To be specific it has recently been found that mixtures of cationic liposomes (positively charged vesicles) and linear DNA chains spontaneously self-assemble into a coupled two-dimensional smectic phase of DNA chains embedded between lamellar lipid bilayers . For the appropriate regime of lipid concentration the same system can also form an inverted hexagonal phase with the DNA encapsulated by cylindrical columns of liposomes (see Fig.9). In both these structures the liposomes may act as non-viral carriers (vectors) for DNA with many potentially important applications in gene therapy . Liposomes themselves have long been studied and utilized in the pharmaceutical industry as drug carriers .
On the materials science side the self-assembling ability of membranes is being exploited to fabricate microstructures for advanced material development. One beautiful example is the use of chiral-lipid based fluid microcylinders (tubules) as a template for metallization. The resultant hollow metal needles may be half a micron in diameter and as much as a millimeter in length , as illustrated in Fig.8. They have potential applications as, for example, cathodes for vacuum field emission and microvials for controlled release .
## 3 The Renormalization Group
The Renormalization Group (RG) has provided an extremely general framework that has unified whole areas of physics and chemistry . It is beyond the scope of this review to discuss the RG formalism in detail but there is an ample literature to which we refer the reader (see the articles in this issue). It is the goal of this review to apply the RG framework to the statistical mechanics of membranes, and for this reason we briefly emphasize and review some well known aspects of the RG and its related $`\epsilon `$-expansion.
The RG formalism elegantly shows that the large distance properties (or equivalently low $`p`$-limit) of different models are actually governed by the properties of the corresponding Fixed Point (FP). In this way one can compute observables in a variety of models, such as a molecular dynamics simulation or a continuum Landau phenomenological approach, and obtain the same long wavelength result. The main idea is to encode the effects of the short-distance degrees of freedom in redefined couplings. A practical way to implement such a program is the Renormalization Group Transformation (RGT), which provides an explicit prescription for integrating out all the high $`p`$-modes of the theory. One obtains the large-distance universal term of any model by applying a very large ($`\mathrm{}`$ to be rigorous) number of RGTs.
The previous approach is very general and simple but presents the technical problem of the proliferation in the number of operators generated along the RG flow. There are established techniques to control this expansion, one of the most successful ones being the $`\epsilon `$-expansion. The $`\epsilon `$-expansion may also performed via a field theoretical approach using Feynman diagrams and dimensional regularization within a minimal subtraction scheme, which we briefly discuss below. Whereas it is true that this technique is rather abstract and intuitively not very close to the physics of the model, we find it computationally much simpler.
Generally we describe a particular model by several fields $`\{\varphi ,\chi ,\mathrm{}\}`$ and we construct the Landau free energy by including all terms compatible with the symmetries and introducing new couplings $`(u,v,\mathrm{})`$ for each term. The Landau free energy may be considered in arbitrary dimension $`d`$, and then, one usually finds a Gaussian FP (quadratic in the fields) which is infrared stable above a critical dimension ($`d_U`$). Below $`d_U`$ there are one or several couplings that define relevant directions. One then computes all physical quantities as a function of $`\epsilon d_Ud`$, that is, as perturbations of the Gaussian theory.
In the field theory approach, we introduce a renormalization constant for each field $`(Z_\varphi ,Z_\chi ,\mathrm{})`$ and a renormalization constant $`(Z_u,Z_v,\mathrm{})`$ for each relevant direction below $`d_U`$. If the model has symmetries, there are some relations among observables (Ward identities) and some of these renormalization constants may be related. This not only reduces their number but also has the added bonus of providing cross-checks in practical calculations. Within dimensional regularization, the infinities of the Feynman diagrams appear as poles in $`\epsilon `$, which encode the short-distance details of the model. If we use these new constants ($`Z`$’s) to absorb the poles in $`\epsilon `$, thereby producing a complete set of finite Green’s functions, we have succeeded in carrying out the RG program of including the appropriate short-distance information in redefined couplings and fields. This particular prescription of absorbing only the poles in $`\epsilon `$ in the $`Z`$’s is called the Minimal Subtraction Scheme (MS), and it considerably simplifies practical calculations.
As a concrete example, we consider the theory of a single scalar field $`\varphi `$ with two independent coupling constants. The one-particle irreducible Green’s function has the form
$$\mathrm{\Gamma }_R^N(𝐤_i;u_R,v_R,M)=Z_\varphi ^{N/2}\mathrm{\Gamma }^N(𝐤_i;u,v;\frac{1}{\epsilon }),$$
(1)
where the function on the left depends on a new parameter $`M`$, which is unavoidably introduced in eliminating the poles in $`\epsilon `$. The associated correlator also depends on redefined couplings $`u_R`$ and $`v_R`$. The rhs depends on the poles in $`\epsilon `$, but its only dependence on $`M`$ arises through $`Z_\varphi `$. This observation allows one to write
$`M{\displaystyle \frac{d}{dM}}\left(Z_\varphi ^{N/2}\mathrm{\Gamma }_R^{(N)}\right)=`$
$`=`$ $`\left(M{\displaystyle \frac{}{M}}+\beta (u_R){\displaystyle \frac{}{u_R}}+\beta (v_R){\displaystyle \frac{}{v_R}}{\displaystyle \frac{N}{2}}\mathrm{\Gamma }_\varphi \right)\mathrm{\Gamma }_R^{(N)}=0,`$
where
$`u_R=M^\epsilon F(Z_\varphi ,Z_\chi ,\mathrm{}|Z_u)u`$ $`,`$ $`v_R=M^\epsilon F(Z_\varphi ,Z_\chi ,\mathrm{}|Z_v)v`$ (3)
$`\beta _u(u_R,v_R)=\left(M{\displaystyle \frac{u_R}{M}}\right)|_{u,v}`$ $`,`$ $`\beta _v(u_R,v_R)=\left(M{\displaystyle \frac{v_R}{M}}\right)|_{u,v}`$
$`\gamma _\varphi `$ $`=`$ $`\left(M{\displaystyle \frac{\mathrm{ln}Z_\varphi }{M}}\right)|_{u,v}.`$
The $`\beta `$-functions control the running of the coupling by
$$M\frac{du_R}{dM}=\beta _u(u_R,v_R),M\frac{dv_R}{dM}=\beta _v(u_R,v_R)$$
(4)
The existence of a FP, at which couplings cease to flow, requires $`\beta (u_R^{},v_R^{})=0`$ for all $`\beta `$-functions of the model. Those are the most important aspects of the RG we wanted to review. In Appendix B we derive more appropriate expressions of the RG-functions for practical convenience. For a detailed exposition of the $`\epsilon `$-expansion within the field theory framework we refer to the excellent book by Amit.
## 4 Crystalline Membranes
A crystalline membrane is a two dimensional fish-net structure with bonds (links) that never break - the connectivity of the monomers (nodes) is fixed. It is useful to keep the discussion general and consider $`D`$-dimensional objects embedded in $`d`$-dimensional space. These are described by a $`d`$-dimensional vector $`\stackrel{}{r}(𝐱)`$, with $`𝐱`$ the $`D`$-dimensional internal coordinates, as illustrated in Fig.10. The case $`(d=3,D=2)`$ corresponds to the physical crystalline membrane.
To construct the Landau free energy of the model, one must recall that the free energy must be invariant under global translations, so the order parameter is given by derivatives of the embedding $`\stackrel{}{r}`$, that is $`\stackrel{}{t}_\alpha =\frac{\stackrel{}{r}}{u_\alpha }`$, with $`\alpha =1,\mathrm{},D`$. This latter condition, together with the invariance under rotations (both in internal and bulk space), give a Landau free energy
$`F(\stackrel{}{r})`$ $`=`$ $`{\displaystyle d^D𝐱\left[\frac{1}{2}\kappa (_\alpha ^2\stackrel{}{r})^2+\frac{t}{2}(_\alpha \stackrel{}{r})^2+u(_\alpha \stackrel{}{r}_\beta \stackrel{}{r})^2+v(_\alpha \stackrel{}{r}^\alpha \stackrel{}{r})^2\right]}`$ (5)
$`+`$ $`{\displaystyle \frac{b}{2}}{\displaystyle d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐲))},`$
where higher order terms may be shown to be irrelevant at long wavelength, as discussed later. The physics in Eq.(5) depends on five parameters,
* $`\kappa `$, bending rigidity : This is the coupling to the extrinsic curvature (the square of the Gaussian mean curvature). Since reparametrization invariance is broken for crystalline membranes, this term may be replaced by its long-wavelength limit. For large and positive bending rigidities flatter surfaces are favored.
* $`t,u,v`$, elastic constants : These coefficients encode the microscopic elastic properties of the membrane. In a flat phase, they may be related to the Lamé coefficients of Landau elastic theory (see sect. 4.1.3).
* $`b`$, Excluded volume or self-avoiding coupling : This is the coupling that imposes an energy penalty for the membrane to self-intersect. The case $`b=0`$, i. e. no self-avoidance, corresponds to a phantom model.
We generally expand $`\stackrel{}{r}(𝐱)`$ as
$$\stackrel{}{r}(𝐱)=(\zeta 𝐱+𝐮(𝐱),h(𝐱)),$$
(6)
with $`𝐮`$ the $`D`$-dimensional phonon in-plane modes, and $`h`$ the $`dD`$ out-of-plane fluctuations. If $`\zeta =0`$ the model is in a rotationally invariant crumpled phase, where the typical surfaces have fractal dimension, and there is no real distinction between the in-plane phonons and out-of plane modes. For a pictorial view, see cases a) and b) in Fig.11.
If $`\zeta 0`$ the membrane is flat up to small fluctuations and the full rotational symmetry is spontaneously broken. The fields $`h`$ are the analog of the Goldstone bosons and they have different naive scaling properties than $`𝐮`$. See Fig.11 for a visualization of a typical configuration in the flat phase.
We will begin by studying the phantom case first. This simplified model may even be relevant to physical systems since one can envision membranes that self-intersect (at least over some time scale). One can also view the model as a fascinating toy model for understanding the more physical self-avoiding case to be discussed later. Combined analytical and numerical studies have yielded a thorough understanding of the phase diagram of phantom crystalline membranes.
### 4.1 Phantom
The Phantom case corresponds to setting $`b=0`$ in the free energy Eq.(5):
$$F(\stackrel{}{r})=d^D𝐱\left[\frac{1}{2}\kappa (_\alpha ^2\stackrel{}{r})^2+\frac{t}{2}(_\alpha \stackrel{}{r})^2+u(_\alpha \stackrel{}{r}_\beta \stackrel{}{r})^2+v(_\alpha \stackrel{}{r}^\alpha \stackrel{}{r})^2\right].$$
(7)
The mean field effective potential, using the decomposition of Eq.(6), becomes
$$V(\zeta )=D\zeta ^2(\frac{t}{2}+(u+vD)\zeta ^2),$$
(8)
with solutions
$$\zeta ^2=\{\begin{array}{cc}\hfill 0:& t0\hfill \\ \hfill \frac{t}{4(u+vD)}:& t<0.\hfill \end{array}$$
(9)
There is, consequently, a flat phase for $`t<0`$ and a crumpled phase for $`t>0`$, separated by a crumpling transition at $`t=0`$ (see Fig.12).
The actual phase diagram agrees qualitatively with the phase diagram of the model shown schematically in Fig.13. The crumpled phase is described by a line of equivalent FPs(GFP). There is a general hyper-surface, whose projection onto the $`\kappa t`$ plane corresponds to a one-dimensional curve (CTH), which corresponds to the crumpling transition. Within the CTH there is an infrared stable FP (CTFP) which describes the large distance properties of the crumpling transition. Finally, for large enough values of $`\kappa `$ and negative values of $`t`$, the system is in a flat phase described by the corresponding infra-red stable FP (FLFP) <sup>1</sup><sup>1</sup>1The FLFP is actually a line of equivalent fixed points.. Although the precise phase diagram turns out to be slightly more complicated than the one depicted in Fig.13, the additional subtleties do not modify the general picture.
The evidence for the phase diagram depicted in Fig.13 comes from combining the results of a variety of analytical and numerical calculations. We present in detail the results obtained from the $`\epsilon `$-expansion since they have wide applicability and allow a systematic calculation of the $`\beta `$-function and the critical exponents. We also describe briefly results obtained with other approaches.
#### 4.1.1 The crumpled phase
In the crumpled phase, the free energy Eq.(7) for $`D2`$ simplifies to
$$F(\stackrel{}{r})=\frac{t}{2}d^D𝐱(_\alpha \stackrel{}{r})^2+\text{Irrelevant terms},$$
(10)
since the model is completely equivalent to a linear sigma model in $`D2`$ dimensions having $`O(d)`$ symmetry, and therefore all derivative operators in $`\stackrel{}{r}`$ are irrelevant by power counting. The parameter $`t`$ labels equivalent Gaussian FPs, as depicted in Fig.13. In RG language, it defines a completely marginal direction. This is true provided the condition $`t>0`$ is satisfied. The large distance properties of this phase are described by simple Gaussian FPs and therefore the connected Green’s function may be calculated exactly with result
$$G(𝐱)\{\begin{array}{cc}|𝐱|^{2D}& D2\hfill \\ \mathrm{log}|𝐱|& D=2\hfill \end{array}$$
(11)
The associated critical exponents may also be computed exactly. The Hausdorff dimension $`d_H`$, or equivalently the size exponent $`\nu =D/d_H`$, is given (for the membrane case $`D=2`$) by
$$d_H=\mathrm{}(\nu =0)R_G^2\mathrm{log}L.$$
(12)
The square of the radius of gyration $`R_G^2`$ scales logarithmically with the membrane size $`L`$. This result is in complete agreement with numerical simulations of tethered membranes in the crumpled phase where the logarithmic behavior of the radius of gyration is accurately checked . Reviews may be found in .
#### 4.1.2 The Crumpling Transition
The Free energy is now given by
$$F(\stackrel{}{r})=d^D𝐱\left[\frac{1}{2}(_\alpha ^2\stackrel{}{r})^2+u(_\alpha \stackrel{}{r}_\beta \stackrel{}{r})^2+\widehat{v}(_\alpha \stackrel{}{r}^\alpha \stackrel{}{r})^2\right],$$
(13)
where the dependence on $`\kappa `$ may be included in the couplings $`u`$ and $`\widehat{v}`$. With the leading term having two derivatives, the directions defined by the couplings $`u`$ and $`\widehat{v}`$ are relevant by naive power counting for $`D4`$. This shows that the model is amenable to an $`\epsilon `$-expansion with $`\epsilon =4D`$. For practical purposes, it is more convenient to consider the coupling $`v=\widehat{v}+\frac{u}{D}`$. We provide the detailed derivation of the corresponding $`\beta `$ functions in Appendix D. The result is
$$\begin{array}{ccc}\beta _u(u_R,v_R)& =& \hfill \epsilon u_R+\frac{1}{8\pi ^2}\left\{(d/3+65/12)u_R^2+6u_Rv_R+4/3v_R^2\right\}\\ \beta _v(u_R,v_R)& =& \hfill \epsilon v_R+\frac{1}{8\pi ^2}\left\{21/16u_R^2+21/2u_Rv_R+(4d+5)v_R^2\right\}\end{array}$$
(14)
Rather surprisingly, this set of $`\beta `$ functions does not possess a FP, except for $`d>219`$. This result would suggest that the crumpling transition is first order for $`d=3`$. Other estimates, however, give results which are consistent with the crumpling transition being continuous. These are
* Limit of large elastic constants: The Crumpling transition is approached from the flat phase, in the limit of infinite elastic constants. The model is
$$H_{NL}=d^D\sigma \frac{\kappa }{2}(\mathrm{\Delta }\stackrel{}{r})^2,$$
(15)
with the further constraint $`_\alpha \stackrel{}{r}_\beta \stackrel{}{r}=\delta _{\alpha \beta }`$. Remarkably, the $`\beta `$-function may be computed within a large $`d`$ expansion, yielding a continuous crumpling transition with size exponent at the transition (for $`D=2`$)
$$d_H=\frac{2d}{d1}\nu =1\frac{1}{d}.$$
(16)
* SCSA Approximation: The Schwinger-Dyson equations for the model given by Eq.(13) are truncated to include up to four point vertices. The result for the Hausdorff dimension and size exponent is
$$d_H=2.73\nu =0.732.$$
(17)
* MCRG Calculation : The crumpling transition is studied using MCRG (Monte Carlo Renormalization Group) techniques. Again, the transition is found to be continuous with exponents
$$d_H=2.64(5)\nu =0.85(9).$$
(18)
Each of these three independent estimates give a continuous crumpling transition with a size exponent in the range $`\nu 0.7\pm .15`$. It would be interesting to understand how the $`\epsilon `$-expansion must be performed in order to reconcile it with these results.
Further evidence for the crumpling transition being continuous is provided by numerical simulations where the analysis of observables like the specific heat (see Fig.14) or the radius of gyration radius give textbook continuous phase transitions, although the precise value of the exponents at the transition are difficult to pin down. Since this model has also been explored numerically with different discretizations on several lattices, there is clear evidence for universality of the crumpling transition , again consistent with a continuous transition. In Appendix C we present more details of suitable discretizations of the energy for numerical simulations of membranes.
#### 4.1.3 The Flat Phase
In a flat membrane (see Fig.15), we consider the strain tensor
$$u_{\alpha \beta }=_\alpha u_\beta +_\beta u_\alpha +_\alpha h_\beta h.$$
(19)
The free energy Eq.(7) becomes
$$F(𝐮,h)=d^D𝐱\left[\frac{\widehat{\kappa }}{2}(_\alpha _\beta h)^2+\mu u_{\alpha \beta }u^{\alpha \beta }+\frac{\lambda }{2}(u_\alpha ^\alpha )^2\right],$$
(20)
where we have dropped irrelevant terms. One recognizes the standard Landau Free energy of elasticity theory, with Lamé coefficients $`\mu `$ and $`\lambda `$, plus an extrinsic curvature term, with bending rigidity $`\widehat{\kappa }`$. These couplings are related to the original ones in Eq.(7) by $`\mu =u\zeta ^{4D}`$, $`\lambda =2v\zeta ^{4D}`$, $`\widehat{\kappa }=\kappa \zeta ^{4D}`$ and $`t=4(\mu +\frac{D}{2}\lambda )\zeta ^{D2}`$.
The large distance properties of the flat phase for crystalline membranes are completely described by the Free energy Eq.(20). Since the bending rigidity may be scaled out at the crumpling transition, the free energy becomes a function of $`\frac{\mu }{\kappa ^2}`$ and $`\frac{\lambda }{\kappa ^2}`$. The $`\beta `$-function for the couplings $`u,v`$ at $`\kappa =1`$ may be calculated within an $`\epsilon `$-expansion, which we describe in detail in Appendix E. Let us recall that the dependence on $`\kappa `$ may be trivially restored at any stage. The result is
$`\beta _\mu (\mu _R,\lambda _R)`$ $`=`$ $`\epsilon \mu _R+{\displaystyle \frac{\mu _R^2}{8\pi ^2}}({\displaystyle \frac{d_c}{3}}+20A)`$ (21)
$`\beta _\lambda (\mu _R,\lambda _R)`$ $`=`$ $`\epsilon \lambda _R+{\displaystyle \frac{1}{8\pi ^2}}({\displaystyle \frac{d_c}{3}}\mu _R^2+2(d_c+10A)\lambda _R\mu _R+2d_c\lambda _R^2),`$
where $`d_c=dD`$, and $`A=\frac{\mu _R+\lambda _R}{2\mu _R+\lambda _R}`$. These $`\beta `$ functions show four fixed points whose actual values are shown in Table 1.
As apparent from Fig.16, the phase diagram of the flat phase turns out to be slightly more involved than the one shown in Fig.13, as there are three FPs in addition to the FLFP already introduced. These additional FPs are infra-red unstable, however, and can only be reached for very specific values of the Lamé coefficients, so for any practical situation we can regard the FLFP as the only existing FP in the flat phase.
The properties of the flat phase
The flat phase is a very important phase as will be apparent once we study the full model, including self-avoidance. For that reason we turn now to a more detailed study of its most important properties.
Fig.11 (c) gives an intuitive visualization of a crystalline membrane in the flat phase. The membrane is essentially a flat two dimensional object up to fluctuations in the perpendicular direction. The rotational symmetry of the model is spontaneously broken, being reduced from $`O(d)`$ to $`O(dD)\times O(D)`$. The remnant rotational symmetry is realized in Eq.(20) as
$`h_i(𝐱)`$ $``$ $`h_i(𝐱)+A^{i\alpha }𝐱_\alpha `$ (22)
$`u_\alpha (𝐱)`$ $``$ $`u_\alpha A^{i\alpha }h_i{\displaystyle \frac{1}{2}}\delta ^{ij}(A^{i\alpha }A^{\beta j}𝐱_\beta ),`$
where $`A^{i\alpha }`$ is a $`D\times (dD)`$ matrix. This relation is very important as it provides Ward identities which simplify enormously the renormalization of the theory.
Let us first study the critical exponents of the model. There are two key correlators, involving the in-plane and the out-of-plane phonon modes. Using the RG equations, it is easy to realize that at any given FP, the low-$`p`$ limit of the model is given by
$`\mathrm{\Gamma }_{uu}(\stackrel{}{p})`$ $``$ $`|\stackrel{}{p}|^{2+\eta _u}`$ (23)
$`\mathrm{\Gamma }_{hh}(\stackrel{}{p})`$ $``$ $`|\stackrel{}{p}|^4\kappa (\stackrel{}{p})|\stackrel{}{p}|^{4\eta },`$
where the last equation defines the anomalous elasticity $`\kappa (\stackrel{}{p})`$ as a function of momenta $`\stackrel{}{p}`$. These two exponents are not independent, since they satisfy the scaling relation
$$\eta _u=4D2\eta ,$$
(24)
which follows from the Ward identities associated with the remnant rotational symmetry (Eqn.(22). Another important exponent is the roughness exponent $`\zeta `$, which measures the fluctuations transverse to the flat directions. It can be expressed as $`\zeta =\frac{4D\eta }{2}`$.
The long wavelength properties of the flat phase are described by the FLFP (see Fig.16). Since the FLFP occurs at non-zero renormalized values of the Lamé coefficients, the associated critical exponents discussed earlier are clearly non-Gaussian. Within an $`\epsilon `$-expansion, the values for the critical exponents are given in Table 1. There are alternative estimates available from different methods. These are
* Numerical Simulation: In a large scale simulation of the model was performed using very large meshes. The results obtained for the critical exponents are very reliable, namely
$$\begin{array}{ccc}\eta _u=0.50(1)\hfill & \eta =0.750(5)\hfill & \zeta =0.64(2)\hfill \end{array}$$
(25)
For a review of numerical results see .
* SCSA Approximation: This consists of suitably truncating the Schwinger-Dyson equations to include up to four-point correlation functions . The result for general $`d`$ is
$$\eta (d)=\frac{4}{d_c+(162d_c+d_c^2)^{1/2}},$$
(26)
which for $`d=3`$ gives
$$\begin{array}{ccc}\eta _u=0.358\hfill & \eta =0.821\hfill & \zeta =0.59\hfill \end{array}$$
(27)
* Large d expansion: The result is
$$\eta =\frac{2}{d}\eta (3)=2/3$$
(28)
We regard the results of the numerical simulation as our most accurate estimates, since we can estimate the errors. The results obtained from the SCSA, which are the best analytical estimate, are in acceptable agreement with simulations.
Finally there are two experimental measurements of critical exponents for the flat phase of crystalline membranes. The static structure factor of the red blood cell cytoskeleton (see Sect.1) has been measured by small-angle x-ray and light scattering, yielding a roughness exponent of $`\zeta =0.65(10)`$ . Freeze-fracture electron microscopy and static light scattering of the conformations of graphitic oxide sheets (Sect.1) revealed flat sheets with a fractal dimension $`d_H=2.15(6)`$. Both these values are in good agreement with the best analytic and numerical predictions, but the errors are still too large to discriminate between different analytic calculations.
The Poisson ratio of a crystalline membrane (measuring the transverse elongation due to a longitudinal stress ) is universal and within the SCSA approximation, which we regard as the more accurate analytical estimate, is given by
$$\sigma (D)=\frac{1}{D+1}\sigma (2)=1/3,$$
(29)
This result has been accurately checked in numerical simulations . Rather remarkably, it turns out to be negative. Materials having a negative Poisson ratio are called auxetic. This highlights potential applications of crystalline membranes to materials science since auxetic materials have a wide variety of potential applications as gaskets, seals etc.
Finally another critical regime of a flat membrane is achieved by subjecting the membrane to external tension . This allows a low temperature phase in which the membrane has a domain structure, with distinct domains corresponding to flat phases with different bulk orientations. This describes, physically, a buckled membrane whose equilibrium shape is no longer planar.
### 4.2 Self-avoiding
Self-avoidance is a necessary interaction in any realistic description of a crystalline membrane. It is introduced in the form of a delta-function repulsion in the full model Eq.(5). We have already analyzed the phantom case and explored in detail the distinct phases. The question before us now is the effect of self-avoidance on each of these phases.
The first phase we analyze is the flat phase. Since self-intersections are unlikely in this phase, it is intuitively clear that self-avoidance should be irrelevant. This may also be seen if one neglects the effects of the in-plane phonons. In the self-avoiding term for the flat phase we have
$`{\displaystyle \frac{b}{2}}{\displaystyle d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐲))}`$
$`=`$ $`{\displaystyle \frac{b}{2}}{\displaystyle d^D𝐱d^D𝐲\delta ^D(\zeta (𝐱𝐲)+𝐮(𝐱)𝐮(𝐲))\delta ^{dD}(h(𝐱)h(𝐲))}`$
$``$ $`{\displaystyle \frac{b}{2}}{\displaystyle d^D𝐱d^D𝐲\delta ^D(\zeta (𝐱𝐲))\delta ^{dD}(h(𝐱)h(𝐲))}=0,`$
as the trivial contribution where the membrane equals itself is eliminated by regularization. The previous argument receives additional support from numerical simulations in the flat phase, where it is found that self-intersections are extremely rare in the typical configurations appearing in those simulations. It seems clear that self-avoidance is most likely an irrelevant operator, in the RG sense, of the FLFP. Nevertheless, it would be very interesting if one could provide a more rigorous analytical proof for this statement. A rough argument can be made as follows. Shortly we will see that the Flory approximation for self-avoiding membranes predicts a fractal dimension $`d_H=2.5`$. For bulk dimension $`d`$ exceeding 2.5 therefore we expect self-avoidance to be irrelevant. A rigorous proof of this sort remains rather elusive, as it involves the incorporation of both self-avoidance and non-linear elasticity, and this remains a difficult open problem.
The addition of self-avoidance in the crumpled phase consists of adding the self-avoiding interaction to the free energy of Eq.(10)
$$F(\stackrel{}{r})=\frac{1}{2}d^D𝐱(_\alpha \stackrel{}{r}(𝐱))^2+\frac{b}{2}d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐲)),$$
(31)
which becomes the natural generalization of the Edwards’ model for polymers to $`D`$-dimensional objects. Standard power counting shows that the GFP of the crumpled phase is infra-red unstable to the self-avoiding perturbation for
$$\epsilon (D,d)2Dd\frac{2D}{2}>0,$$
(32)
which implies that self-avoidance is a relevant perturbation for $`D=2`$-objects at any embedding dimension $`d`$. The previous remarks make it apparent that it is possible to perform an $`\epsilon `$-expansion of the model . In Appendix F, we present the calculation of the $`\beta `$-function at lowest order in $`\epsilon `$ using the MOPE (Multi-local-operator-product-expansion) formalism . The MOPE formalism has the advantage that it is more easily generalizable to higher orders in $`\epsilon `$, and enables concrete proofs showing that the expansion may be carried out to all orders. At lowest order, the result for the $`\beta `$-function is
$`\beta _b(b_R)`$ $`=`$ $`\epsilon b_R+{\displaystyle \frac{(2D)^{1+\frac{d}{2}}}{(4\pi )^{\frac{d}{2}}}}\left({\displaystyle \frac{2\pi ^{\frac{D}{2}}}{\mathrm{\Gamma }(D/2)}}\right)^{2+\frac{d}{2}}\left[{\displaystyle \frac{\mathrm{\Gamma }(\frac{D}{2D})^2}{\mathrm{\Gamma }(\frac{2D}{2D})}}+{\displaystyle \frac{d}{2}}{\displaystyle \frac{(2D)^2}{2D}}\right]{\displaystyle \frac{b_R^2}{2}}`$ (33)
$``$ $`\epsilon b_R+a_1b_R^2.`$
The infra-red stable FP is given at lowest order in $`\epsilon `$ by $`b_R^{}=\frac{\epsilon }{a_1}`$, which clearly shows that the GFP of the crumpled phase is infra-red unstable in the presence of self-avoidance.
The preceding results are shown in Fig.17 and may be summarized as
* The flat phase of self-avoiding crystalline membranes is exactly the same as the flat phase of phantom crystalline tethered membranes.
* The crumpled phase of crystalline membranes is destabilized by the presence of any amount of self-avoidance.
The next issue to elucidate is whether this new SAFP describes a crumpled self-avoiding phase or a flat one and to give a more quantitative description of the critical exponents describing the universality class. Supposing that the SAFP is, in fact, flat we must understand its relation to the FLFP describing the physics of the flat phase and the putative phase transitions between these two.
#### 4.2.1 The nature of the SAFP
Let us study in more detail the model described in Eq.(31). The key issue is whether this model still admits a crumpled phase, and if so to determine the associated size exponent. On general grounds we expect that there is a critical dimension $`d_c`$, below which there is no crumpled phase (see Fig.18).
An estimate for the critical dimension may be obtained from a Flory approximation in which minimizes the free energy obtained by replacing both the elastic and self-avoiding terms with the radius of gyration raised to the power of the appropriate scaling dimensions. Within the Flory treatment a $`D`$-dimensional membrane is in a crumpled phase, with a size exponent given by
$$\nu =(D+2)/(d+2).$$
(34)
From this it follows that $`d_c=D`$ (see Fig.18). The Flory approximation, though very accurate for polymers ($`D=1`$), remains an uncontrolled approximation.
In contrast the $`\epsilon `$-expansion provides a systematic determination of the critical exponents. For the case of membranes, however, some extrapolation is required, as the upper critical dimension is infinite. This was done in , where it is shown that reconsidering the $`\epsilon `$-expansion as a double expansion in $`\epsilon `$ and $`D`$, critical quantities may be extrapolated for $`D=2`$-dimensional objects. At lowest order in $`\epsilon `$, the membrane is in a crumpled phase. The enormous task of calculating the next correction ($`\epsilon ^2`$) was successfully carried out in , employing more elaborate extrapolation methods than those in . Within this calculation, the $`d=3`$ membrane is still in a crumpled phase, but with a size exponent now closer to 1. It cannot be ruled out that the $`\epsilon `$-expansion, successfully carried out to all orders could give a flat phase $`\nu =1`$. In fact, the authors in present some arguments in favor of a scenario of this type, with a critical dimension $`d_c4`$.
Other approaches have been developed with different results. A Gaussian approximation was developed in . The size exponent of a self-avoiding membrane within this approach is
$$\nu =4/d,$$
(35)
and since one has $`\nu >1`$ for $`d4`$, one may conclude that the membrane is flat for $`dd_c=4`$. Since we cannot determine the accuracy of the Gaussian approximation this estimate must be viewed largely as interesting speculation. Slightly more elaborate arguments of this type yield an estimated critical dimension $`d_c=3`$.
Numerical simulations
We have seen that numerical simulations provide good support for analytic results in the case of phantom membranes. When self-avoidance is included, numerical simulations become invaluable, since analytic results are harder to come by. It is for this reason that we discuss them in greater detail than in previous sections.
A possible discretization of membranes with excluded volume effects consists of a network of $`N`$ particles arranged in a triangular array. Nearest neighbors interact with a potential
$$V_{NN}(\stackrel{}{r})=\{\begin{array}{cc}0& \text{for }|\stackrel{}{r}|<b\text{ }\\ \mathrm{}& \text{for }|\stackrel{}{r}|>b\end{array},$$
(36)
although some authors prefer a smoothened version, with the same general features. The quantity $`b`$ is of the order of a few lattice spacings. This is a lattice version of the elastic term in Eq.(31). The discretization of the self-avoidance is introduced as a repulsive hard sphere potential, now acting between any two atoms in the membrane, instead of only nearest neighbors. A hard sphere repulsive potential is, for example,
$$V_{Exc}(\stackrel{}{r})=\{\begin{array}{cc}\mathrm{}& \text{for }|\stackrel{}{r}|<\sigma \\ 0& \text{ for }|\stackrel{}{r}|>\sigma \text{ }\end{array},$$
(37)
where $`\sigma `$ is the range of the potential, and $`\sigma <b`$. Again, some smoothened versions, continuous at $`|\stackrel{}{r}|=\sigma `$, have also been considered. This model may be pictured as springs, defined by the nearest-neighbor potential Eq.(36), with excluded volume effects enforced by balls of radius $`\sigma `$ (Eq.(37)). This model represents a lattice discretization of Eq.(31).
Early simulations of this type of model provided a first estimate of the size exponent at $`d=3`$ fully compatible with the Flory estimate Eq.(34). The lattices examined were not very large, however, and subsequent simulations with larger volumes found that the $`d=3`$ membrane is actually flat. This result is even more remarkable if one recalls that there is no explicit bending rigidity.
The flat phase was a very surprising result, in some conflict with the insight provided from the analytical estimates discussed in the previous subsection. An explanation for it came from the observation that excluded volume effects induce bending rigidity, as depicted in Fig.19. The reason is that the excluded volume effects generate a non-zero expectation value for the bending rigidity, since the normals can be parallel, but not anti-parallel (see 19). This induced bending rigidity was estimated and found to be big enough to drive the self-avoiding membrane well within the flat phase of the phantom one. This means that this particular discretization of the model renders any potential SAFP inaccessible and the physics is described by the FLFP. In the structure function of the self-avoiding model is numerically computed and found to compare well with the analytical structure function for the flat phase of phantom crystalline membranes, including comparable roughness exponents.
The natural question then to ask is whether it is possible to reduce the bending rigidity sufficiently to produce a crumpled self-avoiding phase. Subsequent studies addressed this issue in various ways. The most natural way is obviously to reduce the range of the potential sufficiently that the induced bending rigidity is within the crumpled phase. This is the approach followed in . The flat phase was found to persist to very small values of $`\sigma `$, with eventual signs of a crumpled phase. This crumpled phase may essentially be due to the elimination at self-avoidance at sufficiently small $`\sigma `$. A more comprehensive study, in which the same limit is performed this time with an excluded volume potential which is a function of the internal distance along the lattice , concluded that for large membranes, inclusion of excluded volume effects, no matter how small, leads to flatness. A different approach to weakening the flat phase, bond dilution , found that the flat phase persists until the percolation critical point. In conclusion the bulk of accumulated evidence indicates that flatness is an intrinsic consequence of self-avoidance. If this is indeed correct the SAFP coincides with FLFP and this feature is an inherent consequence of self-avoidance, rather than an artifact of discretization.
Given the difficulties of finding a crumpled phase with a repulsive potential, simulations for larger values of the embedding space dimension $`d`$ have also been performed . These simulations show clear evidence that the membrane remains flat for $`d=3`$ and $`4`$ and undergoes a crumpling transition for $`d5`$, implying $`d_c4`$.
An alternative approach to incorporating excluded volume effects corresponds to discretize a surface with a triangular lattice and imposing the self-avoidance constraint by preventing the triangular plaquettes from inter-penetrating. This model has the advantage that is extremely flexible, since there is no restriction on the bending angle of adjacent plaquettes (triangles) and therefore no induced bending rigidity (see C).
The first simulations of the plaquette model found a size exponent in agreement with the Flory estimate Eq. 34. A subsequent simulation disproved this result, and found a size exponent $`\nu =0.87`$, higher than the Flory estimate, but below one. More recent results using larger lattices and more sophisticated algorithms seem to agree completely with the results obtained from the ball and spring models .
Further insight into the lack of a crumpled phase for self-avoiding crystalline membranes is found in the study of folding . This corresponds to the limit of infinite elastic constants studied by David and Guitter with the further approximation that the space of bending angles is discretized. One quickly discovers that the reflection symmetries of the allowed folding vertices forbid local folding (crumpling) of surfaces. There is therefore essentially no entropy for crumpling. There is, however, local unfolding and the resulting statistical mechanical models are non-trivial. The lack of local folding is the discrete equivalent of the long-range curvature-curvature interactions that stabilize the flat phase. The dual effect of the integrity of the surface (time-independent connectivity) and self-avoidance is so powerful that crumpling seems to be impossible in low embedding dimensions.
#### 4.2.2 Attractive potentials
Self-avoidance, as introduced in Eq.(37) is a totally repulsive force among monomers. There is the interesting possibility of allowing for attractive potentials also. This was pioneered in as a way to escape to the induced bending rigidity argument (see Fig.19), since an attractive potential would correspond to a negligible (or rather a negative) bending rigidity. Remarkably, in , a compact (more crumpled) self-avoiding phase was found, with fractal dimension close to 3.
This was further studied in , where it was found that with an attractive Van der Waals potential, the crystalline membrane underwent a sequence of folding transitions leading to a crumpled phase. In similar results were found, but instead of a sequence of folding transitions a crumpled phase was found with an additional compact (more crumpled phase) at even lower temperatures. Subsequent work gave some support to this scenario .
On the analytical side, the nature of the $`\mathrm{\Theta }`$-point for membranes and its relevance to the issue of attractive interactions has been addressed in .
We think that the study of a tether with an attractive potential remains an open question begging for new insights. A thorough understanding of the nature of the compact phases produced by attractive interactions would be of great value.
#### 4.2.3 The properties of the SAFP
The enormous efforts dedicated to study the SAFP have not resulted in a complete clarification of the overall scenario since the existing analytical tools do not provide a clear picture. Numerical results clearly provide the best insight. For the physically relevant case $`d=3`$, the most plausible situation is that there is no crumpled phase and that the flat phase is identical to the flat phase of the phantom model. For example, the roughness exponents $`\zeta _{SA}`$ from numerical simulations of self-avoidance at $`d=3`$ using ball-and-spring models and the roughness exponent at the FLFP, Eq.(25), compare extremely well
$$\zeta _{SA}=0.64(4),\zeta =0.64(2),$$
(38)
So the numerical evidence allows us to conjecture that the SAFP is exactly the same as the FLFP, and that the crumpled self-avoiding phase is absent in the presence of purely repulsive potentials (see Fig.20). This identification of fixed points enhances the significance of the FLFP treated earlier. It would be very helpful if analytical tools were developed to further substantiate this statement.
## 5 Anisotropic Membranes
An anisotropic membrane is a crystalline membrane having the property that the elastic or the bending rigidity properties in one distinguished direction are different from those in the $`D1`$ remaining directions. As for the isotropic case we keep the discussion general and describe the membrane by a $`d`$-dimensional $`\stackrel{}{r}(𝐱_{},y)`$, where now the $`D`$ dimensional coordinates are split into $`D1𝐱_{}`$ coordinates and the orthogonal distinguished direction $`y`$.
The construction of the Landau free energy follows the same steps as in the isotropic case. Imposing translational invariance, $`O(d)`$ rotations in the embedding space and $`O(D1)`$ rotations in internal space, the equivalent of Eq.(5) is now
$`F(\stackrel{}{r}(𝐱))`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }d^{D1}𝐱_{}dy[\kappa _{}(_{}^2\stackrel{}{r})^2+\kappa _y(_y^2\stackrel{}{r})^2`$ (39)
$`+\kappa _y_y^2\stackrel{}{r}_{}^2\stackrel{}{r}+t_{}(_\alpha ^{}\stackrel{}{r})^2+t_y(_y\stackrel{}{r})^2`$
$`+{\displaystyle \frac{u_{}}{2}}(_\alpha ^{}\stackrel{}{r}_\beta ^{}\stackrel{}{r})^2+{\displaystyle \frac{u_{yy}}{2}}(_y\stackrel{}{r}_y\stackrel{}{r})^2`$
$`+u_y(_\alpha ^{}\stackrel{}{r}_y\stackrel{}{r})^2+{\displaystyle \frac{v_{}}{2}}(_\alpha ^{}\stackrel{}{r}_\alpha ^{}\stackrel{}{r})^2`$
$`+v_y(_\alpha ^{}\stackrel{}{r})^2(_y\stackrel{}{r})^2]`$
$`+{\displaystyle \frac{b}{2}}{\displaystyle d^D𝐱d^D𝐱^{}\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐱^{}))}.`$
This model has eleven parameters, representing distinct physical interactions:
* $`\kappa _{},\kappa _y,\kappa _y`$ bending rigidity: the anisotropic versions of the isotropic bending rigidity splits into three distinct terms.
* $`t_{},t_y,u_{},u_{yy},v_{},v_y`$ elastic constants: there are six quantities describing the microscopic elastic properties of the anisotropic membrane.
* $`b`$, self-avoidance coupling: This particular term is identical to its isotropic counterpart.
Following the same steps as in the isotropic case, we split
$$\stackrel{}{r}(𝐱)=(\zeta _{}𝐱_{}+𝐮_{}(𝐱),\zeta _yy+u_y(𝐱),h(𝐱)),$$
(40)
with $`𝐮_{}`$ being the $`D1`$-dimensional phonon in-plane modes, $`u_y`$ the in-plane phonon mode in the distinguished direction $`y`$ and $`h`$ the $`dD`$ out-of-plane fluctuations. If $`\zeta _{}=\zeta _y=0`$, the membrane is in a crumpled phase and if both $`\zeta _{}0`$ and $`\zeta _y0`$ the membrane is in a flat phase very similar to the isotropic case (how similar will be discussed shortly). There is, however, the possibility that $`\zeta _{}=0`$ and $`\zeta _y0`$ or $`\zeta _{}0`$ and $`\zeta _y=0`$. This describes a completely new phase, in which the membrane is crumpled in some internal directions but flat in the remaining ones. A phase of this type is called a tubular phase and does not appear when studying isotropic membranes. In Fig.21 we show an intuitive visualization of a tubular phase along with the corresponding flat and crumpled phases of anisotropic membranes.
We will start by studying the phantom case. We show, using both analytical and numerical arguments, that the phase diagram contains a crumpled, tubular and flat phase. The crumpled and flat phases are equivalent to the isotropic ones, so anisotropy turns out to be an irrelevant interaction in those phases. The new physics is contained in the tubular phase, which we describe in detail, both with and without self-avoidance.
### 5.1 Phantom
#### 5.1.1 The Phase diagram
We first describe the mean field theory phase diagram and then the effect of fluctuations. There are two situations depending on the particular values of the function $`\mathrm{\Delta }`$, which depends on the elastic constants $`u_{},v_y,u_{yy}`$ and $`v_{}`$. Since the derivation is rather technical, we refer to Appendix G for the details.
* Case A ($`\mathrm{\Delta }>0`$): the mean field solution displays all possible phases. When $`t_y>0`$ and $`t_{}>0`$ the model is in a crumpled phase. Lowering the temperature, one of the $`t`$ couplings becomes negative, and we reach a tubular phase (either $``$ or $`y`$-tubule). A further reduction of the temperature eventually leads to a flat phase.
* Case B ($`\mathrm{\Delta }<0`$): in this case the flat phase disappears from the mean field solution. Lowering further the temperature leads to a continuous transition from the crumpled phase to a tubular phase. Tubular phases are the low temperature stable phases in this regime.
This mean field result is summarized in Fig.22.
Beyond mean field theory, the Ginsburg criterion applied to this model tells us that the phase diagram should be stable for physical membranes $`D=2`$ at any embedding dimension $`d`$, so the mean field scenario should give the right qualitative picture for the full model.
Numerical simulations have spectacularly confirmed this result. We have already shown in Fig.21 the results from the numerical simulation in , where it was shown that changing the temperature generates a sequence of transitions crumpled-to-tubular and tubular-to-flat, in total agreement with case A) in the mean field result illustrated in Fig.22.
We now turn to a more detailed study of both the crumpled and flat anisotropic phases. Since we have already studied crumpled and flat phases we just outline how those are modified when anisotropy is introduced.
#### 5.1.2 The Crumpled Anisotropic Phase
In this phase $`t_y>0`$ and $`t_{}>0`$, and the free energy Eq.(39) reduces for $`D2`$ to
$$F(\stackrel{}{r}(𝐱))=\frac{1}{2}d^{D1}𝐱_{}𝑑y\left[t_{}(_\alpha ^{}\stackrel{}{r})^2+t_y(_y\stackrel{}{r})^2\right]+\text{Irrelevant}.$$
(41)
By redefining the $`y`$ direction as $`y^{}=\frac{t_{}}{t_y}y`$ this reduces to Eq.(10), with $`tt_{}`$. We have proved that anisotropy is totally irrelevant in this particular phase.
#### 5.1.3 The Flat Phase
This phase becomes equivalent to the isotropic case as well. Intuitively, this may be obtained from the fact that if the membrane is flat, the intrinsic anisotropies are only apparent at short-distances, and therefore by analyzing the RG flow at larger and larger distances the membrane should become isotropic. This argument may be made slightly more precise .
### 5.2 The Tubular Phase
We now turn to the study of the novel tubular phase, both in the phantom case and with self-avoidance. Since the physically relevant case for membranes is $`D=2`$ the $`y`$-tubular and $``$-tubular phase are the same. So we concentrate on the properties of the $`y`$-tubular phase.
The key critical exponents characterizing the tubular phase are the size (or Flory) exponent $`\nu `$, giving the scaling of the tubular diameter $`R_g`$ with the extended ($`L_y`$) and transverse ($`L_{}`$) sizes of the membrane, and the roughness exponent $`\zeta `$ associated with the growth of height fluctuations $`h_{rms}`$ (see Fig.23):
$`R_g(L_{},L_y)`$ $``$ $`L_{}^\nu S_R(L_y/L_{}^z)`$ (42)
$`h_{rms}(L_{},L_y)`$ $``$ $`L_y^\zeta S_h(L_y/L_{}^z)`$
Here $`S_R`$ and $`S_h`$ are scaling functions and $`z`$ is the anisotropy exponent.
The general free energy described in Eq.(39) may be simplified considerably in a $`y`$-tubular phase. The analysis required is involved and we refer the interested reader to . We just quote the final result. It is
$`F(u,\stackrel{}{h})`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle }d^{D1}𝐱_{}dy[\kappa (_y^2\stackrel{}{h})^2+t(_\alpha \stackrel{}{h})^2`$ (43)
$`+g_{}(_\alpha u+_\alpha \stackrel{}{h}_y\stackrel{}{h})^2`$
$`+g_y(_yu+{\displaystyle \frac{1}{2}}(_y\stackrel{}{h})^2)^2]`$
$`+{\displaystyle \frac{b}{2}}{\displaystyle 𝑑yd^{D1}𝐱_{}d^{D1}𝐱_{}^{}\delta ^{d1}(\stackrel{}{h}(𝐱_{},y)\stackrel{}{h}(𝐱_{}^{},y))}.`$
Comparing with Eq.(39), this free energy does represent a simplification as the number of couplings has been reduced from eleven to five. Furthermore, the coupling $`g_{}`$ is irrelevant by standard power counting. The most natural assumption is to set it to zero. In that case the phase diagram one obtains is shown in Fig.24. Without self-avoidance $`b=0`$, the Gaussian Fixed Point (GFP) is unstable and the infra-red stable FP is the tubular phase FP (TPFP). Any amount of self-avoidance, however, leads to a new FP, the Self-avoiding Tubular FP (SAFP), which describes the large distance properties of self-avoiding tubules.
We just mention, though, that other authors advocate a different scenario . For sufficiently small embedding dimensions $`d`$, including the physical $`d=3`$ case, these authors suggest the existence of a new bending rigidity renormalized FP (BRFP), which is the infra-red FP describing the actual properties of self-avoiding tubules (see Fig. 25).
Here we follow the arguments presented in and consider the model defined by Eq.(43) with the $`g_{}`$-term as the model describing the large distance properties of tubules. One can prove then than there are some general scaling relations among the critical exponents. All three exponents may be expressed in terms of a single exponent
$`\zeta `$ $`=`$ $`{\displaystyle \frac{3}{2}}+{\displaystyle \frac{1D}{2z}}`$ (44)
$`\nu `$ $`=\zeta z.`$
Remarkably, the phantom case as described by Eq.(43) can be solved exactly. The result for the size exponent is
$$\nu _{Phantom}(D)=\frac{52D}{4},\nu _{Phantom}(2)=\frac{1}{4}$$
(45)
with the remaining exponents following from the scaling relations Eq.(44).
The self-avoiding case may be treated with techniques similar to those in isotropic case. The size exponent may be estimated within a Flory approach. The result is
$$\nu _{Flory}=\frac{D+1}{d+1}.$$
(46)
The Flory estimate is an uncontrolled approximation. Fortunately, a $`\epsilon `$-expansion, adapting the MOPE technique described for the self-avoiding isotropic case to the case of tubules, is also possible . The $`\beta `$-functions are computed and provide evidence for the phase diagram shown in Fig.24. Using rather involved extrapolation techniques, it is possible to obtain estimates for the size exponent, which are shown in Table 2. The rest of the exponents may be computed from the scaling relations.
## 6 Defects in membranes: The Crystalline-Fluid transition and Fluid membranes
A flat crystal melts into a liquid when the temperature is increased. This transition may be driven by the sequential liberation of defects, as predicted by the KTNHY theory. The KTNHY theory is schematically shown in Fig.26. With increasing temperature, a crystal melts first to an intermediate hexatic phase via a continuous transition, and finally goes to a conventional isotropic fluid phase via another continuous transition.
We will not review here either the KTNHY theory or the experimental evidence in its favor – . We just want to emphasize here that the KTNHY theory is in general agreement with existing experiments, although there are two main points worth keeping in mind when studying the more difficult case of fluctuating geometries. 1) The experimental evidence for the existence of the hexatic phase is not completely settled in those transitions which are continuous. 2) Some 2D crystals (like Xenon absorbed on graphite) melt to a fluid phase via a first order transition without any intermediate hexatic phase.
The straight-forward translation of the previous results to the tethered membrane would suggest a similar scenario. There would then be a crystalline to hexatic transition and a hexatic to fluid transition, as schematically depicted in Fig.26. Although the previous scenario is plausible, there are no solid experimental or theoretical results that establish it. From the theoretical point of view, for example, an important open problem is how to generalize the RG equations of the KTNHY theory to the case of fluctuating geometry. The situation looks even more uncertain experimentally, especially considering the elusive nature of the hexatic phase even in the case of flat monolayers.
In this review we will assume the general validity of the KTNHY scenario and we describe models of hexatic membranes, as well as fluid membranes. The study of the KTNHY theory in fluctuating geometries is a fascinating and challenging problem that deserves considerable effort. In this context, let us mention recent calculations of defects on frozen topographies , which show that even in the more simplified case when the geometry is frozen, defects proliferate in an attempt to screen out Gaussian curvature, even at zero temperature, and organize themselves in rather surprising and unexpected structures. These results hint at a rich set of possibilities for the more general case of fluctuating geometries.
### 6.1 Topological Defects
A crystal may have different distortions from its ground state. Thermal fluctuations are the simplest. Thermal fluctuations are small displacements from the ground state, and therefore one may bring back the system to its original positions by local moves without affecting the rest of the lattice. There are more subtle lattice distortions though, where the lattice cannot be taken to its ground state by local moves. These are the topological defects. There are different possible topological defects that may occur on a lattice, but we just need to consider dislocations and disclinations. Let us review the most salient features.
* Dislocation: represents the breaking of the translational holonomy. A path that would naturally close in a perfect lattice fails to close by a vector $`\stackrel{}{b}`$, the Burgers vector, as illustrated in Fig.27. In a flat monolayer, the energy is
$$E=\frac{K_0\stackrel{}{b}^2}{8\pi }\mathrm{ln}(\frac{R}{a}),$$
(47)
where $`K_0`$ is the Young Modulus. It diverges logarithmically with system size.
* Disclination: represents the breaking of the rotational holonomy. The bond angle around the point defect is a multiple of the natural bond angle in the ground state ($`\frac{\pi }{3}`$ in a triangular lattice), as illustrated in Fig.28 for a $`+`$ and $``$ disclination. The energy for a disclination in a flat monolayer is given by
$$E=\frac{K_0s^2}{32\pi }R^2.$$
(48)
Note the quadratic divergence of the energy with system size $`R`$.
Inspection of Fig.27 shows that a dislocation may be regarded as a tightly bound +,- disclination pair.
#### 6.1.1 Topological Defects in fluctuating geometries
The problem of understanding topological defects when the geometry is allowed to fluctuate was addressed in (see for a review). The important new feature is that the energy of a disclination defect may be lowered considerably if the membrane buckles out-of-the-plane. That is, the membrane trades elastic energy for bending rigidity. The energy for a buckled free disclination is given by
$$E=f(\kappa ,K_0,q_i)\mathrm{ln}(\frac{R}{a}),$$
(49)
where $`f`$ is some complicated function that may be evaluated numerically for given values of the parameters. It depends explicitly on $`q_i`$, which implies that the energies for positive and negative disclinations may be different, unlike the situation in flat space. The extraordinary reduction in energy from $`R^2`$ to $`\mathrm{ln}R`$ is possible because the buckled membrane creates positive Gaussian curvature for the plus-disclination and negative curvature for the negative-disclination. This is a very important physical feature of defects on curved surfaces. The defects attempt to screen out like-sign curvature, and analogously, like-sign defects may force the surface to create like-sign curvature in order to minimize the energy.
The reduction in energy for a dislocation defect is even more remarkable, since the energy of a dislocation becomes a constant, independent of the system size, provided the system is larger than a critical radius $`R_c`$. Again, by allowing the possibility of out-of-plane buckling, a spectacular reduction in energy is achieved (from $`R^2`$ to $`\mathrm{ln}R`$).
The study of other topological defects, e.g. vacancies, interstitials, and grain boundaries, may be carried out along the same lines. Since we are not going to make use of it, we refer the reader to the excellent review in .
#### 6.1.2 Melting and the hexatic phase
The celebrated Kosterlitz-Thouless argument shows that defects will necessarily drive a 2D crystal to melt. The entropy of a dislocation grows logarithmically with the system size, so for sufficiently high temperature, entropy will dominate over the dislocation energy (Eq.(47)) and the crystal will necessarily melt. If the same Kosterlitz-Thouless is applied now to a tethered membrane, the entropy is still growing logarithmically with the system size, while the energy becomes independent of the system size, as explained in the previous subsection, so any finite temperature drive the crystal to melt, and the low temperature phase of a tethered membrane will necessarily be a fluid phase, either hexatic if the KTNHY melting can be applied, or a conventional fluid if a first order transition takes place, or even some other more perverse possibility. This problem has also been investigated in numerical simulations , which provide some concrete evidence in favor of the KTNHY scenario, although the issue is far from being settled.
It is apparent from these arguments that a hexatic membrane is a very interesting and possibly experimentally relevant membrane to understand.
### 6.2 The Hexatic membrane
The hexatic membrane is a fluid membrane that, in contrast to a conventional fluid, preserves the orientational order of the original lattice (six-fold (hexatic) for a triangular lattice). The mathematical description of a fluid membrane is very different from those with crystalline order. Since the description cannot depend on internal degrees of freedom, the free energy must be invariant under reparametrizations of the internal coordinates (that is, should depend only on geometrical quantities, or in more mathematical terminology, must be diffeomorphism invariant). The corresponding free energy was proposed by Helfrich and it is given by
$$\frac{_H}{T}=\mu \sqrt{g}+\frac{\kappa }{2}𝑑𝐱\sqrt{g}\stackrel{}{H}^2,$$
(50)
where $`\mu `$ is the bare string tension, $`\kappa `$ the bending rigidity, $`g`$ is the determinant of the metric of the surface
$$g_{\mu \nu }(𝐱)=_\mu \stackrel{}{r}(𝐱)_\nu \stackrel{}{r}(𝐱)$$
(51)
and $`\stackrel{}{H}`$ is the mean Gaussian curvature of the surface. For a good description of the differential geometry relevant to the study of membranes we refer to A hexatic membrane has an additional degree of freedom, the bond angle, which is introduced as a field on the surface $`\theta `$. The hexatic free energy is obtained from adjoining to the fluid case of Eq.(50), the additional energy of the bond angle
$$_{hex}/T=\frac{K_A}{2}𝑑𝐱\sqrt{g}g^{\mu \nu }(_\mu \theta +\mathrm{\Omega }_{sing}\mathrm{\Omega }_\mu ^L)(_\nu \theta +\mathrm{\Omega }_{sing}\mathrm{\Omega }_\nu ^L)$$
(52)
where $`K_A`$ is called the hexatic stiffness, and $`\mathrm{\Omega }_\mu `$ is the connection two form of the metric, which may be related to the Gaussian curvature of the surface by
$$K(𝐱)=\frac{1}{\sqrt{g}}ϵ^{\mu \nu }_\mu \mathrm{\Omega }_\nu $$
(53)
and $`\mathrm{\Omega }_{sing}`$ is similarly related to the topological defect density
$`s(𝐱)`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{g}}}ϵ^{\mu \nu }_\mu \mathrm{\Omega }_{sing\nu }`$
$`s(𝐱)`$ $`=`$ $`{\displaystyle \frac{\pi }{3}}{\displaystyle \frac{1}{\sqrt{g}}}{\displaystyle \underset{i=1}{\overset{N}{}}}q_i\delta (𝐱,𝐱_i).`$ (54)
From general theorems on differential geometry one has the relation
$$\sqrt{g}s(𝐱)=4\pi \underset{i=1}{\overset{i}{}}q_i=2\chi ,$$
(55)
where $`\chi `$ is the Euler characteristic of the surface.
Therefore the total free energy for the hexatic membrane is given by
$`/T`$ $`=`$ $`\mu {\displaystyle \sqrt{g}}+{\displaystyle \frac{\kappa }{2}}{\displaystyle 𝑑𝐱\sqrt{g}\stackrel{}{H}^2}+`$
$`+`$ $`{\displaystyle \frac{K_A}{2}}{\displaystyle 𝑑𝐱\sqrt{g}g^{\mu \nu }(_\mu \theta +\mathrm{\Omega }_{sing}\mathrm{\Omega }_\mu ^L)(_\nu \theta +\mathrm{\Omega }_{sing}\mathrm{\Omega }_\nu ^L)}.`$
The partition function is therefore
$`𝒵(\beta )`$ $`=`$ $`{\displaystyle \underset{N_+,N_{}}{}}{\displaystyle \frac{\delta _{N_+N_{},2\chi }}{N_+!N_{}!}}y^{N_++N_{}}\times `$
$`{\displaystyle D[\stackrel{}{r}]D[\theta ]\underset{\mu =1}{\overset{N_+}{}}d𝐱_\mu ^+\sqrt{g}\underset{\nu =1}{\overset{N_{}}{}}d𝐱_\nu ^{}\sqrt{g}e^{(\stackrel{}{r}(𝐱),\theta (𝐱))/T}},`$
where $`y`$ is the fugacity of the disclination density. The partition function includes a discrete sum over allowed topological defects, those satisfying the topological constraint Eq.(55), and a path integral over embeddings $`\stackrel{}{r}`$ and bond angles $`\theta `$. The previous model remains quite intractable since the sum over defects interaction is very difficult to deal with.
Fortunately, the limit of very low fugacity $`y0`$ is analytically tractable as was shown in the beautiful paper . The RG functions can be computed within a combined large $`d`$ and large bending rigidity expansion. The $`\beta `$ functions in that limit is given by
$`\beta (\alpha )`$ $`=`$ $`{\displaystyle \frac{1}{4\pi K_A}}\left(D\alpha ^2+{\displaystyle \frac{3}{4}}\alpha ^3+𝒪(1/K_A^2)\right),`$
$`\beta (K_A)`$ $`=`$ $`0`$ (58)
where $`\alpha =1/\kappa `$. The physics of hexatic membranes in the limit of very low fugacity is very rich and show a line of fixed points parametrized by the hexatic stiffness $`K_A`$. The normal-normal correlation function, for example, reads
$$\stackrel{}{n}(𝐫)\stackrel{}{n}(\mathrm{𝟎})|𝐱|^\eta ,$$
(59)
with $`\eta =\frac{2}{3\pi }d(d2)\frac{k_BT}{K_A}`$. The FPs of Eq.(6.2) describe a new crinkled phase, more rigid than a crumpled phase but more crumpled than a flat one. The Hausdorff dimension at the crinkled phase is given by
$$d_H=2+\frac{d(d2)}{3\pi }\frac{k_BT}{K_A}.$$
(60)
From the RG point of view, the properties of these crinkled phases are really interesting, since they involve a line of Fixed Points which are inequivalent in the sense that the associated critical exponents depend continuously on $`K_A`$, a situation reminiscent of the $`XY`$-model. In the phase diagram is discussed, and the authors propose the scenario depicted in Fig.29. How these scenarios are modified when the fugacity is considered is not well established and we refer the reader to the original papers .
The shape fluctuations of hexatic vesicles, for large defect core energies, have also been investigated .
## 7 The Fluid Phase
The study of fluid membranes is a broad subject, currently under intense experimental and theoretical work. The Hamiltonian is given by
$$/T=\mu \sqrt{g}+\frac{\kappa }{2}𝑑𝐱\sqrt{g}\stackrel{}{H}^2+\frac{\widehat{\kappa }}{2}\sqrt{g}K,$$
(61)
This corresponds to the Helfrich hamiltonian together with a term that allows for topology changing interactions. For fixed topology it is well-known that the one loop beta function for the inverse-bending rigidity has a fixed point only at $`\kappa =0`$, which corresponds to the bending rigidity being irrelevant at large length scales. The RG flow of the bending rigidity is given by
$$\kappa (l)=\kappa _o\frac{3T}{4\pi }\mathrm{ln}(l/a),$$
(62)
where a is a microscopic cutoff length. The fluid membrane is therefore crumpled, for arbitrary microscopic bending rigidity $`\kappa _0`$, at length scales beyond a persistence length which grows exponentially with $`\kappa _0`$. For a fluid membrane out-of-plane fluctuations cost no elastic energy (the membrane flows internally to accommodate the deformation) and the bending rigidity is therefore softened by thermal undulations at all length scales, rather than stiffening at long length scales as in the crystalline membrane.
So far we have assumed an infinite membrane, which is not always a realistic assumption. A thickness may be taken into account via a spontaneous extrinsic curvature $`\stackrel{}{H}_0`$. The model described by Eq. 61 gets replaced then by
$$/T=\mu \sqrt{g}+\frac{\kappa }{2}𝑑𝐱\sqrt{g}(\stackrel{}{H}\stackrel{}{H}_0)^2+\frac{\widehat{\kappa }}{2}\sqrt{g}K.$$
(63)
Further effects of a finite membrane size for spherical topology have been discussed in . The phase diagram of fluid membranes when topology change is allowed is fascinating and not completely understood. A complete description of these phases goes beyond the scope of this review – we refer the reader to and references therein.
## 8 Conclusions
In this review we have described the distinct universality classes of membranes with particular emphasis on crystalline membranes. In each case we discussed and summarized the key models describing the interactions of the relevant large distance degrees of freedom (at the micron scale). The body of the review emphasizes qualitative and descriptive aspects of the physics with technical details presented in extensive appendices. We hope that the concreteness of these calculations gives a complete picture of how to extract relevant physical information from these membrane models.
We have also shown that the phase diagram of the phantom crystalline membrane class is theoretically very well understood both by analytical and numerical treatments. To complete the picture it would be extremely valuable to find experimental realizations for this particular system. An exciting possibility is a system of cross-linked DNA chains together with restriction enzymes that catalyze cutting and rejoining . The difficult chemistry involved in these experiments is not yet under control, but we hope that these technical problems will be overcome in the near future.
There are several experimental realizations of self-avoiding polymerized membranes discussed in the text. The experimental results compare very well with the theoretical estimates from numerical simulations. As a future theoretical challenge, analytical tools need to be sharpened since they fail to provide a clear and unified picture of the phase diagram. On the experimental side, there are promising experimental realizations of tethered membranes which will allow more precise results than those presently available. Among them there is the possibility of very well controlled synthesis of DNA networks to form physical realizations of tethered membranes.
The case of anisotropic polymerized membranes has also been described in some detail. The phase diagram contains a new tubular phase which may be realized in nature. There is some controversy about the precise phase diagram of the model, but definite predictions for the critical exponents and other quantities exist. Anisotropic membranes are also experimentally relevant. They may be created in the laboratory by polymerizing a fluid membrane in the presence of an external electric field.
Probably the most challenging problem, both theoretically and experimental, is a complete study of the role of defects in polymerized membranes. There are a large number of unanswered questions, which include the existence of hexatic phases, the properties of defects on curved surfaces and its relevance to the possible existence of more complex phases. This problem is now under intense experimental investigation. In this context, let us mention very recent experiments on Langmuir films in a presumed hexatic phase . The coalescence of air bubbles with the film exhibit several puzzling features which are strongly related to the curvature of the bubble.
Crystalline membranes also provide important insight into the fluid case, since any crystalline membrane eventually becomes fluid at high temperature. The physics of fluid membranes is a complex and fascinating subject in itself which goes beyond the scope of this review. We highlighted some relevant experimental realizations and gave a quick overview of the existing theoretical models. Due to its relevance in many physical and biological systems and its potential applications in material science, the experimental and theoretical understanding of fluid membranes is, and will continue to be, one of the most active areas in soft condensed matter physics.
We have not been able in this review, simply for lack of time, to address the important topic of the role of disorder. We hope to cover this in a separate article.
We hope that this review will be useful for physicists trying to get a thorough understanding of the fascinating field of membranes. We think it is a subject with significant prospects for new and exciting developments.
Note: The interested reader may also find additional material in a forthcoming review by Wiese , of which we have seen only the table of contents.
Acknowledgements
This work was supported by the U.S. Department of Energy under contract No. DE-FG02-85ER40237. We would like to thank Dan Branton and Cyrus Safinya for providing us images from their laboratories and Paula Herrera-Siklódy for assistance with the figures. MJB would like to thank Riccardo Capovilla, Chris Stephens, Denjoe O’Connor and the other organizers of RG2000 for the opportunity to attend a wonderful meeting in Taxco, Mexico.
## Appendix A Useful integrals in dimensional regularization
In performing the $`\epsilon `$-expansion, we will be considering integrals of the form
$$I_{\alpha _1,\mathrm{},\alpha _n}(a,b)(\stackrel{}{p})=d^D\widehat{q}\frac{q_{\alpha _1}\mathrm{}q_{\alpha _n}}{(\stackrel{}{p}+\stackrel{}{q})^{2a}\stackrel{}{q}^{2b}},$$
(64)
These integrals may be computed exactly for general $`D,a,b`$ and $`\alpha =1,\mathrm{},N`$. The result will be published elsewhere. We will content ourselves by quoting what we need, the poles in $`\epsilon `$, for the integrals that appear in the diagrammatic calculations. We just quote the results
$$I_{\alpha _1\alpha _2}(2,2)=\frac{1}{8\pi ^2p^4}p_{\alpha _1}p_{\alpha _2}\frac{1}{\epsilon }$$
(65)
$$I_{\alpha _1\alpha _2\alpha _3}(2,2)=\frac{1}{8\pi ^2p^4}p_{\alpha _1}p_{\alpha _2}p_{\alpha _3}\frac{1}{\epsilon }$$
(66)
$`I_{\alpha _1\alpha _2\alpha _3\alpha _4}(2,2)={\displaystyle \frac{1}{8\pi ^2p^4}}(p_{\alpha _1}p_{\alpha _2}p_{\alpha _3}p_{\alpha _4}`$
$`{\displaystyle \frac{p^4}{24}}(\delta _{\alpha _1\alpha _2}\delta _{\alpha _3\alpha _4}+\delta _{\alpha _1\alpha _3}\delta _{\alpha _2\alpha _4}+\delta _{\alpha _1\alpha _4}\delta _{\alpha _2\alpha _3})){\displaystyle \frac{1}{\epsilon }}`$ (67)
$`I_{\alpha _1\alpha _2\alpha _3}(2,1)`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2p^2}}({\displaystyle \frac{p^2}{6}}(p_{\alpha _1}\delta _{\alpha _2\alpha _3}+p_{\alpha _2}\delta _{\alpha _1\alpha _3}+p_{\alpha _3}\delta _{\alpha _2\alpha _1})`$ (68)
$`p_{\alpha _1}p_{\alpha _2}p_{\alpha _3}){\displaystyle \frac{1}{\epsilon }}`$
## Appendix B Some practical identities for RG quantities
The beta functions defined in Eq. 3 may be re-expressed as
$$\left(\begin{array}{cc}\beta _u(u_R,v_R)& \\ \beta _v(u_R,v_R)& \end{array}\right)=\epsilon \left(\begin{array}{cc}\frac{\mathrm{ln}u}{u_R}& \frac{\mathrm{ln}u}{v_R}\\ \frac{\mathrm{ln}v}{u_R}& \frac{\mathrm{ln}v}{v_R}\end{array}\right)^1\left(\begin{array}{cc}1& \\ 1& \end{array}\right)$$
(69)
The previous expression may be further simplified noticing
$$A=\left(\begin{array}{cc}\frac{\mathrm{ln}u}{u_R}& \frac{\mathrm{ln}u}{v_R}\\ \frac{\mathrm{ln}v}{u_R}& \frac{\mathrm{ln}v}{v_R}\end{array}\right)=\left(\begin{array}{cc}\frac{1}{u_R}& 0\\ 0& \frac{1}{v_R}\end{array}\right)+D,$$
(70)
so that
$`A^1`$ $`=`$ $`\left(1+\left(\begin{array}{cc}u_R& 0\\ 0& v_R\end{array}\right)D\right)^1\left(\begin{array}{cc}u_R& 0\\ 0& v_R\end{array}\right)`$ (75)
$`=`$ $`\left(\begin{array}{cc}u_R& 0\\ 0& v_R\end{array}\right)\left(\begin{array}{cc}u_R& 0\\ 0& v_R\end{array}\right)D\left(\begin{array}{cc}u_R& 0\\ 0& v_R\end{array}\right)+\mathrm{}`$ (82)
where the last result follows from Taylor-expanding. These formulas easily allow to compute the corresponding $`\beta `$-functions. If
$`u`$ $`=`$ $`M^\epsilon \left[u_R+{\displaystyle \frac{1}{\epsilon }}(a_{11}u_R^2+a_{12}u_Rv_R+a_{13}v_R^2)\right]`$ (83)
$`v`$ $`=`$ $`M^\epsilon \left[v_R+{\displaystyle \frac{1}{\epsilon }}(a_{21}u_R^2+a_{22}u_Rv_R+a_{23}v_R^2)\right],`$
from Eq. 75 and Eq. 69 we easily derive the leading two orders in the couplings
$`\beta _u(u_R,v_R)`$ $`=`$ $`\epsilon u_R+a_{11}u_R^2+a_{12}u_Rv_R+a_{13}v_R^2`$ (84)
$`\beta _v(u_R,v_R)`$ $`=`$ $`\epsilon v_R+a_{21}u_R^2+a_{22}u_Rv_R+a_{23}v_R^2.`$
The formula for $`\gamma `$ in Eq. 3 may also be given a more practical expression. It is given by
$$\gamma =(\beta _u\frac{}{u_R}+\beta _v\frac{}{v_R})\mathrm{ln}Z_\varphi ,$$
(85)
Those are the formulas we need in the calculations we present in this review.
## Appendix C Discretized Model for tethered membranes
In this appendix we present appropriate discretized models for numerical simulation of tethered membranes. The surface is discretized by a triangular lattice defined by its vertices $`\{\stackrel{}{r}\}_{a=1,\mathrm{}}`$, with a corresponding discretized version of the Landau elastic term Eq. 20 given by
$$F_s=\frac{\beta }{2}\underset{ab}{}(|\stackrel{}{r}_a\stackrel{}{r}_b|1)^2,$$
(86)
where $`a,b`$ are nearest-neighbor vertices. If we write $`\stackrel{}{r}_a=𝐱_a+𝐮_a`$ with $`𝐱_a`$ defining the vertices of a perfectly regular triangular lattice and $`𝐮`$ the small perturbations around it, one gets
$$|\stackrel{}{r}_a\stackrel{}{r}_b|=1+u_{\alpha \beta }x^\alpha x^\beta +\mathrm{},$$
(87)
with $`u_{\alpha \beta }`$ being a discretized strain tensor and we we have neglected higher order terms. Plugging the previous expression into Eq. 86 and passing from the discrete to the continuum language we obtain
$$F_s=\frac{\sqrt{3}}{8}\beta d^2𝐱(2u_{\alpha \beta }^2+u_{\alpha \alpha }^2)$$
(88)
which is the elastic part of the free energy Eq. 20 with $`\lambda =\mu =\frac{\sqrt{3}}{4}\beta `$.
The bending rigidity term is written in the continuum as
$$S_{ext}=d^2𝐱\sqrt{g}K_{\alpha \beta }^\mu K_\mu ^{\alpha \beta }=d^2𝐮\sqrt{g}g^{\alpha \beta }_\alpha \stackrel{}{n}_\beta \stackrel{}{n}$$
(89)
where $`\stackrel{}{n}`$ is the normal to the surface and $``$ is the covariant derivative (see for a detailed description of these geometrical quantities). We discretize the normals form the the previous equation by
$$d^2𝐮\sqrt{g}g^{\alpha \beta }_\alpha \stackrel{}{n}_\beta \stackrel{}{n}\underset{ab}{}(\stackrel{}{n}_a\stackrel{}{n}_b)^2=2\underset{ab}{}(1\stackrel{}{n}_a\stackrel{}{n}_b)$$
(90)
The two terms Eq. 86 and Eq. 90 provide a suitable discretized model for a tethered membranes. However, in actual simulations, the even more simplified discretization
$$F=\underset{a,b}{}(\stackrel{}{r}_a\stackrel{}{r}_b)^2+\kappa \underset{i,j}{}(1\stackrel{}{n}_i\stackrel{}{n}_j)$$
(91)
is preferred since it is simpler and describes the same universality class (see for a discussion). Anisotropy may be introduced in this model by ascribing distinct bending rigidities to bending across links in different intrinsic directions .
Self-avoidance can be introduced in this model by imposing that the triangles that define the discretized surface cannot self-intersect. There are other possible discretizations of self-avoidance that we discus in sect. 4.2.
In order to numerically simulate the model Eq. 91 different algorithms have been used. A detailed comparison of the performance of each algorithm may be found in .
## Appendix D The crumpling Transition
The Free energy is given by Eq. 13
$$F(\stackrel{}{r})=d^D𝐱\left[\frac{1}{2}(_\alpha ^2\stackrel{}{r})^2+u(_\alpha \stackrel{}{r}_\beta \stackrel{}{r}\frac{\delta _{\alpha \beta }}{D}(_{\alpha \stackrel{}{r}})^2)^2+v(_\alpha \stackrel{}{r}^\alpha \stackrel{}{r})^2\right],$$
(92)
where the dependence on $`\kappa `$ is trivially scaled out. The Feynman rules for the model are given in Fig.30.
We need three renormalized constants, namely $`Z`$, $`Z_u`$ and $`Z_v`$ in order to renormalize the theory. We define the renormalized quantities by
$`\stackrel{}{r}`$ $`=`$ $`Z^{1/2}\stackrel{}{r}`$ (93)
$`u_R=M^\epsilon Z^2Z_u^1u`$ $`,`$ $`v_R=M^\epsilon Z^2Z_v^1v.`$
Then Eq. 92 becomes
$`F(\stackrel{}{r})`$ $`=`$ $`{\displaystyle }d^D𝐱[{\displaystyle \frac{Z}{2}}(_\alpha ^2\stackrel{}{r}_R)^2+M^\epsilon Z_uu(_\alpha \stackrel{}{r}_R_\beta \stackrel{}{r}_R{\displaystyle \frac{\delta _{\alpha \beta }}{D}}(_\alpha \stackrel{}{r}_R)^2)^2+`$ (94)
$`+`$ $`M^\epsilon Z_vv(_\alpha \stackrel{}{r}_R^\alpha \stackrel{}{r}_R)^2],`$
In order to compute the renormalized couplings, one must compute all relevant diagrams at one loop. Those are depicted in fig. 31. Within dimensional regularization, diagrams (1a) and (1b) are zero, which in turns imply that the renormalized constant is $`Z=1`$ at leading order in $`\epsilon `$, similarly as in linear $`\sigma `$ models.
Using the integrals in dimensional regularization (see Sect. A) Diagram (2a) gives the result
$$\frac{d}{8\pi ^2}\frac{1}{\epsilon }\delta ^{i_1i_2}\delta ^{j_1j_2}\left\{\frac{u^2}{24}(\stackrel{}{p}_1\stackrel{}{p}_3\stackrel{}{p}_2\stackrel{}{p}_4+\stackrel{}{p}_1\stackrel{}{p}_4\stackrel{}{p}_2\stackrel{}{p}_3)+(v^2\frac{u^2}{48})\stackrel{}{p}_1\stackrel{}{p}_2\stackrel{}{p}_3\stackrel{}{p}_4\right\}$$
(95)
And diagram (2b) and (2c) may be computed at once, since the result of (2c) is just (2b) after interchanging $`\stackrel{}{p}_3\stackrel{}{p}_4`$, so the total result (2a)+(2b) is
$`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \frac{1}{\epsilon }}\delta ^{i_1i_2}\delta ^{j_1j_2}\{({\displaystyle \frac{61}{96}}u^2+{\displaystyle \frac{7}{12}}uv+{\displaystyle \frac{v^2}{6}})(\stackrel{}{p}_1\stackrel{}{p}_3\stackrel{}{p}_2\stackrel{}{p}_4+\stackrel{}{p}_1\stackrel{}{p}_4\stackrel{}{p}_2\stackrel{}{p}_3)`$
$`+({\displaystyle \frac{v^2}{6}}+{\displaystyle \frac{uv}{12}}+{\displaystyle \frac{u^2}{96}})\stackrel{}{p}_1\stackrel{}{p}_2\stackrel{}{p}_3\stackrel{}{p}_4\}`$ (96)
And the result for (2d) and (2e) is just identical, so the total result (2d)+(2e) is
$`{\displaystyle \frac{1}{16\pi ^2}}{\displaystyle \frac{1}{\epsilon }}\delta ^{i_1i_2}\delta ^{j_1j_2}\{({\displaystyle \frac{u^2}{24}}+{\displaystyle \frac{1}{6}}uv)(\stackrel{}{p}_1\stackrel{}{p}_3\stackrel{}{p}_2\stackrel{}{p}_4+\stackrel{}{p}_1\stackrel{}{p}_4\stackrel{}{p}_2\stackrel{}{p}_3)`$
$`+(v^2+{\displaystyle \frac{13}{6}}uv{\displaystyle \frac{u^2}{48}})\stackrel{}{p}_1\stackrel{}{p}_2\stackrel{}{p}_3\stackrel{}{p}_4\}`$ (97)
Adding up all these contributions taking into account the different combinatorial factors (4 the first contribution, 8 the last two ones) and recalling $`Z=1`$, we get
$`u`$ $`=`$ $`M^\epsilon \left[u_R+{\displaystyle \frac{1}{\epsilon }}{\displaystyle \frac{1}{8\pi ^2}}\left(({\displaystyle \frac{d}{3}}+{\displaystyle \frac{65}{12}})u_R^2+6u_Rv_R+{\displaystyle \frac{4}{3}}v_r^2\right)\right]`$
$`v`$ $`=`$ $`M^\epsilon \left[v_R+{\displaystyle \frac{1}{\epsilon }}{\displaystyle \frac{1}{8\pi ^2}}\left({\displaystyle \frac{21}{16}}u_R+{\displaystyle \frac{21}{2}}u_Rv_R+(4d+5)v_R^2\right)\right].`$ (98)
The resultant $`\beta `$-functions are then readily obtained by applying Eq.(84).
## Appendix E The Flat Phase
The free energy is given in Eq. 20,and it is given by
$$F(𝐮,h)=d^D𝐱\left[\frac{\widehat{\kappa }}{2}(_\alpha _\beta h)^2+\mu u_{\alpha \beta }u^{\alpha \beta }+\frac{\lambda }{2}(u_\alpha ^\alpha )^2\right].$$
(99)
The Feynman rules are shown in fig. 32, it is apparent that the in-plane phonons couple different from the out-of-plane, which play the role of Goldstone bosons.
We apply standard field theory techniques to obtain the RG-quantities. Using the Ward identities, the theory can be renormalized using only three renormalization constants $`Z`$, $`Z_\mu `$ and $`Z_\lambda `$, corresponding to the wave function and the two coupling renormalizations. Renormalized quantities read
$`h_R=Z^{1/2}h`$ $`,`$ $`𝐮=Z^1𝐮`$ (100)
$`\mu _R=M^\epsilon Z^2Z_\mu ^1\mu `$ $`,`$ $`\lambda _R=M^\epsilon Z^2Z_\lambda ^1\lambda ,`$
Then Eq. 99 becomes
$$F(𝐮,h)=d^D𝐱\left[Z(_\alpha _\beta h_R)^2+2M^\epsilon Z_\mu \mu _Ru_{R\alpha \beta }u_R^{\alpha \beta }+M^\epsilon Z_\lambda \lambda (u_{R\alpha }^\alpha )^2\right].$$
(101)
We now compute the renormalized quantities from the leading divergences appearing in the Feynman diagrams. The diagrams to consider are given in Fig.33. These can be computed using the integrals given in Sect.A.
The result of diagram (1a) is given by
$$\frac{1}{\epsilon }\frac{d_c}{6\pi ^2}\left[\mu ^2(\delta _{\alpha \beta }\frac{p_\alpha p_\beta }{p^2})+3(\mu ^2+2\mu \lambda +2\lambda ^2)\frac{p_\alpha p_\beta }{p^2}\right]p^2.$$
(102)
Diagrams (2b) and (2c) are identically zero, so (2a) is the only additional diagram to be computed. The result is
$$\frac{1}{\epsilon }\frac{\delta ^{ij}}{8\pi ^2}\frac{\mu (\mu +\lambda )}{2\mu +\lambda }10(p^2)^2$$
(103)
from Eq. 103 and the definitions in Eq. 100
$$Z=1\frac{1}{\epsilon }\frac{10}{8\pi ^2}\frac{\mu _R(\mu _R+\lambda _R)}{2\mu _R+\lambda _R}.$$
(104)
Using the previous result in the diagrams (1a) whose result is in Eq. 102 we obtain
$`Z_\mu `$ $`=`$ $`1+{\displaystyle \frac{1}{\epsilon }}{\displaystyle \frac{d_c}{24\pi ^2}}\mu _R`$ (105)
$`Z_\lambda `$ $`=`$ $`1+{\displaystyle \frac{1}{\epsilon }}{\displaystyle \frac{d_c}{24\pi ^2}}(\mu _R^2+6\mu _R\lambda _R+6\lambda _R^2)/\lambda _R`$
and we deduce the renormalized couplings
$`\mu `$ $`=`$ $`M^\epsilon \left[\mu _R+{\displaystyle \frac{1}{\epsilon }}\left({\displaystyle \frac{10}{4\pi ^2}}{\displaystyle \frac{(\mu _R+\lambda _R)}{2\mu _R+\lambda _R}}+{\displaystyle \frac{d_c}{24\pi ^2}}\right)\mu _R^2\right]`$ (106)
$`\lambda `$ $`=`$ $`M^\epsilon \left[\lambda _R+{\displaystyle \frac{1}{\epsilon }}\left({\displaystyle \frac{10}{4\pi ^2}}{\displaystyle \frac{(\mu _R+\lambda _R)}{2\mu _R+\lambda _R}}\mu _R\lambda _R+{\displaystyle \frac{d_c}{24\pi ^2}}(\mu _R^2+6\mu _R\lambda _R+6\lambda _R^2)\right)\right]`$
from which the $`\beta `$-functions trivially follow with the aid of Eq.(84).
## Appendix F The Self-avoiding phase
The model has been introduced in Eq. 31 and is given by
$$F(\stackrel{}{r})=\frac{1}{2}d^D𝐱(_\alpha \stackrel{}{r}(𝐱))^2+\frac{b}{2}d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐲)),$$
(107)
We follow the usual strategy of defining the renormalized quantities by
$`\stackrel{}{r}`$ $`=`$ $`Z^{1/2}\stackrel{}{r}_R`$ (108)
$`b`$ $`=`$ $`M^\epsilon Z_bZ^{d/2}b_R,`$
and the renormalized Free energy by
$$F(\stackrel{}{r})=\frac{Z}{2}d^D𝐱(_\alpha \stackrel{}{r}_R(𝐱))^2+M^\epsilon Z_b\frac{b_R}{2}d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}_R(𝐱)\stackrel{}{r}_R(𝐲)).$$
(109)
The $`\delta `$-function being non-local adds some technical difficulties to the calculation of the renormalized constants $`Z`$ and $`Z_b`$. There are different approaches available but we will follow the MOPE (Multilocal-Operator-Product-Expansion), which we will just explain in a very simplified version. A rigorous description of the method may be found in the literature.
The idea is to expand the $`\delta `$-function term in Eq. 109
$$e^{F(\stackrel{}{r})}=e^{\frac{Z}{2}{\scriptscriptstyle d^D𝐱(_\alpha \stackrel{}{r}_R(𝐱))^2}}\times \underset{n=0}{\overset{\mathrm{}}{}}\left(M^\epsilon Z_b\frac{b_R}{2}d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}_R(𝐱)\stackrel{}{r}_R(𝐲))\right)^n,$$
(110)
with this trick, the delta-function term may be treated as expectation values of a Gaussian free theory. This observation alone allows to isolate the poles in $`\epsilon `$. We write the identity
$$e^{i\stackrel{}{k}(\stackrel{}{r}(𝐱_1)\stackrel{}{r}(𝐱_2))}=:e^{i\stackrel{}{k}(\stackrel{}{r}(𝐱_1)\stackrel{}{r}(𝐱_2))}:e^{k^2G(x_1x_2)},$$
(111)
where $`G(x)`$ is the two point correlator
$$G(𝐱)=\stackrel{}{r}(𝐱)\stackrel{}{r}(0)=\frac{|𝐱|^{2D}}{(2D)S_D},$$
(112)
with $`S_D`$ being the volume of the $`D`$-dimensional sphere. The symbol $`::`$ stands for normal ordering. A normal ordered operator is non-singular at short-distances, so it may be Taylor-expanded
$`e^{i\stackrel{}{k}(\stackrel{}{r}(𝐱_1)\stackrel{}{r}(𝐱_2))}`$ $`=`$ $`(1+i(𝐱_\mathrm{𝟏}𝐱_\mathrm{𝟐})^\alpha (\stackrel{}{k}_\alpha \stackrel{}{r})`$
$`{\displaystyle \frac{1}{2}}(𝐱_\mathrm{𝟏}𝐱_\mathrm{𝟐})^\alpha (𝐱_\mathrm{𝟏}𝐱_\mathrm{𝟐})^\beta (\stackrel{}{k}_\alpha \stackrel{}{r})(\stackrel{}{k}_\beta \stackrel{}{r})+\mathrm{})e^{k^2G(x_1x_2)}.`$
To isolate the poles in $`\epsilon `$ we do not need to consider higher order terms as it will become clear. If we now integrate over $`\stackrel{}{k}`$, we get
$`\delta ^d(\stackrel{}{r}(𝐱_1)\stackrel{}{r}(𝐱_2))`$ $`=`$ $`{\displaystyle \frac{1}{(4\pi )^{d/2}(G(𝐱_1𝐱_2))^{d/2}}}1`$
$`{\displaystyle \frac{1}{4}}{\displaystyle \frac{(𝐱_1𝐱_2)^\alpha (𝐱_1𝐱_2)^\beta }{(4\pi )^{d/2}(G(𝐱_1𝐱_2))^{d/2+1}}}_\beta \stackrel{}{r}(𝐱)_\alpha \stackrel{}{r}(𝐱)+\mathrm{}`$
$``$ $`C^1(𝐱_1𝐱_2)1+C^{\alpha \beta }(𝐱_1𝐱_2)_\beta \stackrel{}{r}(𝐱)_\alpha \stackrel{}{r}(𝐱)+\mathrm{}`$
where we omit higher dimensional operators in $`\stackrel{}{r}`$, which are irrelevant by power counting, so since the theory is renormalizable they cannot have simple poles in $`\epsilon `$. Additionally, we have defined $`𝐱=\frac{𝐱_1+𝐱_2}{2}`$. One recognizes in Eq. F Wilson’s Operator product expansion, applied to the non-local delta-function operator.
Following the same technique of splitting the operator into a normal ordered part and a singular part at short-distances, it just takes a little more effort to derive the OPE for the product of two delta functions, the result is
$$\delta ^d(\stackrel{}{r}(𝐱_1)\stackrel{}{r}(𝐲_1))\delta ^d(\stackrel{}{r}(𝐱_2)\stackrel{}{r}(𝐲_2))=C(𝐱_1𝐱_2,𝐲_1𝐲_2)\delta ^d(\stackrel{}{r}(𝐱)\stackrel{}{r}(𝐲))+\mathrm{}$$
(115)
with $`C(𝐱_1𝐱_2,𝐲_1𝐲_2)=\frac{1}{(4\pi )^{d/2}(G(𝐱_1𝐱_2)G(𝐲_1𝐲_2))^{d/2}}`$. The terms omitted are again higher dimensional by power counting so they. The OPE Eq. F and Eq. 115 is all we need to compute the renormalization constants at lowest order in $`\epsilon `$, but the calculation may be pursued to higher orders in $`\epsilon `$. In order to do that, one must identify where poles in $`\epsilon `$ arise. In the previous example poles in $`\epsilon `$ appear whenever the internal coordinates ($`𝐱_1`$ and $`𝐱_2`$ in Eq. F, $`𝐱_1`$,$`𝐱_2`$,$`𝐲_3`$ and $`𝐲_2`$ in Eq. 115) are pairwise made to coincide. This is diagrammatically shown in fig. 34. It is possible to show, that higher poles appear in the same way, if more $`\delta `$-product terms are considered.
Let us consider the first delta-function term corresponding to $`n=1`$ in the sum Eq. 110. Using Eq. F we have
$`{\displaystyle \frac{b_rM^\epsilon }{2}}Z_b{\displaystyle d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}_R(𝐱)\stackrel{}{r}_R(𝐲))}`$
$`=`$ $`{\displaystyle \frac{b_rM^\epsilon }{2}}Z_b{\displaystyle }d^D𝐱d^D𝐲(C^1(𝐱𝐲)+C^{\alpha \beta }(𝐱𝐲)_\beta \stackrel{}{r}_R(𝐱)_\alpha \stackrel{}{r}_R(𝐱)+\mathrm{}`$
$`=`$ $`{\displaystyle \frac{b_rM^\epsilon }{2}}Z_b{\displaystyle d^D𝐱_\alpha \stackrel{}{r}_R(𝐱)^\alpha \stackrel{}{r}_R(𝐱)d^D𝐲\frac{\delta _{\alpha \beta }C^{\alpha \beta }}{D}}+\mathrm{}`$
The first term just provides a renormalization of the identity operator, which we can neglect. From
$$_{|𝐲|>1/M}d^D𝐲\frac{\delta _{\alpha \beta }C^{\alpha \beta }}{D}(𝐲)=\frac{1}{4D}\frac{M^\epsilon }{\epsilon }(4\pi )^{d/2}(2D)^{1+d/2}\left(\frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}\right)^{2+d/2},$$
(117)
and we can absorb the pole by $`Z`$ if we define
$$Z=1+\frac{b_R}{\epsilon }\frac{(4\pi )^{d/2}}{4D}(2D)^{1+d/2}\left(\frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}\right)^{2+d/2},$$
(118)
From the short distance behavior in the sum Eq. 110 corresponding to $`n=2`$ we get
$`{\displaystyle \frac{b_r^2M^{2\epsilon }}{8}}{\displaystyle d^D𝐱_1d^D𝐲_1d^D𝐱_2d^D𝐲_2\delta ^d(\stackrel{}{r}_R(𝐱_1)\stackrel{}{r}_R(𝐲_1))\delta ^d(\stackrel{}{r}_R(𝐱_1)\stackrel{}{r}_R(𝐲_2))}`$
$`{\displaystyle \frac{b_r^2M^{2\epsilon }}{8}}{\displaystyle d^D𝐱d^D𝐲\delta ^d(\stackrel{}{r}_R(𝐱)\stackrel{}{r}_R(𝐲))d^D𝐳d^D𝐰C(z,w)},`$ (119)
where, in order to isolate the pole we can perform the following tricks
$`{\displaystyle d^D𝐳d^D𝐰C(z,w)}`$
$`=`$ $`(4\pi )^{d/2}S_D^{d/2}(2D)^{d/2}\left({\displaystyle \frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}}\right)^2{\displaystyle _0^{M^1}}𝑑z{\displaystyle _0^{M^1}}𝑑w{\displaystyle \frac{z^{D1}w^{D1}}{(z^{2D}+w^{2D})^{d/2}}}`$
$`=`$ $`(4\pi )^{d/2}S_D^{d/2}(2D)^{d/2}\left({\displaystyle \frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}}\right)^2{\displaystyle \frac{M^\epsilon }{(2D)^2}}{\displaystyle _0^1}{\displaystyle _0^1}𝑑x𝑑y{\displaystyle \frac{x^{\frac{D}{2D}}y^{\frac{D}{2D}}}{(x+y)^{d/2}}}`$
$`=`$ $`(4\pi )^{d/2}S_D^{d/2}(2D)^{d/2}\left({\displaystyle \frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}}\right)^2{\displaystyle \frac{M^\epsilon }{(2D)^2}}{\displaystyle _{x^2+y^21}}𝑑x𝑑y{\displaystyle \frac{x^{\frac{D}{2D}}y^{\frac{D}{2D}}}{(x+y)^{d/2}}}`$
$`=`$ $`(4\pi )^{d/2}S_D^{d/2}(2D)^{d/2}\left({\displaystyle \frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}}\right)^2{\displaystyle \frac{M^\epsilon }{(2D)^3}}{\displaystyle \frac{\mathrm{\Gamma }(\frac{D}{2D})^2}{\mathrm{\Gamma }(\frac{2D}{2D})}}{\displaystyle \frac{1}{\epsilon }}`$
since changing the boundary of integration from a square to a circle does not affect the residue of the pole. We finally have
$$Z_b=1+\frac{b_R}{\epsilon }\frac{1}{2}(2D)^{1+d/2}\frac{\mathrm{\Gamma }(\frac{D}{2D})^2}{\mathrm{\Gamma }(\frac{2D}{2D})}\left(\frac{2\pi ^{D/2}}{\mathrm{\Gamma }(D/2)}\right)^{2+d/2}(4\pi )^{d/2},$$
(121)
and the $`\beta `$-function follows from the definitions Eq.(108) together with Eq.(118) and Eq.(84).
## Appendix G The mean field solution of the anisotropic case
The free energy has been introduced in Eq. 39. Let us first show the constraints on the couplings so that the Free energy is bounded from below.
* $`u_{yy}>0`$: This follows trivially.
* $`u_{}^{}v_{}+\frac{u_{}}{D1}>0`$ : Define $`A_\alpha ^i=_\alpha r^i(𝐱)`$ then from Eq. 39 we get
$`{\displaystyle \frac{u_{}}{2}}Tr(AA^T)^2+{\displaystyle \frac{v_{}}{2}}(TrAA^T)^2({\displaystyle \frac{u_{}}{D1}}+v_{})/2(TrAA^T)^2`$ (122)
$`=`$ $`{\displaystyle \frac{u_{}^{}}{2}}(TrAA^T)^2,`$
which implies $`u_{}^{}>0`$.
* $`v_y>(u_{}^{}u_{yy})^{1/2}`$ : defining $`\stackrel{}{b}=_y\stackrel{}{r}(𝐱)`$, It is derived from
$$\frac{u_{}^{}}{2}(Tr(A^TA))^2+\frac{u_{yy}}{2}(b^2)^2+v_yb^TbTrAA^T>0.$$
(123)
Introducing the variables
$$A=\left(\begin{array}{cc}v_{}+\frac{u_{}}{D1}& v_y\\ v_y& u_{yy}\end{array}\right),b=(t_{},t_y)$$
(124)
and $`w=((D1)\zeta _{}^2,\zeta _y^2)`$, the mean field effective potential may be written as
$$V(w)=\frac{1}{2}L_{}^{D1}L_y\left[wb+\frac{1}{2}wAw\right].$$
(125)
In this form, it is easy to find the four minima of Eq. 125, those are
1. Crumpled phase:
$$\begin{array}{c}\zeta _{}^2=0\hfill \\ \zeta _y^2=0\hfill \end{array}V_{min}=0$$
(126)
2. Flat phase:
$$\begin{array}{c}\zeta _{}^2=\frac{u_{yy}t_{}v_{}t_y}{\mathrm{\Delta }(D1)}\hfill \\ \zeta _y^2=\frac{v_yt_{}+u_{}^{}t_y}{\mathrm{\Delta }}\hfill \end{array}V_{min}=\frac{L_{}^{D1}L_y}{4\mathrm{\Delta }}\left[u_{}^{}t_y^2+u_{yy}t_{}^22v_yt_{}t_y\right]$$
(127)
3. $``$-Tubule:
$$\begin{array}{c}\zeta _{}^2=\frac{t_{}}{u_{}^{}}\hfill \\ \zeta _y^2=0\hfill \end{array}V_{min}=\frac{L_{}^{D1}L_y}{4}\frac{t_{}^2}{u_{}^{}}$$
(128)
4. $`y`$-Tubule:
$$\begin{array}{c}\zeta _{}^2=0\hfill \\ \zeta _y^2=\frac{t_y}{u_{yy}}\hfill \end{array}V_{min}=\frac{L_{}^{D1}L_y}{4}\frac{t_y^2}{u_yy}$$
(129)
The regions in which each of the four minima prevails depend on the sign of $`\mathrm{\Delta }`$.
* $`\mathrm{\Delta }>0`$: Let us see under which conditions the flat phase may occur. We must satisfy the equations
$`u_{yy}t_{}`$ $`<`$ $`v_yt_y`$
$`u_{}^{}t_y`$ $`<`$ $`v_yt_y`$ (130)
If $`v_y>0`$ this inequalities can only be satisfied if both $`t_{}`$ and $`t_y0`$ have the same sign. If they are positive, Eq. G imply $`\mathrm{\Delta }t_{}<0`$ or $`\mathrm{\Delta }t_y<0`$, which by the assumption $`\mathrm{\Delta }>0`$ cannot be satisfied. The flat phase exists for $`t_y<0`$ and $`t_{}<0`$ and satisfying Eq. G. If $`t_y>0`$ and $`t_{}>0`$ then the flat phase or the tubular cannot exist (see Eq. 128 and Eq. 129) so those are the conditions for the crumpled phase. Any other case is a tubular phase, either $``$-tubule or $`y`$-tubule, depending on which of the inequalities Eq. G is not satisfied. If $`v_y<0`$ it easily checked from Eq. 127 that the flat phase exists as well and the same analysis apply.
* $`\mathrm{\Delta }<0`$: From inequality Eq. 123 we have $`v_y>0`$. The inequalities are now
$`t_y`$ $`<`$ $`{\displaystyle \frac{u_{yy}}{v_{}}}t_{}`$
$`t_y`$ $`>`$ $`{\displaystyle \frac{v_y}{u_{}^{}t_y}}`$ (131)
Now, in order to have a solution for both inequalities we must have $`\frac{u_{yy}}{v_{}}>\frac{v_y}{u_{}^{}}`$ which requires $`\mathrm{\Delta }>0`$. This proves that the flat phase cannot exist. There is then a crumpled phase for $`t_y>0`$ and $`t_{}>0`$ and tubular phase when either one of this two conditions are not satisfied.
|
warning/0002/cond-mat0002173.html
|
ar5iv
|
text
|
# Disorder Induced Diffusive Transport In Ratchets
## Abstract
The effects of quenched disorder on the overdamped motion of a driven particle on a periodic, asymmetric potential is studied. While for the unperturbed potential the transport is due to a regular drift, the quenched disorder induces a significant additional chaotic “diffusive” motion. The spatio-temporal evolution of the statistical ensemble is well described by a Gaussian distribution, implying a chaotic transport in the presence of quenched disorder.
Stochastic models known as thermal ratchets or correlation ratchets , in which a non-zero net drift velocity may be obtained from time correlated fluctuations interacting with broken symmetry structures , have recently received much attention. This interest is due to the wide range of possible applications of these models for understanding such systems as molecular motors , nano-scale friction , surface smoothening , coupled Josephson junctions , optical ratchets and directed motion of laser cooled atoms , mass separation and trapping schemes at the microscale . Recently, spatial disorder in thermal ratchets has been shown to reduce the characteristic drift speed . Little is known, however, about the effects of quenched spatial disorder on the regular or diffusive motion in otherwise periodic potentials .
Diffusion-like motion is observed in many types of deterministic systems. In particular, it has been shown that in deterministic chaotic systems, diffusion can be normal , with the mean-square displacement $`x^2`$ proportional to time $`t`$ ($`x^2t`$), or it can be anomalous , with $`x^2t^\gamma `$, (enhanced for $`\gamma >2`$, dispersive for $`1<\gamma <2`$), or have a logarithmic time dependence ($`\gamma =0`$) .
In the present work we report on an unusual behavior that occurs in the case of an overdamped ratchet subject to an external oscillatory drive: quenched disorder induces a normal diffusive transport in addition to the drift due to the external drive. For the parameter range considered, this process is observed even for very small perturbations. Moreover, this diffusive motion is enhanced by higher values of the quenched disorder. In fact, for high enough disorder the diffusive motion is of the same order of magnitude as the regular drift. The possibility of having large fluctuations, of the same order of magnitude as the average velocity, can be of great importance for a correct interpretation of experimental results. This may be of particular importance in studies of friction, in understanding the motion of nanoclusters or monolayers sliding on surfaces, as well as for designing particles separation techniques.
We consider the one-dimensional, overdamped motion of a particle (in dimensionless units) on a disordered ratchet:
$$\gamma \frac{dx}{dt}=\mathrm{cos}(x)+\mu \mathrm{cos}(2x)+\mathrm{\Gamma }\mathrm{sin}(\omega t)+\alpha \xi (x).$$
(1)
Here, $`\gamma `$ is the damping coefficient, $`\mathrm{\Gamma }`$ and $`\omega `$ are, respectively, the amplitude and frequency of an external oscillatory forcing, and $`\alpha \xi (x)`$ is the force due to the quenched disorder. For the present study, $`\xi (x)[1,1]`$ are independent, uniformly distributed random variables with no spatial correlations, and $`\alpha 0`$ is the amount of quenched disorder. This corresponds to a piecewise constant force on the interval $`\mathbf{[}2k\pi ,2(k+1)\pi \mathbf{)}`$, $`k`$. The unperturbed ratchet potential,
$$U(x)=\mathrm{sin}(x)\mu \mathrm{sin}(2x)$$
(2)
has been the subject of extensive recent studies . The quenched disorder term ($`\alpha 0`$) is expected to give either a more realistic representation of a real substrate or potential landscape, or to model fluctuations in DC current amplitude, as for arrays of Josephson junctions.
It is well known that in the absence of quenched disorder ($`\alpha =0`$) there are unbounded solutions of Eq. (1), provided that the driving amplitude $`\mathrm{\Gamma }`$ is large enough. These solutions tend asymptotically to a constant average velocity independent of the initial conditions . We have identified a set of parameters where, in the absence of disorder, the system shows non-zero current (regular transport). Specifically, we have selected $`\gamma =1.0`$, $`\mu =0.25`$, $`\omega =0.1`$, and we have studied the behavior for several values of $`\mathrm{\Gamma }`$ with $`\mathrm{\Gamma }1.4`$ (see below).
For $`\alpha >0`$ the periodicity of the unperturbed potential is destroyed as a result of the spatial randomness, and solutions of Eq. (1) begin to show a very complex behavior, including chaotic motion. The chaotic behavior is characterized by the rate of divergence of trajectories starting from very close initial conditions, in other words by the leading positive Lyapunov exponent. For a given, fixed realization of the quenched disorder, and for several values of $`\mathrm{\Gamma }1.4`$, we have computed Lyapunov exponents over trajectories starting from origin. We have found positive Lyapunov exponents $`\mathrm{\Lambda }`$ ranging from 2.55, for $`\mathrm{\Gamma }=1.4`$, to 3.22 for $`\mathrm{\Gamma }=1.76`$, which shows a very strong chaotic behavior. As a consequence of this chaotic behavior, the motion of the particle in the perturbed potential should be characterized by ensemble averages performed not only over realizations of disorder, but also over the spatial distribution of the positions of the particle in a given realization of the quenched disorder.
Numerical solutions of Eq. (1) were obtained using a variable step Runge-Kutta-Fehlberg method . Averages were performed over ensembles of 5000 trajectories starting from different initial conditions very close to the origin $`x=0`$. The ensemble described above was left to evolve for 1000 external drive periods $`T`$, and every 10 periods the positions $`x(t)`$ were stored for further analysis.
We have first analyzed the motion in a given realization of disorder. In Fig. 1 we show results for the second moment, $`C_2(t)=(x(t)x(t))^2`$, where $`\mathrm{}`$ means average over the ensemble, as a function of the time $`t`$, for two different, fixed realizations of quenched disorder, and for two values of the disorder parameter $`\alpha `$, $`\alpha =0.05`$ (panel (a)), respectively $`\alpha =0.10`$ (panel (b)). The most striking feature is the fact that the second moment which is zero in the absence of disorder (corresponding to a purely deterministic motion), becomes non-zero in the presence of the perturbation. The non-zero second moment confirms the chaotic behavior mentioned above, showing a disorder-induced sensitive dependence on the initial conditions. It can be seen also that the time-dependence of the second moment is very complicated, and it is dependent on both the realization of quenched disorder and the amplitude $`\alpha `$ of disorder.
In order to perform averages over the realizations of disorder, we have used for each trajectory a different random sequence $`\xi (x)`$. In this way, the averages over the ensemble of trajectories include also averages over realizations of disorder. Figure 2 shows results for the first two moments, $`C_1(t)=x(t)`$ and $`C_2(t)=(x(t)x(t))^2`$, as a function of the time $`t`$ for $`\mathrm{\Gamma }=1.5`$ for several values of disorder parameter $`\alpha `$. In contrast to averages over a given ”landscape”, in this case both first and second moment show an asymptotic linear dependence on time $`t`$, $`C_1(t)v(\alpha )t`$, $`C_2(t)D(\alpha )t`$. We have considered several other values $`1.40\mathrm{\Gamma }1.76`$, and we have observed the same linear behavior for all $`\alpha `$ values below a threshold value which depends on $`\mathrm{\Gamma }`$. The quenched disorder induces fluctuations in the spatial position around the average value in our system, and the dynamics is no longer regular, but rather consists in a superposition of regular drift and diffusion-like chaotic motion. Moreover, even for reasonably small amounts of disorder, for example $`\alpha =0.1`$, it can be seen that these spatial fluctuations are of the same order of magnitude as the first moment, so the knowledge of a particular $`x(t)`$ no longer gives relevant information about the position of the center of mass of the distribution, an observation that can be of importance in studies of friction, particularly the sliding motion of clusters on surfaces .
The fluctuations in the position are characterics of chaotic behavior in deterministic systems. The description of the initial ensemble is then given by a probability distribution function $`p_t(x)`$, whose first two moments are linear in time as we have shown above. We have also calculated the higher order cumulants $`C_n(t)`$, for $`n6`$, and we have found that they increase slower than $`t^{n/2}`$. Therefore, $`p_t(x)`$ is asymptotically a Gaussian, and it is determined by the first two moments . In Figure 3 we show the distributions $`P(z)`$, where $`z=xx`$, for $`\mathrm{\Gamma }=1.50`$ and two values of disorder parameter $`\alpha =0.05`$ (panel (a)), and $`\alpha =0.10`$ (panel (b)), at several times $`t`$, and the scaled distributions $`f(y)=P(z)\times \sqrt{(}t)`$, where $`y=z/\sqrt{t}`$; one can see that the distribution is indeed well aproximated by a Gaussian. This asymptotic Gaussian behavior also supports the conclusion that the motion is chaotic, as it was shown by Jung et al . The reason for this chaotic behavior is the existence of discontinuities in the velocity at $`x_k=2\pi k`$, where $`k`$ is an integer, introduced by the quenched disorder perturbation. These random kicks keep into a transitory regime the trajectories that in the absence of disorder would have asymptotically converged to the asymptotic constant speed state mentioned above. This “mixing” of transitory regimes causes the chaotic behavior, and we emphasize again that it is an effect due solely to the perturbation induced by quenched spatial disorder.
For several values of the external drive amplitude $`\mathrm{\Gamma }`$, we have computed from the slopes of the first two moments, the drift velocity $`v(\alpha )`$, and the diffusion coefficient $`D(\alpha )`$, as functions of the amount of disorder $`\alpha `$. The results shown in Fig. 4 indicate that below a ($`\mathrm{\Gamma }`$ dependent) threshold value of $`\alpha `$ the drift is slightly decreasing with increased quenched disorder, while the diffusion coefficient is steadily increasing and tends to saturate at high amounts of disorder. The fact that disorder has little effect on the drift motion is explained by the fact that the drift is a consequence of the initial asymmetry in the potential, and this asymmetry is only weakly influenced by small perturbations. We note here that there is no decrease in the diffusion coefficient over the range of disorder considered in this study. This is in contrast to the decrease of the diffusion coefficient observed in other systems . The “divergence” of $`D(\alpha )`$ above the threshold can be understood if we consider the fact that $`v(\alpha )`$ decreases to zero. For large enough $`\alpha `$, some of the trajectories in the ensemble become bounded, and their contribution to the second moment is proportional to the displacement of the center of mass, thus with $`t^2`$. The number of bounded trajectories increases with time, as shown by the steady decrease of the drift velocity toward zero. The contribution to the second moment (fluctuations) of the $`t^2`$ term thus increases in time, and becomes dominant at late time, leading to the above mentioned “divergence” of $`D(\alpha )`$.
There is a number of experimental situations where small perturbations of a ratchet potential are relevant, including such systems as surface electromigration , dielectrophoretic trapping, and particle separation techniques . Our preliminary results for the case of a non-negligible inertial term in Eq. (1) are qualitatively similar to the ones for the over-damped case, showing disorder induced chaotic diffusion. In this case, however, both the “diffusion coefficient” $`D`$ and the drift velocity $`v`$ depend on the mass of the particle. Based on the similarities mentioned, our results may be relevant for experiments where the mass dependence of the drift velocity or diffusion coefficient is essential. The efficiency of a nano-scale surface smoothening by an ac field suggested by Derényi et al could be actually significantly smaller than theoretically predicted because of the chaotic diffusion, induced by the inherent ”disorder” of a real surface, superimposed on the net downhill current. On the other hand, the rough, imperfect surface of the electrodes in the dielectrophoretic separation technique suggested by Gorre-Talini et al can actually lead to a better efficiency of the process by superimposing the chaotic diffusion and drift on top of the thermal, Brownian motion. Moreover, the ac-separation techniques using a two-dimensional sieve discussed by Derényi and Astumian can be modified in a very natural way to take advantage of the inherent imperfection of the two-dimensional structure. This can be done by replacing the Brownian diffusion along the drift direction with an additional ac-field along that direction. Also, in this way one can fine tune both the drift velocity and the diffusion coefficient along the separation direction by a convenient choice of the ac-field parameters. The temperature can then be used for an independent tuning of the electrophoretic mobility, thus for the transverse displacement.
In summary we have shown that the addition of small amounts of quenched disorder in the equation of motion of a continuous time system induces a strong diffusive motion. In addition, we have found that the presence of small amounts of disorder slightly decreases the regular current (drift motion), but significantly increases the transport by chaotic diffusion. We have shown also that in the presence of disorder the spatial distribution of positions, averaged over the realizations of disorder, is described by a time-dependent Gaussian distribution, which is a signature of chaotic motion. These unexpected results may help in the interpretation of experimental results in studies of friction, particularly at the nanoscale, as well as in understanding transport processes in molecular motors or designing particle separation techniques.
Acknowledgments
This work was supported by grants from the Office of Naval Research, and from the Universidad Nacional de Mar del Plata. A. L. Salas-Brito wants to thank M. Mina and C. Ch. Ujaya for their friendly support and acknowledges the partial support of CONACyT through grant 1343P-E9607.
|
warning/0002/cond-mat0002232.html
|
ar5iv
|
text
|
# Stripe Dynamics, Global Phase Ordering and Quantum Criticality in High 𝑇_𝑐 Superconductors
## I Introduction
The relevance of quantum criticality to the mechanism of high $`T_c`$ superconductivity in cuprates has captured considerable interest in the theoretical community. One scenario argues that the proximity to a quantum critical point associated with anti-ferromagnetic(AFM) ordering is responsible for the anomalous normal state properties and the pairing mechanism that leads to d-wave superconductivity. There is also recipe that emphasizes the competition between AFM and superconductivity(SC) orders , which is likely controlled by hidden quantum critical points .
Recently there have been convincing experimental evidences that support the presence of stripe ordering (both dynamic and static) in typical cuprate materials such as $`La_{2\delta }Sr_\delta CuO_4`$, $`YBa_2Cu_3O_{7\delta }`$ and $`Bi_2Sr_2CaCu_2O_{8+\delta }`$ (Bi-2212) etc . This offers new possibilities of quantum critical scenarios, considering that stripe phase requires charge order to be locked to local AFM order in its competition with SC ordering. Motivated by these results, suggestions of new critical point associated with charge ordering were advanced.
Among many unresolved issues on relations between various orders, the interplay between stripe order and SC order has been under hot discussions. In the theory suggested in ref , it is argued that Cooper pairing is induced in hole-rich stripes through ”spin proximity” effect caused by pair tunneling between stripe and insulating background, and global phase ordering occurs at a lower energy scale determined by the inter-stripe Josephson coupling which is enhanced by transverse zero-point fluctuations of stripes . In parallel to the above recipe, I recently suggested a scenario where stripes are coupled to an RVB ( Resonating Valence Bond) spin liquid background through single-particle hopping, which results in the generation of two quantitatively different gaps ( normal state pseudo-gap and superconducting gap) by strong pairing correlation inherent in the RVB environment. In both of these scenarios, the dynamical stripes play the central role in accommodating the seeds of Copper pairs for the later establishment of superconductivity order which results from the overlaps of the SC wave function between neighboring one dimensional superconducting stripes, while the hole deficient regions between stripes are less relevant and treated as being remnant of the undoped antiferromagnet with reduced local magnetization which could only compete with the superconducting order. It is then natural to ask: Is it the only possible route toward high $`T_c`$ superconductivity in striped phase of cuprates? In this paper, we will attempt to explore an alternative scenario based on the same microscopic model proposed in ref , albeit from a different viewpoint of the role of stripe dynamics, namely, the effect of dynamical stripes in turning a pre-paired RVB spin liquid into a superconductor.
The relevance of RVB spin liquid to high $`T_c`$ superconductivity was suggested by Anderson more than a decade ago . The basic idea is that the undoped cuprates have a novel quantum disordered ground state which is the resonating superposition of different configurations of local singlet pairs (so called valence bonds). Upon hole doping, these localized electron pairs are gradually liberated and become Cooper pairs which condense into a superconducting ground state. Around this proposal, there have been a lot of discussions and controversies in the theoretical community, and it is still inconclusive . We note that beyond the detailed formulation of RVB theory, there is one important aspect which has enjoyed a broader acceptance , that is, strong local pairing correlation is present even in the normal state of under-doped cuprates although the global phase coherence is established only at a lower temperature. Despite the general argument by Emery and Kivelson about the effect of low carrier density on reducing phase stiffness, the concrete mechanism that is responsible for turning a pre-paired but incoherent RVB liquid into a superconductor with global phase coherence, is albeit unclear. It is interesting to return to this issue now, thanks to the development of experiments which provide precious information and constraints on any serious theoretical effort to understand the high $`T_c`$ mechanism, such as the presence of stripe correlation in the charge degree of freedom which must be taken into account in discussing the above issue.
In this work, we will try to address this issue, based on the consideration of stripe dynamics and its effect on phase ordering in the RVB spin liquid which then gives rise to a global superconducting order. The detailed formulation is provided in next section and the last section is devoted to discussions and conclusions.
## II Formulations
Based on the two-component stripe picture suggested in ref , one can start with the total microscopic Hamiltonian as follows,
$`H(c,c^+,d,d^+)`$ $`=`$ $`H_{1D}(d,d^+)+H_{RVB}(c,c^+)`$ (1)
$`+`$ $`H_{couple}(c,c^+,d,d^+),`$ (2)
where $`c`$, $`c^+`$ and $`d`$, $`d^+`$ represent the annihilation and creation operators of a single electron in 2D RVB background and 1D stripe, respectively. In the undoped background, there is on average one electron per site and the charge degree of freedom is frozen (which is however gradually mobilized when the coupling with stripes develops with hole doping), and only spin exchange interaction is relevant at low energy scale, which is responsible for the singlet formation in RVB state. Within the stripes, where doped holes concentrate, both charge and spin degrees of freedom are active at low energy scale.
Now, let us treat these three parts one by one as follows in order to obtain the low energy effective theory:
Hamiltonian of 2D RVB spin liquid $`H_{RVB}`$: Since the undoped background is at half-filling, one can start with the 2D antiferromagnetic Heisenberg model and perform a routine Hartree-Fork decoupling which leads to the RVB Hamiltonian $`H_{RVB}`$ ,
$`H_{RVB}`$ $`=`$ $`J{\displaystyle \underset{<ij>}{}}(S_iS_j1/4)`$ (3)
$`=`$ $`J{\displaystyle \underset{<ij>}{}}b_{ij}^+b_{ij}`$ (4)
$`=`$ $`J{\displaystyle \underset{<ij>}{}}(\mathrm{\Delta }_{ij}b_{ij}^++h.c.|\mathrm{\Delta }_{ij}|^2),`$ (5)
where $`b_{ij}^+=\frac{1}{\sqrt{2}}[c_{i,}^+c_{j,}^+c_{i,}^+c_{j,}^+]`$. Integrating over fermion variables, one can get the ”free energy ” of RVB order parameters(OP) $`\mathrm{\Delta }_{ij}`$ :
$`F_{RVB}`$ $``$ $`a{\displaystyle \underset{<ij>}{}}|\mathrm{\Delta }_{ij}|^2+b{\displaystyle \underset{<ij>}{}}|\mathrm{\Delta }_{ij}|^4`$ (6)
$`+`$ $`c{\displaystyle \underset{plaquette[ijkl]}{}}(\mathrm{\Delta }_{ij}^{}\mathrm{\Delta }_{jk}\mathrm{\Delta }_{kl}^{}\mathrm{\Delta }_{li}+h.c.)+\mathrm{},`$ (7)
where $`a`$,$`b`$,$`c`$ are parameters derived from microscopic model calculations. Then approaching the continuous limit by coarse graining: $`\mathrm{\Psi }(\stackrel{}{r}\frac{\stackrel{}{r}_i+\stackrel{}{r}_j}{2})\mathrm{𝑙𝑜𝑐𝑎𝑙}\mathrm{𝑎𝑣𝑒𝑟𝑎𝑔𝑒}\mathrm{𝑜𝑓}|\mathrm{\Delta }_{ij}|\mathrm{exp}(i\theta _{ij})`$, one can arrive at the following effective action
$`S_{RVB}^{eff}={\displaystyle 𝑑\tau 𝑑x𝑑y[a^{}|\mathrm{\Psi }|^2+b^{}|\mathrm{\Psi }|^4c^{}|\mathrm{\Psi }|^2|\mathrm{\Psi }|^2]},`$ (8)
where $`a^{}<0`$, $`b^{}>0`$, $`c^{}<0`$ are renormalized parameters from $`a`$,$`b`$,$`c`$ by coarse graining . Note that the temporal phase stiffness ($`\frac{1}{U}`$, where $`U`$ is the local charging energy) is rather weak compared with the spatial phase stiffness $`E_s|\mathrm{\Psi }|^2`$ and thus does not appear in Eq. This is because charge fluctuation is significantly suppressed by strongly repulsive interactions, which leads to severe phase fluctuations thanks to the conjugating relation between phase variable and pair density. This accounts for the absence of phase coherence in half-filled RVB spin liquid. We note that the lack of phase coherence (or put it in a different way, the freezing of charge fluctuations) is what makes RVB spin liquid different from a superconductor of Cooper pairs, later on we will see how stripe dynamics helps to establish phase coherence in RVB spin liquid and turns it into a superconductor.
Hamiltonian of 1D stripes $`H_{1D}`$: As shown by extensive experiments, stripes are dynamical in nature , and its dynamics includes the longitudinal charge and spin fluctuations along the stripe and the transverse motion which is relatively slow. Therefore it is valid to treat the stripes as 1D electron gas(1DEG) at first and then take the transverse degree of freedom into account by doing the suitable average over the transverse configurations. The general theory of Luttinger Liquid provides a powerful tool for the description of 1DEG. Following the notation of Luttinger Liquid , the low energy effective theory of 1D stripes is described by separated charge modes and spin modes as follows :
$`H_{1D}^{eff}`$ $`=`$ $`{\displaystyle 𝑑x[\frac{K_cu_c}{2}\mathrm{\Pi }_c^2+\frac{u_c}{2K_c}(_x\mathrm{\Phi }_c)^2]}`$ (9)
$`+`$ $`{\displaystyle 𝑑x[\frac{u_s}{2}\mathrm{\Pi }_s^2+\frac{u_s}{2}(_x\mathrm{\Phi }_s)^2+g_1\mathrm{cos}(\sqrt{8\pi }\mathrm{\Phi }_s)]},`$ (10)
where $`\mathrm{\Phi }_c,\mathrm{\Pi }_c`$, and $`\mathrm{\Phi }_s,\mathrm{\Pi }_s`$ , are conjugated boson operators representing density fluctuations in charge and spin sectors of 1D Luttinger Liquid, respectively. $`u_c`$, $`u_s`$ are the corresponding propagating velocities, and $`K_c`$ is a parameter of interaction. The last term $`g_1\mathrm{cos}(\sqrt{8\pi }\mathrm{\Phi }_s)`$ is the spin gap correction caused by the coupling between 1D stripes and 2D RVB background with strong pairing correlations (see ref for detail).
Coupling between 1D stripes and 2D RVB background $`H_{couple}`$:
$`H_{couple}(c,c^+,d,d^+)=_{k,q,\sigma }Vc_{k,\sigma }^+d_{q,\sigma }\delta _{k_x,q}+h.c.`$, where only horizontal stripe (along x direction) is considered, and momentum conservation is ensured by requiring $`k_x=q`$. The coupling term accounts for the single electron hopping between stripes and the background and $`V`$ gives the hopping matrix element. In order to discuss the low energy effective theory with boson variables only, we need to integrate out fermion variables so the lowest order relevant process happens at the second order of $`V`$. For the present discussion, we adopt the Cooper pair tunneling process as the only effective coupling between 2D RVB environment and 1D stripes, that is
$`H_{couple}^{eff}(\mathrm{\Psi },\mathrm{\Delta })=g{\displaystyle 𝑑x𝑑y\mathrm{\Psi }(x,y)\mathrm{\Delta }^{}(x)f(yY)}+h.c.,`$ (11)
where $`g`$ is the continuous limit of pair tunneling amplitude, $`\mathrm{\Psi }(x,y)`$ stands for the coarse grained RVB order parameter , $`\mathrm{\Delta }(x)`$ is the singlet pairing order parameter of stripes, $`f(yY)`$ gives the transverse distribution function of the stripe position due to vibration , and $`Y`$ stands for the transverse displacement of the stripe diffusion. Here we assume that the stripe transverse modes can be described as the superposition of the fast mode of vibration and the slow mode of diffusion. Under the harmonic approximation, one can assume $`f(y)\sqrt{\alpha }\mathrm{exp}(\alpha y^2)`$, where $`1/\sqrt{\alpha }`$ represents the amplitude of stripe vibration mode which is determined by the microscopic details which are responsible for the stripe formation and stabilization. In Nd doped LSCO where stripes are pinned by the Low Temperature Tetragonal lattice structure , $`1/\sqrt{\alpha }`$ is expected to be small (of the order of one lattice unit); while in optimum doped LSCO and YBCO, stripes are more disordered and transverse fluctuations are strong, so $`1/\sqrt{\alpha }`$ can be as large as the order of inter-stripe distance which is about $`4a`$ ( $`a`$ is the lattice unit).
In the formulation of Luttinger Liquid ,
$$\mathrm{\Delta }(x)=d_u\mathrm{exp}(i\sqrt{2\pi }\mathrm{\Theta }_c)\mathrm{sin}(\sqrt{2\pi }\mathrm{\Phi }_s).$$
We note that this pair tunneling process is very important in the present picture. Through this process the previously localized spin singlets in RVB background become mobile and the charge degree of freedom is resumed, in this sense one can no longer distinguish spin singlets of electrons from Cooper pairs which constitute the basis of superconductivity, and the RVB phase variable can be continuously connected to the phase degree of freedom of d-wave superconductivity order parameter.
After establishing the effective Hamiltonians of the coupled RVB and stripe variables, we can study the effect of stripe dynamics on the RVB background ( especially the phase degree of freedom ) by integrating out the stripe variables ( including both the OP field $`\mathrm{\Delta }(x)`$ and the transverse mode variable $`Y`$)
$`\mathrm{exp}[\mathrm{\Delta }S_{RVB}^{eff}]`$ $`=`$ $`{\displaystyle D\mathrm{\Phi }_c(x,\tau )D\mathrm{\Phi }_s(x,\tau )DY(\tau ,x)}`$ (13)
$`\mathrm{exp}[{\displaystyle _0^\beta }𝑑\tau (H_{1D}^{eff}+H_{couple}^{eff}+H_{tm})],`$
where $`H_{tm}`$is the Hamiltonian of the transverse modes of stripes, $`\beta =\frac{1}{k_BT}`$. Therefore, the integration up to the second order gives
$`\mathrm{\Delta }S_{RVB}^{eff}`$ $``$ $`{\displaystyle \frac{g^2}{2}}{\displaystyle 𝑑\tau 𝑑\tau ^{}𝑑x𝑑y𝑑x^{}𝑑y^{}\mathrm{\Psi }(x,y,\tau )\mathrm{\Psi }^{}(x^{},y^{},\tau ^{})}`$ (15)
$`\mathrm{\Delta }^{}(x,\tau )\mathrm{\Delta }(x^{},\tau ^{})_{1D}f(yY(\tau ,x))f(y^{}Y(\tau ^{},x^{}))_{tm}`$
where $`_{1D}`$ and $`_{tm}`$ stand for average over longitudinal and transverse stripe variables, respectively. According to the Luttinger Liquid theory,
$$\mathrm{\Delta }^{}(x,\tau )\mathrm{\Delta }(x^{},\tau ^{})_{1D}\{\begin{array}{cc}\frac{d_u^2}{|(\tau \tau ^{})+i\frac{(xx^{})}{u_c}|^{\frac{1}{K_c}}|(\tau \tau ^{})+i\frac{(xx^{})}{u_s}|}\hfill & |(\tau \tau ^{})+i\frac{(xx^{})}{u_{s,c}}|<<\frac{\xi _s}{u_{s,c}}\hfill \\ \frac{d_u^2}{|(\tau \tau ^{})+i\frac{(xx^{})}{u_c}|^{\frac{1}{K_c}}\xi _s/u_s}\hfill & |(\tau \tau ^{})+i\frac{(xx^{})}{u_{s,c}}|>>\frac{\xi _s}{u_{s,c}}\hfill \end{array}$$
where $`\xi _s1/\mathrm{\Delta }_s`$ is the cutoff of length scale given by the spin gap $`\mathrm{\Delta }_s\sqrt{|g_1|}\mathrm{exp}(\frac{v}{2\pi g_1})`$ .
In order to extract the spatial and temporal phase stiffness coefficients, one can expand the integrand with respect to $`\mathrm{\Delta }x=x^{}x`$, $`\mathrm{\Delta }y=y^{}y`$, $`\mathrm{\Delta }\tau =\tau ^{}\tau `$, over which one can perform integrations , then reach
$`\mathrm{\Delta }S_{RVB}^{eff}{\displaystyle 𝑑\tau 𝑑x𝑑y[E_\tau |\frac{\mathrm{\Psi }}{\tau }|^2+\mathrm{\Delta }E_x|\frac{\mathrm{\Psi }}{x}|^2+\mathrm{\Delta }E_y|\frac{\mathrm{\Psi }}{y}|^2]},`$ (16)
where the induced incipient temporal phase stiffness is
$`E_\tau `$ $``$ $`\delta g^2\alpha {\displaystyle _0^{\frac{\xi _s}{u_{c,s}}}}𝑑\mathrm{\Delta }\tau \mathrm{exp}[\alpha (Y(\tau +\mathrm{\Delta }\tau ,x)Y(\tau ,x))^2/2]_{tm}`$ (19)
$`(\mathrm{\Delta }\tau )^{11/K_c}+\delta g^2\alpha {\displaystyle _{\frac{\xi _s}{u_{c,s}}}^{\mathrm{}}}𝑑\mathrm{\Delta }\tau \mathrm{exp}[\alpha (Y(\tau +\mathrm{\Delta }\tau ,x)Y(\tau ,x))^2/2]_{tm}`$
$`(\mathrm{\Delta }\tau )^{21/K_c},`$
Considering that $`(Y(\tau +\mathrm{\Delta }\tau ,x)Y(\tau ,x))^2_{tm}=2D(\mathrm{\Delta }\tau )`$ (where the stripe diffusion is modeled as random walk and $`D`$ is the diffusive coefficient), then the above expression can be simplified to
$`E_\tau \delta g^2\alpha (\alpha D)^{1/K_c2}F({\displaystyle \frac{\alpha D\xi _s}{u_c}})+{\displaystyle \frac{u_{c,s}}{\xi _s}}\delta g^2\alpha (\alpha D)^{1/K_c3}G({\displaystyle \frac{\alpha D\xi _s}{u_c}}),`$ (20)
where $`F(X)=_0^Xx^{11/K_c}e^x𝑑x`$, $`G(X)=_X^{\mathrm{}}x^{21/K_c}e^x𝑑x`$. At the limit $`\frac{\alpha D\xi _s}{u_c}>>1`$, $`E_\tau \delta g^2\alpha (\alpha D)^{1/K_c2}`$; while for $`\frac{\alpha D\xi _s}{u_c}<<1`$, $`E_\tau \delta g^2\alpha (\alpha D)^{1/K_c3}/\xi _s`$.
The corrections to spatial phase stiffness $`\mathrm{\Delta }E_x`$ and $`\mathrm{\Delta }E_y`$ can be calculated similarly, however, compared with the unperturbed $`E_s`$ of 2D RVB effective theory in eq they are negligibly small when $`\delta `$ is small enough.
Now combined with eq where the spatial phase stiffness $`E_s\rho ^2`$ is given, and retain only phase variables (assuming frozen amplitude $`|\mathrm{\Psi }|=\rho `$), one can discuss the phase ordering process in the RVB liquid with the following effective action:
$`S_{eff}`$ $`=`$ $`{\displaystyle _0^\beta }d\tau [{\displaystyle \underset{i}{}}E_\tau \rho ^2|{\displaystyle \frac{\varphi (\stackrel{}{r}_i)}{\tau }}|^2`$ (21)
$``$ $`E_s\rho ^2{\displaystyle \underset{<ij>}{}}\mathrm{cos}(\varphi (\stackrel{}{r}_i)\varphi (\stackrel{}{r}_j))].`$ (22)
Notice here we turn from the Ginzburg-Landau like ”soft-spin” effective model into ”hard spin” XY model, because in (2+1) dimension they belong to the same universal class and thus have the same critical behavior.
This effective action has been under heavy discussions in the study of granular superconducting film and Josephson junction array. To reveal the quantum critical physics inside it, one can perform a standard Hubbard-Stratonovich transformation to decouple the Josephson term , which introduces the complex order-parameter field $`\psi `$ in proportion to the expectation value of $`\mathrm{exp}(i\varphi )`$. The resulting Ginzburg-Laudau action in (2+1)D reads:
$`F_{eff}[\psi ]`$ $`=`$ $`{\displaystyle }dxdyd\tau [{\displaystyle \frac{1}{8E_s\rho ^2}}|\psi |^2+128E_\tau ^3\rho ^6|{\displaystyle \frac{\psi }{\tau }}|^2`$ (23)
$`+`$ $`({\displaystyle \frac{1}{2E_s\rho ^2}}4E_\tau \rho ^2)|\psi |^2+\kappa |\psi |^4].`$ (24)
So the quantum critical point(QCP) is given by $`\frac{1}{2E_s\rho ^2}4E_\tau \rho ^2=0`$, which separates the zero-temperature phase diagram into superconducting ordered phase ($`E_\tau E_s\rho ^4>1/8`$) and non-superconducting disordered phase ($`E_\tau E_s\rho ^4<1/8`$). At finite temperature, there exists a crossover temperature $`T_{cr}(\rho ,E_\tau )\sqrt{\frac{|E_\tau \rho ^2\frac{1}{8E_s\rho ^2}|}{E_\tau ^3\rho ^6}}`$, above which lies the quantum critical region where physical quantities obey scaling laws with $`T`$. On the SC ordered side the crossover temperature becomes the transition temperature corresponding to the well-known KT transition . The phase diagram is shown in Fig.1(a). Note that strong asymmetry exists in $`T_{cr}`$ around the QCP, and the much higher crossover temperature on the disordered side compared with the SC ordered side can explain why the anomalous $`T`$ dependent scaling behaviors are prevalent in the normal states of superconducting cuprates while in slightly doped insulating cuprates the critical regime eludes experiments (it is however likely that stripe ordering itself can lead to critical scaling behavior which is not considered here).
## III Discussions and Conclusions
Now let us discuss how to connect this QCP with the general phase diagram of high $`T_c`$ cuprates. One can see the quantum phase transition is tuned by a single parameter $`E_\tau E_s\rho ^4\delta g^2\rho ^6H(\xi _s,\alpha ,D)`$ which is a complicated function of doping density $`\delta `$, RVB OP amplitude $`\rho `$, spin gap $`1/\xi _s`$ and stripe transverse mode parameters $`\alpha `$ and $`D`$. In realistic experiments, upon hole doping, all the other parameters change accordingly. For example, RVB OP amplitude $`\rho `$ and spin gap both decrease with doping (experimentally spin gap closes around $`\delta =0.2`$), while stripe transverse modes may depend on material-dependent properties like lattice distortions and impurity effects. Therefore a comprehensive understanding of this issue can be formidable and will not be pursued here. However For the purpose of qualitative demonstration of the physical mechanism , I will attempt to take some of the relevant parameters into account (while leave the others like those of stripe transverse modes as external inputs) and mark the route followed by a cuprate under hole doping in the ground state phase diagram (Fig.1(b)).At first, with slightly doping from the parent cuprate, $`E_\tau `$ increases from zero (roughly in proportion to $`\delta `$) while $`\rho `$ gradually decreases from the maximum, therefore at a critical doping value ($`\delta _{c1}0.06`$) the system crosses the phase boundary into the SC ordered state. Then upon further doping from under-doped to over-doped regions, $`\delta `$ gradually becomes saturated, meanwhile a diminishing spin gap pushes $`E_\tau `$ toward the limit value controlled by $`\alpha `$ and $`D`$. Therefore the route is bent toward $`E_\tau `$ axis thanks to the decreasing $`\rho `$ (because over-doping reduces RVB correlations significantly with excessive holes ”overflowing” into the background, which is also consistent with the result of RVB mean field calculations ). Finally as $`\delta >\delta _{c2}0.3`$ the system crosses the phase boundary again and returns to the disordered non-superconducting ground state. During the above process, $`T_{cr}`$ increases from zero to its maximum and then decreases back to zero, as it is the case for the transition temperature .
Before end, two comments are in order. First, I will comment on the role of transverse stripe modes in affecting the SC transition. According to eq, lower $`\alpha `$ and $`D`$ tends to strengthen $`E_\tau `$ (which is especially effective in under doped region where the spin gap is substantial and the limit $`\frac{\alpha D\xi _s}{u_c}<<1`$ can be approached, assuming $`K_c>1/2`$ which coincides with the condition under which SC fluctuations along stripes are relevant at low energy ). This suggests that in the present mechanism, larger transverse vibration amplitude ($`1/\sqrt{\alpha }`$) favors SC while the diffusion mode does not. Considering the various stripe phases proposed in literature , it is interesting to note that SC order is favored only in the intermediate region between the stripe crystal phase ( with small vibration amplitude, or large $`\alpha `$ ) and stripe liquid phase ( where stripes are meandering strings and diffusions dominate), which implies a very subtle relation between SC order and stripe charge order. Secondly, I will briefly compare the present picture with the one suggested in ref : in that work, the superconducting order is induced by the Josephson tunneling between neighboring stripes and it is natural to expect this coupling to be strongly dependent on the inter-stripe distance (presumably decays exponentially with the distance ) and also the extent of disorder in stripe configurations, which makes it a subtle issue to explain the simple and well-defined relation between $`T_c`$, zero temperature superfluid density and doping density, and the fact that higher $`T_c`$ is found in the cuprates with more disordered stripe correlations. In the present work, the induced temporal phase stiffness only depends on the local coupling between one stripe and its neighboring background and is therefore not sensitive to the disorder in the coupling between the neighboring stripes.
In conclusion, the low energy effective theory of the RVB phase variable coupled to the stripe dynamics is obtained, where the effect of stripe dynamics induces doping dependent incipient temporal phase stiffness in the RVB liquid, which tunes a quantum phase transition toward a superconducting ground state with global phase order.
I am grateful to S.A.Kivelson for discussions . The support from Stanford Graduate Fellowship (SGF) and SSRL is gratefully acknowledged.
|
warning/0002/cond-mat0002211.html
|
ar5iv
|
text
|
# Spin-flux phase in the Kondo lattice model with classical localized spins
## Abstract
We provide numerical evidence that a spin-flux phase exists as a ground state of Kondo lattice model with classical local spins on a square lattice. This state manifests itself as a double-Q magnetic order in the classical spins with spin density at both $`(0,\pi )`$ and $`(\pi ,0)`$ and further exhibits fermionic spin currents around an elementary plaquette of the square lattice. We examine the spin-wave spectrum of this phase. We further discus an extension to a face centered cubic (FCC) lattice where a spin-flux phase may also exist. On the FCC lattice the spin-flux phase manifests itself as a triple-Q magnetically ordered state and may exist in $`\gamma `$-Mn alloys.
The Kondo lattice model with classical local spins has emerged as one of the simplest models that can account for some of the physics of the manganites and the cuprates . For the manganites the ferromagnetic Kondo lattice model gives rise to the double exchange model which has been argued to be the relevant model to explain the physics in these materials . For the cuprates the assumption of classical local spins is clearly unrealistic, however the antiferromagnetic Kondo lattice model gives rise to many insights into high $`T_c`$ materials. For example, it has been used to understand the appearance of $`d`$-wave superconductivity and it also gives rise to incommensurate magnetic and stripe structures that have been experimentally observed . One aspect of this model that has recently gained interest is the appearance of a Berry phase in the fermion wave function that arises when the fermion spin is strongly pinned to the local classical spin orientation . This Berry phase has been argued to give rise to a flux phase ground state on a square lattice in the manganites , to an anomalous hall-effect in ferromagnets , and to a quantized hall conductance in Kagome lattices . In this paper we examine the conditions under which this Berry phase gives rise to novel ground state structures. In particular, we give numerical evidence that a spin-flux phase appears as a ground state structure of this model on a square lattice. This phase is analogous to but quite different from the flux phases that are usually discussed in the context of the cuprates . The difference arises because the latter flux phases exhibit a finite current around each elementary plaquette of the square lattice, but in our case spin currents exist (for which the up and down electrons have opposite currents around an elementary plaquette). On the square lattice the phase discussed here has a spin-flux of $`\pi `$ through each elementary plaquette. In this regard, it is of interest to note that a $`\pi `$ spin-flux phase has been argued to be central in explaining the normal state properties of the cuprates by John and co-workers .
In this paper we will first demonstrate numerically that a double-Q magnetic structure exists as a ground state of the Kondo lattice model. We then demonstrate that such a state is a spin-flux state with circulating spin currents and estimate the stability region of this phase. We also determine the spin-wave spectrum of the double-Q magnetic phase and demonstrate that a spin flux phase may exist for this model on a face centered cubic (FCC) lattice. The spin-flux phase on the FCC lattice manifests itself as a triple-Q magnetically ordered state and has a spin flux of $`\pi /2`$ through each elementary triangular plaquette that lies in the planes having Miller indices $`(1,1,1)`$ (and equivalent symmetry planes).
The model we study here is
$$\mathrm{H}=\mathrm{t}\underset{\mathrm{𝐢𝐣}\alpha }{}(\mathrm{c}_{𝐢\alpha }^{}\mathrm{c}_{𝐣\alpha }+\mathrm{h}.\mathrm{c}.)\mathrm{J}\underset{𝐢}{}𝐬_𝐢𝐒_𝐢+\mathrm{J}^{}\underset{\mathrm{𝐢𝐣}}{}𝐒_𝐢𝐒_𝐣,$$
(1)
where $`\mathrm{c}_{𝐢\alpha }^{}`$ creates an electron at site $`𝐢=(i_x,i_y)`$ with spin projection $`\alpha `$, $`𝐬_𝐢`$=$`_{\alpha \beta }\mathrm{c}_{𝐢\alpha }^{}𝝈_{\alpha \beta }\mathrm{c}_{𝐢\beta }`$ is the spin of the mobile electron, the Pauli matrices are denoted by $`𝝈`$, $`𝐒_𝐢`$ is the localized spin at site $`𝐢`$, $`\mathrm{𝐢𝐣}`$ denotes nearest-neighbor (NN) lattice sites, $`\mathrm{t}`$ is the NN-hopping amplitude for the electrons, $`\mathrm{J}`$ is a coupling between the spins of the mobile and localized degrees of freedom, and $`\mathrm{J}^{}>0`$ is a direct AF coupling between the localized classical spins. Throughout this article the unit of energy will correspond to $`t=1`$. For the numerical studies a Monte Carlo technique was used. This involves no “sign problems” so that by this procedure temperatures as low as T=0.005 at any density can be reached. The present study has been performed mostly on 6$`\times `$6 lattices with periodic boundary conditions (PBC), but occasional runs were made also using open and antiperiodic BC as well as different lattice sizes (up to 12$`\times `$12 lattices). The specific numerical technique used here involves a standard Metropolis algorithm for the classical spins and an exact diagonalization for the itinerant electrons. The details of the method are described in Ref. .
The spin-flux phase was identified numerically by studying the classical spin structure factor which is the Fourier transform of the static spin-spin correlation function $`S(𝐪)=\frac{1}{N}_{𝐧,𝐦}e^{i𝐪(𝐧𝐦)}𝐒_𝐧𝐒_𝐦`$. In particular, it was found that for various $`JS`$ and $`J^{}S^2`$ in the vicinity of electron density $`n=0.5`$ the structure factor was peaked at $`𝐐=(0,\pi )`$ ($`𝐐_𝐲`$) and $`𝐐=(\pi ,0)`$ ($`𝐐_𝐱`$) (see Fig. 1). To understand the possible ground states for a spin density with this wave vector it is useful to look at a Ginzburg Landau free energy. The order parameter is determined by the two vectors $`𝐌_{0,\pi }`$ and $`𝐌_{\pi ,0}`$. The free energy can be simply constructed by noting that the relevant space group representation transforms as a vector under spin rotations and as a scalar (that is as an $`A_{1g}`$ representation) under the little co-group $`D_{2H}`$ of the wavevector $`𝐐=(0,\pi )`$. The most general dimensionless Ginzburg Landau free energy is
$$F=(𝐌_{0,\pi }^2+𝐌_{\pi ,0}^2)+(𝐌_{0,\pi }^2+𝐌_{\pi ,0}^2)^2+\beta 𝐌_{0,\pi }^2𝐌_{\pi ,0}^2+\beta _2(𝐌_{0,\pi }𝐌_{\pi ,0})^2$$
(2)
The minimization of this energy leads to three possible ground states: (a) $`(𝐌_{0,\pi },𝐌_{\pi ,0})=(𝐌,0)`$, (b) $`(𝐌_{0,\pi },𝐌_{\pi ,0})=(𝐌_1,𝐌_2)`$ with $`𝐌_1𝐌_2=0`$, (c) $`(𝐌_{0,\pi },𝐌_{\pi ,0})=(𝐌,𝐌)`$. The double-Q phase (b) we will argue below is the spin-flux phase which in fact corresponds to a particular representation of flux phase proposed by Yamanaka et al. (note that this phase does not lead to a peak in $`S(𝐪)`$ at $`(\pi /2,\pi /2)`$ as suggested in Ref. ). The double-Q phase (c) corresponds to ordering only one half of the local moments and is therefore not a likely ground state for this model (note however that there exists numerical evidence for this phase in a periodic Anderson model on a square lattice ). To distinguish numerically which of these three phases corresponds to the phase found here the spin correlations were examined by evaluating $`𝐒_𝐢𝐒_𝐣=S^2\mathrm{cos}\theta _{ij}`$ (the spin dot product of NN spins) for each pair of NN spins and plotting the value of $`\mathrm{cos}\theta _{ij}`$ in a histogram. The results are shown in Fig. 1 for $`JS=2`$ and $`J^{}S^2=0`$. From this figure it is clear that NN spins are orthogonal which implies the double-Q order of phase (b) above.
To understand the electronic properties of this double-Q magnetic phase we fix the classical spins and find the fermion eigenstates (for the double-Q state this is reasonable because the spin structure factor is very strongly peaked at $`(0,\pi )`$ and $`(\pi ,0)`$ with little weight at other $`𝐪`$ values as can be seen in Fig 1). The classical spin orientation is given by $`𝐒_𝐢=(S/2)[(1)^{i_x}+(1)^{i_y},(1)^{i_x}(1)^{i_y},0]`$. Solving for the eigenstates of the resulting electronic Hamiltonian results in four bands with dispersions
$$ϵ_𝐤=\pm \sqrt{(JS)^2+4(\mathrm{cos}^2k_x+\mathrm{cos}^2k_y)\pm 2\sqrt{2(JS)^2(\mathrm{cos}^2k_x+\mathrm{cos}^2k_y)+16\mathrm{cos}^2k_x\mathrm{cos}^2k_y}}$$
(3)
where $`(k_x,k_y)\{|k_x+k_y|\pi \}\{|k_xk_y|\pi \}`$ are restricted to one half the original Brillouin zone. The density of states (DOS) is linear in $`|k|`$ for $`n=0.5`$ which is characteristic of the Dirac spectrum that appears for $`\pi `$-flux phases . Also note that the dispersion relation is independent of the sign of $`J`$; consequently if this flux phase is the ground state for positive $`J`$ then it must also be the ground state for negative $`J`$. In the limit $`J=\mathrm{}`$ the dispersion reduces to that found in Ref. . To identify this phase as a spin-flux state the spin current from site $`i`$ to $`j`$ was determined
$$𝐣_{i,j}=it\underset{\alpha ,\beta }{}c_{\alpha ,i}^{}𝝈_{\alpha ,\beta }c_{\beta ,j}c_{\alpha ,j}^{}𝝈_{\alpha ,\beta }c_{\beta ,i}.$$
(4)
It was found that only $`[𝐣_{ij}]_z`$ is non-zero and it is non-zero only for NN sites. The resulting spin currents circulate neighboring plaquettes in opposite directions. The charge current was found to be zero. This spin current pattern implies a spin-flux of $`\pi `$ exists in each elementary plaquette. Note that in the limit $`J=\mathrm{}`$ this result is intuitively clear; in this limit the spin of the electron is tied to the local moment so that when the fermion travels around a plaquette the spin changes by $`2\pi `$ which implies that the wavefunction changes sign.
The phase diagram for $`n=0.5`$ as a function of $`1/(JS)`$ and $`J^{}S^2`$ is shown in Fig. 2. The solid phase boundaries were found by comparing the energy of the flux phase to that of the canted magnetic and spin density wave (SDW) phases (the energies of the helical SDW phases agree with those found in Ref. ). At larger $`J^{}S^2`$ the flux phase is found to be unstable to a SDW phase characterized by $`𝐒_𝐢=𝐒\sqrt{2}(1)^{i_y}\mathrm{cos}(i_x\pi /2\pi /4)`$ (note this is not a helical SDW state). This agrees with the structure found numerically.
It is of interest to determine the spin-wave spectrum arising from the spin-flux phase. This can be done by using the spin-wave approximation that was introduced by Kubo and Ohata and later used by Furukawa for the double-exchange model . The local spins are described by a local co-ordinate system in which each classical spin is aligned along the $`\widehat{z}`$ direction and the spin-wave operators $`S_i^+\sqrt{2S}a_i`$, $`S_i^{}\sqrt{2S}a_i^{}`$, and $`S_i^z=Sa_i^{}a_i`$ are introduced. The spin-wave spectrum is found by keeping all $`1/S`$ corrections to the magnon self-energy. Here we consider the limit $`J=\mathrm{}`$. This results in the following effective boson Hamiltonian for the spin waves
$$\underset{𝐤}{}[\mathrm{\Pi }(𝐤)a_𝐤^{}a_𝐤+A(𝐤)a_𝐤^{}a_𝐤+h.c.]$$
(5)
with $`𝐤`$ summed over the whole Brillouin zone of the square lattice,
$$\mathrm{\Pi }(𝐤)=\frac{1}{2SN}\underset{𝐪}{}\left\{E_𝐪\mathrm{cos}(\theta _𝐪\theta _{𝐤+𝐪})E_{𝐤+𝐪}\frac{E_{𝐤+𝐪}^2}{E_𝐪+E_{𝐤+𝐪}}\left[1\mathrm{cos}(5\theta _{𝐤+𝐪}\theta _𝐪)\right]\right\}+J^{}S(\mathrm{cos}k_x+\mathrm{cos}k_y),$$
(6)
$$A(𝐤)=\frac{1}{2SN}\underset{𝐪}{}\frac{E_{𝐤+𝐪}E_𝐪}{E_𝐪+E_{𝐤+𝐪}}\left[\mathrm{cos}(2\theta _𝐪+2\theta _{𝐤+𝐪})\mathrm{cos}(\theta _𝐪\theta _{𝐤+𝐪})\right]+J^{}S(\mathrm{cos}k_x+\mathrm{cos}k_y),$$
(7)
$`\mathrm{cos}\theta _𝐤=[\mathrm{cos}k_x+\mathrm{cos}k_y]/E_𝐤`$, $`\mathrm{sin}\theta _𝐤=[\mathrm{cos}k_x\mathrm{cos}k_y]/E_𝐤`$, and $`E_𝐤=\sqrt{2}\sqrt{\mathrm{cos}^2k_x+\mathrm{cos}^2k_y}`$. The spin wave dispersion is given by $`\omega _𝐤^2=\sqrt{\mathrm{\Pi }(𝐤)^2|A(𝐤)|^2}`$. The eigenstate of this mode is given by $`\delta 𝐒_𝐪(𝐫)=e^{i𝐪𝐫}\left[e^{i𝐐_𝐱𝐫}(\mathrm{\Pi }(𝐪)+A(𝐪))\frac{\widehat{x}+\widehat{y}}{2}+i\omega _𝐪\widehat{z}+e^{i𝐐_𝐲𝐫}(\mathrm{\Pi }(𝐪)+A(𝐪))\frac{\widehat{y}\widehat{x}}{2}\right]`$ (recall the ordered moment is $`𝐒_0(𝐫)=e^{i𝐐_𝐱𝐫}\frac{\widehat{x}+\widehat{y}}{2}+e^{i𝐐_𝐲𝐫}\frac{\widehat{x}\widehat{y}}{2}`$). The resulting spin-wave spectrum (see Fig. 3) agrees with the general form required by phenomenological arguments (found by using the method of Zhu and Walker ).
Given that the spin-flux phase was found to be stable on a square lattice it is natural to ask whether such states can be realized on other lattice structures. We argue that a spin-flux phase is likely to be stable on a FCC lattice. For an FCC lattice two degenerate ground states of the NN classical Heisenberg model are $`𝐒_{3Q}(𝐑)=[(1)^{l_1+l_2},(1)^{l_2+l_3},(1)^{l_2+l_3}]/\sqrt{3}`$ and $`𝐒_{1Q}(𝐑)=[(1)^{l_1+l_2},0,0]`$ where the FCC lattice is spanned by $`𝐑=[l_1a(\widehat{x}+\widehat{y})/2,l_2a(\widehat{x}+\widehat{z})/2,l_3a(\widehat{y}+\widehat{z})/2]`$ (note that there exists a continuous degeneracy in the ground state, but in the presence of the Kondo coupling only two states are relevant). In the limit $`J=\mathrm{}`$ the structure $`𝐒_{3Q}`$ gives rise to a spin-flux phase with the spectrum
$$ϵ_𝐤=\pm \frac{4}{\sqrt{3}}\sqrt{\mathrm{cos}^2\frac{k_x}{2}\mathrm{cos}^2\frac{k_y}{2}+\mathrm{sin}^2\frac{k_z}{2}\mathrm{cos}^2\frac{k_y}{2}+\mathrm{sin}^2\frac{k_z}{2}\mathrm{cos}^2\frac{k_x}{2}}$$
(8)
where the momenta are restricted to the region of the Brillouin zone where $`2\pi <k_z<0`$. Note that for $`n=0.5`$ the DOS is again linear in energy. For the structure $`𝐒_{1Q}`$ the dispersion is $`ϵ_k=4t\mathrm{cos}\frac{k_x}{2}\mathrm{cos}\frac{k_y}{2}`$. Assuming that one of these ground states is stable (that is taking $`J^{}`$ to be sufficiently large) then it is found that at $`n=1`$ the $`𝐒_{1Q}`$ state is stable while at $`n=0.5`$ the $`𝐒_{3Q}`$ state is stable. There is a transition between these two states at $`n=0.7`$. As in the case of the square lattice the $`𝐒_{3Q}`$ phase has no net current flowing about any closed loops on the lattice but it has spin currents flowing around the elemental triangular plaquettes that exist in planes with Miller indices $`(1,1,1)`$ (and equivalent symmetry planes) . The spin currents that flow correspond to a spin-flux of $`\pi /2`$ per triangular elemental plaquette (not $`\pi `$ per plaquette as was the case in the square lattice). It is intriguing to note that Hasegawa et al. have pointed out that for a triangular lattice the optimal flux per plaquette in a $`U(1)`$ flux phase is $`\pi /2`$ at $`n=0.5`$ . Both the $`𝐒_{1Q}`$ and the $`𝐒_{3Q}`$ states have been observed in $`\gamma `$-Mn alloys produced by doping with Fe, Ni, or Cu and it would be of interest to see if spin currents can be detected in the $`𝐒_{3Q}`$ phase of these materials.
In conclusion, we have given numerical evidence that a spin-flux phase exists as a ground state of the Kondo lattice model with classical localized spins on a square lattice. This phase gives rise to a spin-flux of $`\pi `$ for electrons circulating an elementary plaquette of the square lattice and manifests itself as a double-Q magnetic order in the classical spins. We have also proposed that a spin-flux phase may be stable on a FCC lattice. This phase manifests itself as a triple-Q magnetic order and gives rise to a spin-flux of $`\pi /2`$ for electrons circulating the elementary triangular plaquettes that lie in the planes with Miller indices $`(1,1,1)`$ (and equivalent symmetry planes).
The authors wish to thank J.R. Schrieffer, C. Buhler, A. Moreo, E. Dagotto, and T. Hotta for useful discussions. This work was supported by NSF DMR 9527035 and the State of Florida.
|
warning/0002/hep-th0002198.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
It was shown in (see also ) that, when we turn on a B field on the D-brane worldvolume, the low-energy effective worldvolume theory is modified to be a non-commutative Super-Yang-Mills (NCSYM) theory. In fact the worldvolume theory of $`N`$ coincident D<sub>p</sub>-branes in the presence of a B field is found to be $`U(N)`$ NCSYM theory .
As is the case with a zero B field, there exists a limit where the bulk modes decouple from the worldvolume non-commutative field theory ; we expect to have a correspondence between string theory on the curved background with B field and non-commutative field theories. In other words we expect to have a holographic picture like AdS/CFT correspondence (for review and complete list of references, see ) for the non-commutative theories. In fact this issue has been investigated in -.
It was also shown that there is a limit in which the theory living on the NS5-branes decouples from the bulk <sup>1</sup><sup>1</sup>1 This idea, that the theory on the NS5-branes could be some sort of string theory, has been also considered in in the context of Matrix theory description of M-theory and its compactifications.. It is believed that the decoupled theory is a string theory without gravity called “Little String Theory” (LST) <sup>2</sup><sup>2</sup>2 For a brief review, see .. Upon compactification this theory inherits the T-duality of Type II string theories and must therefore be a non-local theory. The simplest way to find this decoupling limit is to start with the decoupling of D5-branes and using S-duality. The decoupling limit of D5-branes is defined by $`l_s0`$ keeping $`g_{YM}^2=g_sl_s^2`$ fixed. Using S-duality <sup>3</sup><sup>3</sup>3Under S-duality we have $`l_s^2l_{}^{}{}_{s}{}^{2}g_sl_s^2`$ and $`g_sg_s^{}\frac{1}{g_s}`$ one finds
$$g_s^{}0g_{YM}^2=l_{}^{}{}_{s}{}^{2}=\mathrm{fixed}$$
(1)
which is the decoupling limit of NS5-branes . Using T-duality we can also obtain the limit in which the Type IIA NS5-branes decouple from the bulk, which is the same as the IIB one (1).
The NS5-branes theory, or LST, has been studied from several points of view; in particular it has been considered in using the holographic principle for string theory in an asymptotic linear dilaton background. There, the authors considered the string theory in the near-horizon background of parallel NS5-branes, which is dual to LST. This duality could help them to study some observables of the theory as well as some of their correlation functions. The theory on $`N`$ IIA NS5-branes has also been studied at large $`N`$ using supergravity in .
It is a natural question to ask what the non-commutative deformation of LST is. In the spirit of what we have learnt about D-branes in the presence of a B field and its relation with non-commutative gauge theory due to holography (AdS/CFT correspondence), we would expect that the non-commutative deformation of LST could be obtained by NS5-branes in the presence of a non-zero RR field along its worldvolume. In fact such a background has been considered in , and it is the aim of this letter to study this theory in more detail.
The non-commutative deformation of LST is also expected from the DLCQ description of LST ; there the authors showed that blowing up the singularities by FI terms can give, in the DLCQ context, a space-time interpretation that could be considered as turning on an RR field parallel to the NS5-branes. Therefore we would expect to find a non-commutative version of LST (NLST).
In this letter we study the theory living on the NS5-branes in the presence of an RR field at its decoupling limit. We shall only consider the case with the smallest rank of non-zero RR field. This theory can be considered as a non-commutative deformation of little string theory, although as we shall see the decoupling limits of non-commutative and ordinary LST are completely different. In particular in this case the decoupling limit is defined as a limit where the string scale goes to zero. Although even in this limit we have a scale in the theory (coming from the RR field), it is not clear whether the theory on the“non-commutative NS5-branes” obtained in this way is a string theory. In fact, as we shall see, the theory seems to be equivalent to the SYM theory. This may be because of our definition of the decoupling limit of the theory, but till now, it is the only consistent way this decoupling limit can be defined. Nevertheless this theory has some properties of LST; for example we would expect that the thermodynamical quantities of both theories be the same, both theories are non-local and have some sort of T-duality. It might also be possible that what we are studying here is not really a non-commutative deformation of LST. In other words, what we have obtained by S- or T-duality as a decoupling limit of the NS5-branes would be a limit that leads to a non-commutative field theory, such as SYM theory on the NS5-branes with an RR field, and in fact it might be that the non-commutative little string theory does not have a supergravity description<sup>4</sup><sup>4</sup>4This point was suggested by Y. Oz..
In section 2 we shall consider the Type IIB NS5-branes in the presence of an RR field and we will see that in the UV limit where non-commutative effects are important the theory can be described by smeared D3-branes. In section 3 we will study Type IIA NS5-branes. We find that the decoupling limit of the theory is consistent only when the theory is wrapped on a circle and this theory can be described at UV by smeared D2-branes. We will give a conclusion and some comments in section 4.
## 2 Non-commutative Type IIB NS5-branes
Decoupling limit
As for the ordinary NS5-brane, where there was a non-trivial theory on its worldvolume in its decoupling limit, we would expect to find a non-commutative version of little string theory on the worldvolume of the NS5-brane in the presence of a non-zero RR field in its decoupling limit. The simplest way to define this theory is to start with a D5-brane in Type IIB string theory in the presence of a non-zero B field along its worldvolume and using S-duality to map the theory to its S-dual. Doing so we end up with the NS5-brane solution with an RR field background.
The decoupling limit of D5-branes in the presence of a large B field is defined as a limit where $`l_s0`$ while $`g_s`$ and $`b=l_s^2B`$ are kept fixed; moreover, we have to rescale the directions in which the B field is defined, thereby making them non-commutative. In this limit the modes on the D5-brane decouple from modes on the bulk and we are left with 6-dimensional non-commutative SYM with gauge coupling $`g_{YM}^2g_sb`$. The phase diagram of this theory has been studied in .
Using S-duality we can find a decoupling limit in which we expect to have a decoupled theory on the worldvolume of the NS5-brane in the presence of an RR field:
$$l_s^{}0,b^{}=g_s^{}l_{}^{}{}_{s}{}^{2}A=\mathrm{fixed},g_s^{}=\mathrm{fixed}$$
(2)
where $`A`$ is the RR field obtained from a B field by S-duality. The coupling of the theory will be $`g_{YM}^2=\frac{b}{g_s^{}}=b^{}`$.
We note that the decoupling limit of the NS5-brane with the RR field is completely different from ordinary NS5-brane (1). In fact, in both the ordinary case and the case with an RR field, what we want to send to zero is $`g_s^{}l_{}^{}{}_{s}{}^{2}`$, but the important point is which quantity we would like to keep fixed. Since we want to have a non-trivial theory at the decoupling limit, it is natural to assume that the coupling of the theory should be fixed. In general for an RR field with rank $`2m`$, the coupling is proportional to $`g_s^ml_s^{22m}`$. Now we can see that adding a non-zero RR field will change the decoupling limit in a completely non-trivial way.
We also note that the decoupling limit of NLST looks very much like ordinary D3-branes. We will go back to this point later.
Supergravity description
We can study $`N`$ coincident NS5-branes in the presence of an RR field at large $`N`$ using supergravity. In order to study this theory one can start with the supergravity solution of D5-branes with a B field and using S-duality. The supergravity solution of D5-branes in the presence of a rank-two B field is given by:
$`ds^2`$ $`=`$ $`f^{1/2}[dx_0^2+\mathrm{}+h(dx_4^2+dx_5^2)]+f^{1/2}(dr^2+r^2d\mathrm{\Omega }_3^2),`$ (3)
$`f`$ $`=`$ $`1+{\displaystyle \frac{Ng_sl^2}{\mathrm{cos}\theta r^2}},h^1=\mathrm{sin}^2\theta f^1+\mathrm{cos}^2\theta ,`$ (5)
$`B_{45}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}\theta }{\mathrm{cos}\theta }}f^1h,e^{2\varphi }=g_s^2f^1h.`$ (7)
Under S-duality we have
$$e^\varphi e^\varphi ^{}e^\varphi ,ds^2ds^2g_se^\varphi ds^2.$$
(8)
Using (8) we get the Type IIB NS5-branes background :
$`ds_{}^{}{}_{}{}^{2}`$ $`=`$ $`h^{1/2}[dx_0^2+\mathrm{}+h(dx_4^2+dx_5^2)+f(dr^2+r^2d\mathrm{\Omega }_3^2)],`$ (9)
$`f`$ $`=`$ $`1+{\displaystyle \frac{Nl_{}^{}{}_{s}{}^{2}}{\mathrm{cos}\theta r^2}},h^1=\mathrm{sin}^2\theta f^1+\mathrm{cos}^2\theta ,`$ (11)
$`e^{2\varphi ^{}}`$ $`=`$ $`g_{}^{}{}_{s}{}^{2}fh^1,`$ (13)
and the NS field $`B_{ij}`$ is mapped to the RR field $`A_{ij}`$. The decoupling limit is derived by applying S-duality on the decoupling limit of the D5-branes. As we said above, it is defined by taking the limit $`l_{}^{}{}_{s}{}^{2}0`$ and keeping fixed
$$\begin{array}{cc}u=\frac{r}{l_{}^{}{}_{s}{}^{2}}\hfill & \overline{g}_{}^{}{}_{s}{}^{}=g_{}^{}{}_{s}{}^{1}\hfill \\ b^{}=l_{}^{}{}_{s}{}^{2}\mathrm{tan}\theta \hfill & \overline{x}_{4,5}=\frac{b^{}}{l_{}^{}{}_{s}{}^{2}}x_{4,5}.\hfill \end{array}$$
(14)
Keeping $`u`$ fixed means keeping fixed the mass of a D-string stretched between two NS5-branes. In this limit the supergravity solution reads
$`ds_{}^{}{}_{}{}^{2}`$ $`=`$ $`{\displaystyle \frac{l_{}^{}{}_{s}{}^{2}}{b^{}}}h^{1/2}\left[dx_0^2+\mathrm{}+h^1(dx_4^2+dx_5^2)+{\displaystyle \frac{Nb^{}}{u^2}}(du^2+u^2d\mathrm{\Omega }_3^2)\right],`$ (15)
$`h`$ $`=`$ $`1+(au)^2,a^2={\displaystyle \frac{b^{}}{N}},e^{2\varphi ^{}}=g_s^2{\displaystyle \frac{1+(au)^2}{(au)^2}}.`$ (17)
The curvature of the metric reads
$$l_{}^{}{}_{s}{}^{2}\frac{1}{N}\frac{1}{(1+(au)^2)^{1/2}}.$$
(18)
When $`au1`$, which we are interested in, the supergravity approximation can be trusted for finite $`N`$, which means that we can study NLST even for finite $`N`$. In this limit the background reads
$`l_{}^{}{}_{s}{}^{2}ds_{}^{}{}_{}{}^{2}`$ $`=`$ $`{\displaystyle \frac{u}{R^2}}(dx_0^2+\mathrm{}+dx_3^2)+{\displaystyle \frac{R^2}{u}}(du^2+u^2d\mathrm{\Omega }_3^2)+{\displaystyle \frac{R^2}{b_{}^{}{}_{}{}^{2}u}}(dx_4^2+dx_5^2)`$ (19)
$`e^{2\varphi ^{}}`$ $`=`$ $`\overline{g}_s^2,R^2=\sqrt{Nb^{}},`$ (20)
and the curvature of the metric is $`l_{}^{}{}_{s}{}^{2}\frac{1}{R^2u}`$. From the dilaton we can see that this solution is similar to D3-branes and it is in fact the D3-branes solution smeared in two directions and without RR field.
The same situation has also been studied for D<sub>p</sub>-branes in . There, it was shown that the non-commutative $`(p+1)`$-dimensional SYM theory can be considered as $`(p1)`$-dimensional ordinary YM theory whose gauge group is obtained by the B field. In fact what the authors have shown is as follows: they observed that for D<sub>p</sub>-branes in the presence of a B field with rank two at UV, where the non-commutative effects are important, the supergravity solution reduces to D<sub>(p-2)</sub>-branes smeared in two directions without B field. In fact this was the case because in this limit the physics is described by D<sub>(p-2)</sub>-branes; since in this case the B field that we started with is not along the world-volume of the D<sub>(p-2)</sub>-branes, it can be gauged away, and we end up with an ordinary smeared brane.
In our example the situation is the same as D5-branes where the theory can be described by smeared D3-branes, but in our case we have to gauge away the RR field instead of the B field, which is made possible by the $`SL(2,Z)`$ symmetry of Type IIB string theory. We also note that the smeared D3-brane is also self-dual under S-duality. The reason is that, in smeared D3-branes, only the harmonic function of the metric will change and the dilaton and 4-form field will be the same as localized D3-branes. Therefore we would expect that the $`SL(2,Z)`$ symmetry maps the smeared D3-branes to itself.
Since both the non-commutative D5-branes theory and the non-commutative NS5-branes can be described by ordinary D3-branes smeared in two directions, and moreover, that this D3-brane solution is self-dual under S-duality, we conclude that these two theories must be the same.
For $`au1`$, as long as $`aug_s^{}`$ we can still trust the NS5-branes solution. In this regime, setting $`au=e^{\mathrm{\Phi }/\sqrt{Nb^{}}}`$ we have:
$$ds_{}^{}{}_{}{}^{2}=\frac{l_{}^{}{}_{s}{}^{2}}{b^{}}[dx_{||}^2+d\mathrm{\Phi }^2+Nb^{}d\mathrm{\Omega }_3^2],g_s(\mathrm{\Phi })=g_{}^{}{}_{s}{}^{}e^{Q\mathrm{\Phi }}$$
(21)
where $`Q=\frac{1}{\sqrt{Nb^{}}}`$.
Finally, for the case where $`aug_s^{}`$, we have to use the S-dual picture, which maps the theory to an ordinary commutative D5-branes.
Using the same variable as above, solution (17) can be written as follows:
$`{\displaystyle \frac{b}{l_{}^{}{}_{s}{}^{2}}}ds^2`$ $`=`$ $`k^{1/2}(dx_4^2+dx_5^2)+k^{1/2}(dx_{||}^2+d\mathrm{\Phi }^2+Nb^{}d\mathrm{\Omega }_3^2)`$ (22)
$`g^2(\mathrm{\Phi })`$ $`=`$ $`g_{}^{}{}_{s}{}^{2}e^{2Q\mathrm{\Phi }}k,k=1+e^{2Q\mathrm{\Phi }}.`$ (23)
From this form of the solution, one can see the deformation of the linear dilaton background manifestly. The differential equation for the scalar in this background, setting $`\mathrm{\Psi }=\psi e^{i\omega t}`$, is:
$$_\mathrm{\Phi }^2\psi +2Q_\mathrm{\Phi }\psi +\omega ^2\psi =0.$$
(24)
This equation has a wave-like solution if $`\omega >Q`$. Actually one can use this equation to study the absorption cross section of the polarized graviton in this background. Doing so, we find that the absorption cross section can be non-zero at the decoupling limit for an energy
$$\omega >\frac{b_{}^{}{}_{}{}^{1/2}}{\sqrt{N}}.$$
(25)
We note that this is very similar to what was found in for Type IIA NS5-branes. This relation, together with what we have had till now, could suggest that $`b^{}`$ (or $`b`$) has a role of scale in non-commutative NS5-branes theory. From the smeared D3-branes point of view, this can be considered as a scale that measures the smeared directions
We can also calculate the Wilson loop (or ’t Hooft loop) for the smeared D3-branes that appeared above. Doing so we will find the same problem as ordinary D5-branes namely that the distance $`L`$ between quark and antiquark does not depend on $`u_0`$ (where $`u_0`$ is the minimal value of $`u`$). In fact we have
$$L\sqrt{b^{}}\sqrt{N}.$$
(26)
Since this system is equivalent to a non-local theory, this classical minimal distance should be related to the scale of non-locality and in fact this is the case, except for the fact that it is larger by a factor of $`\sqrt{N}`$ than the naively expected scale which was already mentioned in .
Following we can also study the T-duality of the theory on the D5-branes in the presence of a B field as well as non-commutative NS5-branes, which has to be related to the Morita equivalent of the non-commutative theories on the compact space.
## 3 Non-commutative Type IIA NS5-branes
Decoupling limit
The decoupling limit of the Type IIA NS5-branes can be obtained by T-duality from the Type IIB one. Unlike the ordinary case, where the decoupling limit for both IIB and IIA was the same, here we actually find that these limits are different. The decoupling limit of the Type IIA NS5-brane in the presence of a non-zero RR field along its worldvolume is defined as follows:
$$l_s0,l_s^3A=\mathrm{fixed},g_sl_s^1=\mathrm{fixed}$$
(27)
where $`A`$ is the RR field. As we see, this decoupling limit is completely different from the one we have in the ordinary NS5-branes and in fact it is exactly the decoupling limit of smeared ordinary D2-branes. We will go back to this point later. The power of 3 in the fixed quantity $`l_s^3A`$ can be understood by the fact that in Type IIA we are dealing with an RR 3-form. We also note that this decoupling limit is different from the one considered in , where the same theory has been considered in the DLCQ context; it is not clear to me whether these two definitions are related or if they lead to two different deformations of the NS5-branes theory. <sup>5</sup><sup>5</sup>5I would like to thank O. Aharony for a discussion on this point.
Supergravity description
The supergravity solution of the Type IIA NS5-branes in the presence of a non-zero RR field is as follows:
$`ds`$ $`=`$ $`h^{1/2}[dx_0^2+dx_{1,2}^2+hdx_{3,4,5}^2+f(dr^2+r^2d\mathrm{\Omega }_3^2)]`$ (28)
$`f`$ $`=`$ $`1+{\displaystyle \frac{Nl_s^2}{\mathrm{cos}\theta r^2}},h^1=\mathrm{sin}^2\theta f^1+\mathrm{cos}^2\theta ,`$ (30)
$`e^{2\varphi }`$ $`=`$ $`g_s^2fh^{1/2},A_{345}={\displaystyle \frac{\mathrm{tan}\theta }{g_s}}f^1h`$ (32)
$`A_{012}`$ $`=`$ $`{\displaystyle \frac{\mathrm{sin}\theta }{g_s}}f^1.`$ (33)
The decoupling limit is defined by taking the limit $`l_s0`$ and keeping fixed
$$\begin{array}{cc}u=\frac{r}{l_s^2},\hfill & \overline{g}_s=g_sl_s^1,\hfill \\ b=l_s^2\mathrm{tan}\theta ,\hfill & \overline{x}_{3,4,5}=\frac{b}{l_s^2}x_{3,4,5},\hfill \end{array}$$
(34)
From the NS5-branes point of view, we would expect, in the decoupling limit, the mass of a D2-brane stretched between two NS5-branes to be fixed. If it had been the case, we would have had $`\frac{r}{g_sl_s^3}`$ fixed. But here, in the decoupling limit, we have $`\frac{r}{l_s^2}`$ fixed, which means that we are dealing with wrapped NS5-branes. On the other hand, this is the only decoupling limit of NS5-branes that is consistent with string theory dualities. Therefore what we are really studying in this section is non-commutative NS5-branes theory wrapped on a circle. This can also be understood from the T-duality we used to find the decoupling limit. As a result, beside the fixed quantities defined in (34), we also have $`\frac{R_s}{g_sl_s}`$ fixed (here $`R_s`$ is the radius of the compacted direction). This extra condition will play an interesting role in the phase diagram of the theory. Moreover, there is another parameter in the theory, which we have to take into account: $`\beta =g_s^2b`$ .
In the decoupling limit the supergravity solution reads:
$`ds`$ $`=`$ $`{\displaystyle \frac{l_s^2}{b}}h^{1/2}\left[dx_0^2+dx_{1,2}^2+h^1dx_{3,4,5}^2+{\displaystyle \frac{Nb}{u^2}}(du^2+u^2d\mathrm{\Omega }_3^2)\right]`$ (35)
$`h`$ $`=`$ $`1+(au)^2,a^2={\displaystyle \frac{b}{N}},e^{2\varphi }=\overline{g}_s^2b{\displaystyle \frac{\sqrt{1+(au)^2}}{(au)^2}}`$ (37)
$`A_{012}`$ $`=`$ $`{\displaystyle \frac{l_s^3}{\overline{g}_sb^2}}(au)^2,A_{345}={\displaystyle \frac{l_s^3}{\overline{g}_sb^2}}{\displaystyle \frac{(au)^2}{1+(au)^2}}.`$ (39)
An important point for this theory is that, in the limit where non-commutative effects are not important and where, moreover, we have to lift the theory to M-theory, the extra condition mentioned above means $`\frac{R_s}{R_{11}}`$ fixed, with $`R_{11}`$ the radius of the 11th direction. This means that these circles have to be in the same order or, in other words, that these two circles will decompactify at the same energy. Therefore in the extreme IR we will end up with non-compact (0,2) theory. On the other hand, in the extreme UV limit where the non-commutative effects are important, $`au1`$, the solution reduces to:
$`l_s^2ds`$ $`=`$ $`{\displaystyle \frac{u}{R^2}}(dx_0^2+dx_{1,2}^2)+{\displaystyle \frac{R^2}{u}}(du^2+u^2d\mathrm{\Omega }_3^2)+{\displaystyle \frac{R^2}{b^2u}}dx_{3,4,5}^2`$ (40)
$`R^2`$ $`=`$ $`\sqrt{Nb},e^{2\varphi }=\overline{g}_s^2{\displaystyle \frac{R^2}{u}}.`$ (42)
Moreover, we have a non-zero 3-form along the directions (0,1,2) and (3,4,5), the latter of which can be gauged away. Therefore we end up with an ordinary D2-branes solution smeared in three directions<sup>6</sup><sup>6</sup>6The curvature of the metric is $`l_s^2\frac{1}{R^2u}`$; therefore one can trust the solution at UV. We note that this is in contrast with what we have for non-smeared D2-branes where, at UV, we have perturbative SYM theory.. Actually we could reach the same conclusion from Type IIB NS5-branes as from using T-duality as we did.
As the energy starts increasing from the IR limit, there are two possibilities. If $`\beta 1`$, first we reach a regime where we have the wrapped M5-branes in the presence of a C field here the effects of the C field are important and we then have to go to the Type IIA theory where the theory is described by smeared D2-branes. On the other hand, if $`\beta 1`$, we have to go to the Type IIA description, where the non-commutative effects are not important and we can trust the supergravity solution of NS5-branes; in fact, setting $`au=e^{\mathrm{\Phi }/\sqrt{Nb}}`$, we reach the same linear dilaton regime as in the Type IIB case
$$ds^2=\frac{l_s^2}{b}[dx_{||}^2+d\mathrm{\Phi }^2+Nbd\mathrm{\Omega }_3^2],g_s^2(\mathrm{\Phi })=\overline{g}_s^2be^{2Q\mathrm{\Phi }},$$
(43)
and finally we will reach at UV the regime that is described by smeared D2-branes.
As in the Type IIB case we can write the solution in terms of the new variable defined above
$`{\displaystyle \frac{b}{l_{}^{}{}_{s}{}^{2}}}ds^2`$ $`=`$ $`k^{1/2}(dx_3^2+dx_4^2+dx_5^2)+k^{1/2}(dx_{||}^2+d\mathrm{\Phi }^2+Nb^{}d\mathrm{\Omega }_3^2)`$ (44)
$`g^2(\mathrm{\Phi })`$ $`=`$ $`\overline{g}_s^2be^{2Q\mathrm{\Phi }}k^{1/2},k=1+e^{2Q\mathrm{\Phi }},`$ (45)
which shows the deformation of linear dilaton background in the presence of an RR field.
One can also study this system using M5-branes in the presence on a non-zero C field. Using the solution of M5-branes in the presence of a non-zero C field , in the near-horizon region we have
$`ds^2`$ $`=`$ $`(NA)^{2/3}(1+\eta A^1)^{1/3}l_p^2\left[A^1(dy_{0,1,2}^2+{\displaystyle \frac{dy_{3,4,5}^2}{1+\eta A^1}})+d\chi ^2+d\rho ^2+\rho ^2d\mathrm{\Omega }_3^2\right]`$ (46)
$`A`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{n=\mathrm{}}{}}}{\displaystyle \frac{\pi }{[\rho ^2+(\chi 2\pi n)^2]^{3/2}}},\rho ={\displaystyle \frac{r}{R_{11}}},\chi ={\displaystyle \frac{x_{11}}{R_{11}}};`$ (48)
moreover, we rescaled the coordinates along the branes by $`y_{0,1,2}=\sqrt{\frac{\mathrm{cos}\theta }{l_s^2N}}x_{0,1,2}`$ and $`y_{3,4,5}=\sqrt{\frac{1}{l_s^2N\mathrm{cos}\theta }}x_{3,4,5}`$. The non-commutative NS5-branes solution is the regime where $`\rho 1`$, while $`\rho 1`$ is the M5-branes geometry. We note that $`\eta =\frac{R_{11}^2}{l_s^2N\mathrm{cos}\theta }`$ is equal to $`\frac{\beta }{N}`$ at the decoupling limit; as we can see from the solution for $`\eta 1`$ we will reach the linear dilaton regime before we go to the non-commutative NS5-branes regime, but for $`\eta 1`$ we will reach the non-commutative NS5-branes regime when we go to the Type IIA description.
This solution can be used in order to study the absorption cross section of the polarized graviton along the branes. Using the same notation as in , the equation for minimally scalar of the form $`\mathrm{\Psi }=\psi e^{i\sqrt{s}y_0}`$ reads
$$_\chi ^2\psi +\frac{1}{\rho ^3}_\rho \rho ^3_\rho \psi +\underset{n=\mathrm{}}{\overset{n=\mathrm{}}{}}\frac{s\pi }{[\rho ^2+(\chi 2\pi n)^2]^{3/2}}\psi =0,$$
(49)
where $`s=\frac{\omega ^2Nl_s^2}{\mathrm{cos}\theta }`$. From this notation one can use the results of ; in particular we can compute the two-point functions of the theory for the operator that couples to the graviton, which can be interpreted as a component of the energy-momentum tensor of the theory. In fact the results are the same but of course with our $`s`$ defined above. We also note that as the ordinary NS5-branes, the absorption cross section can be non-zero at the decoupling limit for the case $`s>1`$, or for the energy larger than $`\omega >\frac{b^{1/2}}{\sqrt{N}}`$.
This result is the same as for the ordinary case except, that $`b`$ plays the role of the scale of the theory as expected. From this scale we would expect the mass gap that could appear at $`s=1`$, as the ordinary NS5-branes, to be of the order of this scale;, however it is smaller by a factor of $`\sqrt{N}`$. Actually this is also the case in the ordinary NS5-branes; while the scale of the theory is $`m_s`$, the mass gap is of the order of $`\frac{m_s}{\sqrt{N}}`$. We also note that, as we saw in the Type IIB case, there is a factor $`\sqrt{N}`$ difference between the scale of non-locality, which we see from gravity calculation, and what is expected from world-sheet calculation.
We would also like to note that the same phenomenon has been observed for the Coulomb branch of the $`𝒩=4`$ SYM theory in 4 dimensions, where the mass gap is smaller than expected, from a gauge theory point of view, by a factor of $`\sqrt{g_{YM}^2N}`$ . This fact, together with our previous observation, might mean that there are some stringy effects that change the behaviour of the theory at strong coupling.
## 4 Discussion
We could consider NS5-branes in the presence of a non-zero RR field with higher rank. In this case we would find non-commutative NS5-branes in their decoupling limit. These solutions have been given in . Here we were only interested in the case where, at UV, we can trust the NS5-branes solution; in other words we are interested in the case where, at UV, the theory is thought to be a non-commutative version of the little string theory. In fact this is only the case when we have the smallest rank for the RR field. This can be seen for example from the phase diagram of D5-branes with a higher-rank B field . So in this paper we only considered the solution with smallest rank.
Another point that we observed is that the non-commutative little string theories can be described by ordinary branes smeared in some directions. For example the non-commutative Type IIB NS5-branes can be described by the ordinary D3-branes smeared in two directions. If this is really the case, it means that the theory must be equivalent to a non-commutative SYM theory with 16 supercharges in 6 dimensions. On the other hand we know that ordinary SYM theory in 6 dimensions is not renormalizable; in other words, in order to define the theory we need some more information at UV. A way to solve this problem is to assume that the theory flows to a fixed point at UV where we have the little string theory. Now, from our discussion we see that there is another way to make the theory to be well-defined. In this way the theory flows to a non-commutative theory, which can be considered as either non-commutative 6-dimensional SYM theory or non-commutative NS5-branes theory. So non-commutativity gives us another way to make the theory well-defined. This is also the case for the 5-dimensional theory, where there were two possibilities: either it flows to a fixed point at UV ((0,2) theory) or it has a non-commutative 5-dimensional description at UV.
We found that the decoupling limit of the Type IIA NS5-branes can be defined only when the theory is wrapped on a circle. This is also consistent with our knowledge of a possible non-commutative version of NS5-branes, since this theory is only defined in the DLCQ context, where one direction is automatically wrapped on a circle . Nevertheless, the theory flows to the ordinary non-compact (0,2) theory at the IR limit.
We also note that the relation between non-commutative D5-branes, NS5-branes and D3-branes smeared in two directions might give us some information about the Coulomb branch of the $`𝒩=4`$ SYM theory in four dimensions .<sup>7</sup><sup>7</sup>7This issue has been suggested by A. Giveon
Acknowledgements
I would like to thank O. Aharony, A. Brandhuber, A. Giveon, A. Kumar and G. Mandal for useful discussions. I am also indebted to Y. Oz for interesting discussions, which motivated me to study this subject, as well as for his comments. The work was partially supported by H. Hofer.
|
warning/0002/nucl-th0002015.html
|
ar5iv
|
text
|
# A phenomenological analysis of antiproton interactions at low energies
## 1 Introduction
Recently data on $`\overline{p}`$ annihilations on light nuclei (H, D and <sup>4</sup>He) have become available at very small $`\overline{p}`$ momenta (down to 45 MeV/c). A new data on <sup>20</sup>Ne at 57 MeV/c is also available now. Together with previously available data (for a review see e.g. ref.), and with data on antiprotonic atoms, the full set presents some interesting features, that we will try and correlate in this work. As far as a qualitative physical understanding is concerned, the unifying feature is a mechanism that we call “inversion”, i.e. a repulsion-dominated low energy $`\overline{p}p`$ interaction. From a practical point of view, we will widely rely on the possibility of reproducing the available elastic and annihilation $`\overline{p}p`$ data below 600 MeV/c via an energy independent optical potential.
Let us initially discuss some relevant points of the phenomenology:
1) Annihilation $`\overline{p}p`$ data show, in a log-log plot, a series of roughly rectilinear behaviors (see fig.1). These can be approximately identified with regions where different angular momentum components are dominant, with the S-P transition at about 100 MeV/c. At 50 MeV/c it is possible to assume S-wave dominance and estimate the imaginary part of the scattering length $`\alpha `$. The real part, is extracted from the widths and shifts of the levels of antiprotonic Hydrogen atoms, together with an independent measure of $`Im(\alpha )`$. Elastic $`\overline{p}p`$ data at values of the laboratory $`\overline{p}`$ momentum $`k`$ in the 200-500 MeV/c range were reproduced by Brückner $`et`$ $`al`$ with a phenomenological optical potential. These authors left a wide range of uncertainty for the suggested potential parameters. We have noticed that the same potential, with a finer tuning of the parameters, can fit all the annihilation data which have been later measured at smaller $`k`$, down to 30 MeV/c, by the Obelix Collaboration (see fig.2, and section 3 for details). It can also calculate the real and imaginary parts of the scattering length. Optical potential analysis, partial wave analysis and atomic data agree on $`Re(\alpha )`$ $``$ $`Im(\alpha )`$ $``$ 0.7$`÷`$0.8 fm, with positive sign.
2) The $`\rho `$ parameter, i.e. the ratio between the real and imaginary part of the forward scattering amplitude, can be measured at zero or near zero energy exploiting $`\rho (0)`$ $`=`$ $`Re(\alpha )/Im(\alpha )`$, which means $`\rho `$ $``$ $`1`$. At larger energy, it must be extracted by a $`very`$ delicate (and partly model dependent) analysis of the elastic $`\overline{p}p`$ angular distributions. Despite the behavior of $`\rho `$ is still unclear in the region 100-200 MeV/c, an overview of the experimental data suggest that $`\rho `$ is small but positive (0.1$`÷`$0.3) at projectile momenta over some value which lies somewhere around 500 MeV/c, smaller (with uncertain sign) in the region 180-500 MeV/c, and tends to some negative value $``$ $`1`$ at zero energy. As better described in the following, we have applied the optical potential (whose parameters have been fine-tuned on the $`\overline{p}p`$ annihilation data at 30-100 MeV/c), to predict the $`\rho `$ behavior. The results agree with the large and the zero energy data, and suggest that $`\rho `$ varies monotonously in the less known intermediate momentum region (see fig.3).
3) In the laboratory frame $`\overline{p}`$ total annihilation cross sections (TPA from now on) on Deuteron and <sup>4</sup>He are almost equal, and both are smaller than TPA on Hydrogen. The <sup>20</sup>Ne datum is larger, but not so large as one could expect. See fig.1 for a general view of the data. Taking into account that a large enhance of reaction cross sections is predicted at low energies because of charge effects, this phenomenon is surprising. According with the notations used in previous works we will call this behavior “inversion”. Actually, if the data are represented in the center of mass frame, TPA on D and <sup>4</sup>He are slightly larger than TPA on Hydrogen, however the dependence of the TPA on the mass number is still much smaller than any geometrical expectance (see later for a discussion of the “geometrical expectance” and of the role of the center of mass). For $`k`$ $`>>`$ 100 MeV/c this phenomenon is not observed and the ratio between $`\overline{p}p`$ and $`\overline{p}^4`$He annihilation rates is qualitatively what one would expect. The inversion behavior is confirmed by an analysis of antiprotonic atoms, where it is found that $`|Im(\alpha )|`$ is smaller in antiprotonic Deuterium than in antiprotonic Hydrogen.
4) From an overview of the available $`\overline{p}nucleus`$ and $`\overline{n}nucleus`$ annihilation data below 600 MeV/c it appears that: (i) Where many partial waves dominate the $`\overline{p}`$nucleus interaction the cross sections relative to different nuclear species are parallel, and agree with a law $`\sigma `$ $``$ $`\sigma _oA^{2/3}`$. (ii) Where only a few partial waves are supposed to dominate, a convergency (for decreasing energies) between the different TPA is clearly visible. In a log-log plot, the extrapolations of the different TPA seem to aim at some common intersection point somewhere at $`k_{cm}`$ $``$ 1 MeV/c (see fig.1).
## 2 General theoretical background.
To better understand the significance of the previous nuclear data some considerations are useful. Both the “inversion” and the convergency behavior contraddict the geometrical predictions. Assuming that the imaginary part of the scattering length is roughly equal to the nuclear size $`R`$ $``$ 1.3$`A^{1/3}`$ fm, and exploiting the traditional estimation of the Coulomb focusing effect, one has TPA $``$ $`ZA^{1/3}/k^2`$ at very small momenta. At larger momenta the semiclassical expectance is TPA $``$ $`A^{2/3}`$ (well verified for $`\overline{p}`$ and $`\overline{n}`$ annihilations at any $`k_{lab}`$ $`>`$ 180 MeV/c). Since for most nuclei $`ZA^{1/3}`$ $``$ $`0.5A^{4/3}`$, one should naively expect that TPA on different nuclei increase their separation when momenta decrease below 100 MeV/c, while exactly the opposite in seen. In addition, at any precise lab or c.m. momentum below $`k_{lab}`$ $`=`$ 100 MeV/c, the $`A`$dependence of the known TPA is below both the $`A^{2/3}`$ and the $`ZA^{1/3}`$ prediction.
Regarding the question whether the TPA on different nuclei must be compared at the same laboratory or center of mass momenta, the answer is model-dependent. In Impulse Approximation inspired models, the annihilation process only involves one of the nucleons in the target nucleus, which has average momentum equal to zero in the laboratory. It is then reasonable to compare data at the same laboratory momentum. In compound-nucleus inspired models, the collision process directly transfers momentum from the projectile to the full target. In this case, data taken on different targets should be compared at the same c.m. momentum.
The key point is the generalization of the concept of low energy “inversion”. On the ground of general quantum principles it is possible to demonstrate that, in presence of a $`very`$ effective esothermic hadronic reaction mechanism and in conditions of S-wave dominance: (i) the reaction cross section must stay much below geometrical expectations, and is largely independent on the target nucleus size; (ii) most attempts to increase those model parameters which supposedly should enhance the annihilation rate (e.g. strength or radius of a potential) lead to the opposite or to no result; (iii) a strong non-diffractive elastic scattering accompanies the reaction at low energies, and this scattering has $`repulsive`$ character (i.e. $`Re(\alpha )`$ $`>`$ 0). So, with “inversion” we will refer to the presence of these three features. We have previously demonstrated that strong inversion must be expected whenever disappearance of the projectile S-wave wavefunction $`\mathrm{\Psi }_S`$ (at the nuclear surface) is produced within a range much smaller than $`\mathrm{\Delta }r`$ $``$ 1 fm. Then, regularity conditions on $`\mathrm{\Psi }_S`$ at the nuclear surface necessarily produce a large flux reflection and a $`\mathrm{\Psi }_S`$ which is similar to the one produced by a repulsive potential with little absorbtion. For this reason it is not proper to consider the scattered flux as “diffractive”, although it is a by-product of absorbtion. It is a refractive process, as in elastic potential scattering. Now we can better specify the above required condition of “very effective reaction mechanism” (since at low energies it is not so effective): It means that (i) the reaction is esothermic, (ii) it produces large reaction rates at large energies, (iii) at any energy its free mean path in nuclear matter can be estimated to be shorter than 1 fm. We remark that the described behavior is experimentally confirmed by the fact that for the $`\overline{p}p`$ scattering length $`\alpha `$ we have $`Re(\alpha )`$ $``$ $`Im(\alpha )`$ $`>`$ 0, or equivalently $`\rho (0)`$ $``$ $`1`$. And by the fact that $`\overline{p}`$ annihilation rates on nuclei are not that large.
Also the traditional view of the Coulomb focusing effect must be reconsidered. In a previous paper we have already calculated and compared “charged” and “uncharged” annihilation rates on nuclei with finite size, and demonstrated that the traditional $`Z/\beta `$ Coulomb enhancement factor is exagerated. This factor is estimated with the two assumptions: (i) pointlike target (ii) completely independent action of Coulomb and strong forces. On the contrary, on one side the interplay between Coulomb and strong forces is not negligible, and on the other side finite size effects largely neutralize the Coulomb enhancement factor for intermediate and heavy nuclear targets. E.g., speaking in terms of target effective charge $`Z_e`$, we have $`Z_e(^4He)/Z_e(H)`$ $``$ 1 (instead of 2), $`Z_e(^{20}Ne)/Z_e(^4He)`$ $``$ 2 (instead of 5; comparisons are performed at the same laboratory momentum, but center of mass effects were included in the calculation).
## 3 Optical potential fits on $`\overline{p}p`$ data
.
As previously anticipated, all the data on $`\overline{p}p`$ elastic and annihilation cross section below 600 MeV/c can be reasonably well fitted by the same potential, with Woods-Saxon shape, used by Brückner $`et`$ $`al`$ to fit elastic $`\overline{p}p`$ data at 181, 287 and 505 MeV/c, after a finer tuning of the parameters. We have set the real and imaginary strength to -46 and -8000 MeV, the real and imaginary radius to 1.89 and 0.41 fm, and the diffuseness to the common value 0.2 fm. The fit on the annihilations is very good below 300 MeV/c and good within 10 % at 600 MeV/c (the exact precision over 300 MeV/c depends on which set of data is chosen), and the elastic distributions are still well reproduced. The total potential includes the Coulomb potential of a spherical charge distribution with radius 1.25 fm. In all the calculations center of mass corrections have been included. Together with the outcome of the above potential, in fig.2 we also show a curve corresponding to imaginary strength 1000 MeV. For both cases (strength 8000 and 1000 MeV) we also show the S- and P-wave contributions. Evidently the used potential does produce “inversion”, i.e. a larger annihilation potential produces a smaller annihilation rate. From the same figure it is obvious that this behavior is associated with the S-wave dominance, and is present only below an “inversion point” $`k_{inv}`$. In this case $`k_{inv}`$ $``$ 200 MeV/c.
In fig.3 we show the value of $`\rho `$ in the momentum range 0-600 MeV/c calculated with this potential. The change of sign of $`\rho `$ can be related with the transition from the dominance of the reaction-associated repulsion to the dominance of the direct potential attraction, at least in forward scattering. Indeed, at increasing momenta the Born approximation becomes progressively more reliable, and it permits to estimate $`\rho `$ $``$ $`(V_RR_R^3)/(V_IR_I^3)`$ $``$ $`+`$0.2, using as an effective radius the sum of the potential radius and diffuseness. The positive $`\rho `$ value at large momenta is thus directly due to the presence of a real attracting part in the potential. We notice that the “source” of the “direct” attraction will be the region where absolute value of the elastic potential is roughly equal to the kinetic energy, while the “source” of the reaction-induced repulsion will be the region where most annihilations take place, i.e. 0.5$`÷`$1 fm out of the edge of the annihilation core This distance has been estimated in past years in analysis of both $`\overline{p}p`$ and $`\overline{p}`$nucleus interactions.
In fig.4 the total annihilation and elastic cross sections are reported, compared with the corresponding cross sections calculated after turning off the electric charge. In the former case the contribution of the pure Coulomb forward peak and of the Coulomb-strong interference is excluded. Nevertheless, the elastic strong cross section is largely affected by Coulomb focusing effects. In particular, the figure shows that the ratio between the strong elastic and the annihilation total cross sections is completely dominated by the Coulomb effects. Without them, $`\sigma _{el}/\sigma _a`$ $``$ 0 for $`k`$ $``$ 0. With inclusion of the charge effect, approximately $`\sigma _{el}/\sigma _a`$ $``$ 1/6. We have also calculated angular distributions at momenta between 25 and 100 MeV/c, but they are practically flat up to 50 MeV/c, and at 100 MeV/c present a 20% change between forward and backward scattering, so they are not very interesting. At 100 MeV/c the P-wave contributions are 1 % in the total strong elastic cross section, and 10% in the annihilation. We remark that at such small momenta the Rutherford “forward” peak, which spreads at angles $`\theta `$ $``$ $`1/k`$, becomes the most important source of elastic scattering at large angles too.
## 4 Annihilations on nuclei.
Up to now we did not succeed in fitting light nuclei data perfectly by energy-independent optical potentials (which take nuclear density distributions into account). In fact, at momenta below 100 MeV/c a certain energy dependence is introduced by the nontrivial energy dependence of the $`\overline{p}n`$ annihilation rate). The study of the nuclear optical potential requires taking into account nuclear structure details and $`\overline{p}n`$ interactions, so a more specific and longer work will be devoted to it in the next future. Qualitatively, it is evident that the energy dependence of the cross sections in the range 30-200 MeV/c is much slower in $`\overline{p}`$nucleus than in $`\overline{p}p`$. This can be related to the change of sign of $`\rho `$ in $`\overline{p}p`$ interactions observing that if the $`\overline{p}`$nucleon interaction is repulsive below a certain momentum of scale $``$ 100 MeV/c, in a cluster of nucleons each single nucleon will contribute keeping the projectile far from itself and from all the other ones. In the language of the multiple scattering expansion this is an interference between single and double scattering processes, i.e. elastic scattering of $`\overline{p}`$ on one nucleon prevents annihilation on another one. This interpretations would confirm the suggestion given by Wycech $`et`$ $`al`$ in their analysis of antiprotonic deuterium: they estimate single and double scattering amplitudes contributing to the $`\overline{p}D`$ interaction, and observe that the interference between them decreases the single scattering output. At the same time our calculations (still in progress) show that, in the case of light nuclei, nuclear structure details and $`\overline{p}n`$ features do affect the results.
## 5 Conclusions.
We have shown that the Obelix Collaboration data on $`\overline{p}p`$ annihilation in the range 30 to 180 MeV/c allow us for a finer tuning of the parameters of an optical potential, which was previously used by other authors to interpolate elastic differential cross sections at $`k_{lab}`$ 181, 287 and 505 MeV/c. Without the need of introducing any energy dependence of these parameters, the so-obtained potential can reproduce all the $`\overline{p}p`$ annihilation data between 30 and 600 MeV/c, the zero-energy value of the $`\rho `$ parameter together with its general trend at increasing energies, and the measured values of the scattering length (real and imaginary part) with correct sign. We have also used this potential to predict elastic cross sections and $`\rho `$ values in those regions where data are not available yet. We have also shown that the behavior of all the considered observables is largely affected by a mechanism that we call “inversion”: in presence of a very strong reaction mechanism the reaction cross sections become anomalously small at very low energies, while elastic interactions reverse from attractive to repulsive. We can’t make precise predictions for the $`\overline{p}`$nucleus cross sections yet, but we stress that their smallness is closely related with the low-energy repulsive behavior of the $`\overline{p}p`$ interaction.
|
warning/0002/hep-ph0002162.html
|
ar5iv
|
text
|
# 1 Introduction and summary
## 1 Introduction and summary
Effective field theory allows one to analyze the chiral structure of Quantum Chromodynamics in the low energy domain, which is not accessible to a perturbative expansion in the strong coupling constant. The spontaneous violation of the chiral symmetry of QCD entails the existence of Goldstone bosons, the pions (we consider here the two flavor case of the light up and down quarks). The interactions of the Goldstone bosons among themselves and with matter vanish as the momentum transfer goes to zero. This is a consequence of Goldstone’s theorem. Consequently, such interactions can be analyzed in a perturbative expansion, where all momenta and masses are small compared to the typical scale of hadronic interactions, say the mass of the rho meson. This is the so–called chiral expansion. A systematic investigation of processes involving pions allows therefore to understand in a precise and quantitative manner how the symmetry violation takes place and also to pin down the ratios of the light quark masses. One such process is elastic pion–nucleon scattering. In ref. (called (I) from here on), we considered this reaction in the framework of heavy baryon chiral perturbation theory (HBCHPT) to third order in the chiral expansion. At that order, the first contributions from pion loop graphs, which perturbatively restore unitarity, appear. Besides pion loop diagrams, there are tree graphs. Some of these have fixed coefficients, others are accompanied by coupling constants not fixed by chiral symmetry. These so–called low energy constants (LECs) must be determined by a comparison to data or can be estimated using models. As has been shown in refs., these LECs encode the information from higher mass states not present explicitly in the effective field theory. There are three important reasons to extend the calculations of (I) to fourth order: First, only at that order one has a complete one–loop representation. Second, it is known that these fourth order corrections can be large (for a review see ref. and an update is given in ref.). Third, only after having obtained an accurate representation of the isospin–symmetric amplitude, as done here, one can attack the more subtle problem of isospin symmetry violation in low energy pion–nucleon scattering. First steps in the framework of HBCHPT have been reported in refs..
There have been some interesting new developments since (I) appeared: First, a manifestly Lorentz invariant form of baryon chiral perturbation theory was proposed in ref. and some implications for the nucleon mass and the scalar form factor to fourth order were worked out. The same group is also investigating pion–nucleon scattering in that framework . By construction, their approach leads to the correct analytical structure of the pion–nucleon scattering amplitude, whereas in the heavy baryon approach special care has to be taken in certain regions of the complex plane. Second, a different determination of the dimension two and three LECs by fitting the HBCHPT amplitude to the dispersive representation (based on the Karlsruhe partial wave analysis) inside the Mandelstam triangle was performed in ref.. While that method has the a priori advantage that the chiral expansion is expected to converge best in this special region of the Mandelstam plane, it is difficult to work out the theoretical uncertainties. Another drawback of this procedure is that only three dimension two LECs could be determined with sufficient precision (if one uses the third order HBCHPT amplitude as input). This is related to the fact that close to the point $`\nu =t=0`$ the contribution from the third order terms is accompanied by very small kinematical prefactors.
Here, we will follow the approach used in (I), namely to fit to the phase shifts provided by three different partial wave analyses for pion laboratory momenta between 40 and 100 MeV. This not only allows for a discussion of the uncertainties due to the input but also gives us the possibility to predict the phase shifts at higher and lower momenta, in particular the scattering lengths and the range parameters. Furthermore, we are then able to directly compare to the third order calculation and draw conclusions on the convergence of the chiral expansion. Of course, at the appropriate places we will discuss the relation to the work reported in refs..
The pertinent results of the present investigation can be summarized as follows:
1. We have constructed the complete one–loop amplitude for elastic pion–nucleon scattering in heavy baryon chiral perturbation theory, including all terms of order $`q^4`$. It contains 13 low–energy constants plus one related to fixing the pion–nucleon coupling constant through the Goldberger–Treiman discrepancy. Their values can be determined by fitting to the two S– and four P–wave partial wave amplitudes for three different sets of available pion–nucleon phase shifts in the physical region at low energies (typically in the range of 40 to 100 MeV pion momentum in the laboratory frame).
2. We have performed two types of fits. In the first one, we fit four combinations of the dimension two and four LECs, together with five LECs from the third and five from the fourth order. This means that the dimension two LECs are subject to quark mass renormalizations from certain fourth order terms. Most fitted LECs are of “natural” size. In the second approach, we fix the dimension two LECs as determined from the third order fit and determine the corresponding dimension four LEC combinations separately. We have studied the convergence of the chiral expansion by comparing the best fits based on the second, third and fourth order representation of the scattering amplitudes. The fourth order corrections are in general not large, but they improve the description of most partial waves. This indicates convergence of the chiral expansion.
3. We can predict the phases at lower and at higher energies, in particular the threshold parameters (scattering lengths and effective ranges). The results are not very different from the third order study in (I), but the description of the P–waves is improved, in particular the scattering length in the delta channel and the energy dependence of the small P–waves. The errors on the S-wave scattering lengths are as in (I) since they are due to the differences in the partial wave analyses used as input. Our theoretical predictions are consistent with recent determinations from pionic hydrogen and deuterium .
4. The pion–nucleon sigma term (at zero momentum transfer) can not be predicted without further input since at fourth order a new combination of LECs appears, that is not pinned down by the scattering data. Therefore, we have analyzed the sigma term at the Cheng–Dashen point. Using a family of sum rules which relate this quantity to threshold parameters and known dispersive integrals, we find results consistent with other determinations using the various partial wave analyses.
5. We do not find any improvement of the chiral description of the so–called subthreshold parameters as reported in (I). In some cases, the fourth order prediction is worse than the third order one. Since our amplitude is pinned down in the physical region, we do not expect the extrapolation to the subthreshold region to be very precise. To circumvent this problem, it is mandatory to combine the chiral representation obtained here with dispersion relations, see e.g. ref., or by fitting directly inside the Mandelstam triangle .
The manuscript is organized as follows. In section 2, we briefly discuss the effective Lagrangian underlying our calculation. All details were given in (I), so here we only spell out the new terms at fourth order. Section 3 contains the HBCHPT results for the pion–nucleon scattering amplitudes $`g^\pm ,h^\pm `$ to fourth order. The fitting procedure together with the results for the phase shifts and threshold parameters are presented in section 4. We also spell out the remaining problems related to the sigma term and the subthreshold parameters. The appendix contains the analytical expressions for the various threshold parameters.
## 2 Effective Lagrangian
The starting point of our approach is the most general chiral invariant Lagrangian built from pions, nucleons and external scalar sources (to account for the explicit chiral symmetry breaking). The Goldstone bosons are collected in a 2$`\times `$2 matrix-valued field $`U(x)=u^2(x)`$. We use the so–called sigma model parametrisation (gauge). We work in the framework of heavy baryon chiral perturbation theory, thus the nucleons are described by structureless non–relativistic spin-$`\frac{1}{2}`$ particles, denoted by $`N(x)`$. The effective theory admits a low energy expansion, i.e. the corresponding effective Lagrangian can be written as (for more details and references, see e.g. )
$$_{\mathrm{eff}}=_{\pi \pi }^{(2)}+_{\pi \pi }^{(4)}+_{\pi N}^{(1)}+_{\pi N}^{(2)}+_{\pi N}^{(3)}+_{\pi N}^{(4)}\mathrm{},$$
(2.1)
where the ellipsis denotes terms of higher order. For the explicit form of the meson Lagrangian and the dimension one, two and three pion–nucleon terms, we refer to (I). The complete fourth order Lagrangian is given in ref. and the renormalization is discussed in ref.. For completeness, we display here the finite terms from $`_{\pi N}^{(4)}`$ which contribute to elastic $`\pi `$N scattering
$`_{\pi N}^{(4)}`$ $`=`$ $`\overline{N}\{(\overline{e}_{14}{\displaystyle \frac{1}{16m^2}}c_2)h_{\mu \nu }h^{\mu \nu }+(\overline{e}_{15}{\displaystyle \frac{1}{256m^3}}g_A^2{\displaystyle \frac{1}{16m}}(\overline{d}_{14}\overline{d}_{15}))v^\mu v^\nu h_{\lambda \mu }h_\nu ^\lambda `$
$`+\left(\overline{e}_{16}+{\displaystyle \frac{3}{256m^3}}g_A^2\right)v^\lambda v^\mu v^\nu v^\rho h_{\lambda \mu }h_{\nu \rho }+\overline{e}_{17}[S^\mu ,S^\nu ][h_{\lambda \mu },h_\nu ^\lambda ]`$
$`+\left(\overline{e}_{18}{\displaystyle \frac{1}{128m^3}}\left(2+g_A^2\right){\displaystyle \frac{1}{16m^2}}c_4\right)[S^\mu ,S^\nu ]v^\lambda v^\rho [h_{\lambda \mu },h_{\nu \rho }]+\overline{e}_{19}\chi _+uu`$
$`+\left(\overline{e}_{20}{\displaystyle \frac{1}{32m^2}}g_A^2c_1{\displaystyle \frac{1}{8m}}g_A\overline{d}_{16}\right)\chi _+(vu)^2+\left(\overline{e}_{21}+{\displaystyle \frac{1}{16m^2}}c_1\right)[S^\mu ,S^\nu ]\chi _+[u_\mu ,u_\nu ]`$
$`+\overline{e}_{22}[D_\mu ,[D^\mu ,\chi _+]]+\overline{e}_{35}iv^\mu v^\nu \stackrel{~}{\chi }_{}h_{\mu \nu }+\overline{e}_{36}iu_\mu [D^\mu ,\stackrel{~}{\chi }_{}]`$
$`+\overline{e}_{37}i[S^\mu ,S^\nu ][u_\mu ,[D_\nu ,\stackrel{~}{\chi }_{}]]+\overline{e}_{38}\chi _+\chi _++\overline{e}_{115}\chi \chi ^{}+\overline{e}_{116}\left(det\chi +det\chi ^{}\right)`$
$`{\displaystyle \frac{1}{16m^2}}c_2h_\mu ^\mu h_\nu ^\nu \left({\displaystyle \frac{1}{128m^3}}g_A^2{\displaystyle \frac{1}{16m}}\left(\overline{d}_{14}\overline{d}_{15}\right)\right)v^\mu v^\nu h_{\mu \nu }h_\rho ^\rho `$
$`{\displaystyle \frac{1}{4m^2}}c_2u_\mu [D^\mu ,h_\nu ^\nu ]\left({\displaystyle \frac{1}{128m^3}}g_A^2+{\displaystyle \frac{1}{8m}}\left(\overline{d}_{14}\overline{d}_{15}\right)\right)vu[vD,h_\mu ^\mu ]`$
$`\left({\displaystyle \frac{1}{128m^3}}g_A^2{\displaystyle \frac{1}{8m}}\left(\overline{d}_{14}\overline{d}_{15}\right)\right)v^\mu u^\nu [vD,h_{\mu \nu }]+{\displaystyle \frac{1}{32m^3}}g_A^2v^\mu v^\nu vu[vD,h_{\mu \nu }]`$
$`+\left({\displaystyle \frac{1}{32m^3}}\left(1+g_A^2\right)+{\displaystyle \frac{1}{8m^2}}c_4\right)[S^\mu ,S^\nu ]v^\rho [u_\nu ,[vD,h_{\mu \rho }]]{\displaystyle \frac{1}{8m}}g_A\overline{d}_{18}ivu[vD,\stackrel{~}{\chi }_{}]`$
$`+({\displaystyle \frac{1}{128}}g_A^2+{\displaystyle \frac{1}{32m^2}}c_4)([h_\mu ^\mu ,u^\lambda ]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`({\displaystyle \frac{1}{128m^3}}g_A^2+{\displaystyle \frac{1}{32m^2}}c_4{\displaystyle \frac{1}{4m}}(\overline{d}_1+\overline{d}_2))([h^{\lambda \mu },u_\mu ]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+({\displaystyle \frac{1}{128m^3}}(g_A^21)+{\displaystyle \frac{1}{2m}}\overline{d}_3)(v_\mu [h^{\lambda \mu },vu]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+({\displaystyle \frac{1}{128m^3}}+{\displaystyle \frac{1}{4m}}\overline{d}_3)(v^\mu v^\nu [h_{\mu \nu },u^\lambda ]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+({\displaystyle \frac{1}{64m^3}}g_A^2{\displaystyle \frac{1}{8m^2}}c_3)([S^\lambda ,S^\mu ]h_{\mu \nu }u^\nu D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+{\displaystyle \frac{1}{4m}}(\overline{d}_{14}\overline{d}_{15})([S^\mu ,S^\nu ]h_\mu ^\lambda u_\nu D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`({\displaystyle \frac{1}{64m^3}}g_A^2+{\displaystyle \frac{1}{8m^2}}c_2{\displaystyle \frac{1}{4m}}(\overline{d}_{14}\overline{d}_{15}))([S^\lambda ,S^\mu ]v^\nu h_{\mu \nu }vuD_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+({\displaystyle \frac{1}{64m^3}}g_A^2{\displaystyle \frac{1}{4m}}(\overline{d}_{14}\overline{d}_{15}))([S^\lambda ,S^\mu ]v^\nu v^\rho u_\mu h_{\nu \rho }D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+{\displaystyle \frac{1}{32m^3}}g_A(iS^\lambda [vD,h_\mu ^\mu ]D_\lambda +\mathrm{h}.\mathrm{c}.)+{\displaystyle \frac{1}{16m^3}}g_A(i[SD,h_\mu ^\mu ]vD+\mathrm{h}.\mathrm{c}.)`$
$`{\displaystyle \frac{1}{32m^3}}g_A(iS^\nu [D^\mu ,h_{\mu \nu }]vD+\mathrm{h}.\mathrm{c}.){\displaystyle \frac{1}{32m^3}}g_A(iS^\mu v^\nu [vD,h_{\mu \nu }]vD+\mathrm{h}.\mathrm{c}.)`$
$`{\displaystyle \frac{1}{32m^3}}g_A(iS^\lambda v^\mu v^\nu [vD,h_{\mu \nu }]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`({\displaystyle \frac{1}{4m^2}}g_Ac_1+{\displaystyle \frac{1}{2m}}\overline{d}_{16})(iS^\lambda \chi _+vuD_\lambda +\mathrm{h}.\mathrm{c}.){\displaystyle \frac{1}{4m^2}}c_1([S^\lambda ,S^\mu ][D_\mu ,\chi _+]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`+{\displaystyle \frac{1}{2m}}\overline{d}_5(i[\stackrel{~}{\chi }_{},u^\lambda ]D_\lambda +\mathrm{h}.\mathrm{c}.)+{\displaystyle \frac{1}{2m}}\overline{d}_{18}(S^\lambda [vD,\stackrel{~}{\chi }_{}]D_\lambda +\mathrm{h}.\mathrm{c}.)`$
$`{\displaystyle \frac{1}{2m^2}}c_2D_\mu u^\mu u^\nu D_\nu +\left({\displaystyle \frac{1}{64m^3}}g_A^2{\displaystyle \frac{1}{8m^2}}c_3\right)D_\mu uuD^\mu `$
$`\left({\displaystyle \frac{1}{64m^3}}g_A^2+{\displaystyle \frac{1}{8m^2}}c_2\right)D_\mu \left(vu\right)^2D^\mu \left({\displaystyle \frac{1}{32m^3}}g_A^2+{\displaystyle \frac{1}{8m^2}}c_4\right)D_\mu [S^\rho ,S^\tau ][u_\rho ,u_\tau ]D^\mu `$
$`+({\displaystyle \frac{1}{16m^3}}g_A^2+{\displaystyle \frac{1}{4m^2}}c_4)(D_\mu [S^\mu ,S^\rho ][u^\nu ,u_\rho ]D_\nu +\mathrm{h}.\mathrm{c}.){\displaystyle \frac{1}{4m^2}}c_1D_\mu \chi _+D^\mu `$
$`{\displaystyle \frac{1}{4m^3}}g_A(iuDSDvD+\mathrm{h}.\mathrm{c}.)+{\displaystyle \frac{1}{8m^3}}g_A(iSuD^2vD+\mathrm{h}.\mathrm{c}.)`$
$`+{\displaystyle \frac{3}{8m^3}}g_A(ivuSDvDvD+\mathrm{h}.\mathrm{c}.){\displaystyle \frac{1}{8m^3}}g_A(iSuvDvDvD+\mathrm{h}.\mathrm{c}.)\}N,`$
with $`v_\mu `$ the nucleon’s four–velocity, $`S_\mu `$ the covariant spin–vector, $`D_\mu `$ the chiral covariant derivative, $`u_\mu =i(_\mu uu^{}+u^{}_\mu u)`$, $`h_{\mu \nu }=[D_\mu ,u_\nu ]+[D_\nu ,u_\mu ]`$ and $`\chi _\pm =u^{}\chi u^{}\pm u\chi ^{}u`$ encoding the explicit chiral symmetry breaking. Traces in flavor space are denoted by $`\mathrm{}`$ and $`\stackrel{~}{\chi }_{}=\chi _{}\chi _{}/2`$. We remark that the various parameters like $`g_A,m,\mathrm{}`$ appearing in the effective Lagrangian have to be taken in the chiral SU(2) limit ($`m_u=m_d=0,m_s`$ fixed) and should be denoted as $`\underset{A}{\overset{}{g}},\stackrel{}{m},\mathrm{}`$. Throughout this manuscript, we will not specify this but it should be kept in mind. It is also worth mentioning the particular role of the terms $`e_{115,116}`$. These operators have no pion matrix elements but are simply contact interactions of the external scalar source with the matter fields and thus contribute to the nucleon mass and the scalar form factor. This will be of importance later on. Having constructed the effective pion–nucleon Lagrangian to order $`q^4`$, we now turn to use it in order to describe elastic pion–nucleon scattering. To account for isospin breaking, one has to extend this Lagrangian to include virtual photons. This has already been done in and we refer the reader to these papers. For a systematic study of isospin violation in the elastic and charge exchange channels, one first has to find out to what accuracy the low energy $`\pi N`$ phase shifts can be described in the isospin symmetric framework.<sup>#5</sup><sup>#5</sup>#5An investigation of isospin violation in the threshold amplitudes to third order was reported in ref.. This is the question which will be addressed in the remaining sections of this investigation.
## 3 Pion–nucleon scattering
### 3.1 Basic definitions
In this section, we only give a few basic definitions pertinent to elastic pion–nucleon scattering. For a more detailed discussion, we refer to (I). In the center-of-mass system (cms), the amplitude for the process $`\pi ^a(q_1)+N(p_1)\pi ^b(q_2)+N(p_2)`$ takes the following form (in the isospin basis):
$`T_{\pi N}^{ba}`$ $`=`$ $`\left({\displaystyle \frac{E+m}{2m}}\right)\{\delta ^{ba}[g^+(\omega ,t)+i\stackrel{}{\sigma }(\stackrel{}{q}_2\times \stackrel{}{q}_1)h^+(\omega ,t)]`$ (3.1)
$`+iϵ^{bac}\tau ^c[g^{}(\omega ,t)+i\stackrel{}{\sigma }(\stackrel{}{q}_2\times \stackrel{}{q}_1)h^{}(\omega ,t)]\}`$
with $`\omega =vq_1=vq_2`$ the pion cms energy, $`E_1=E_2E=(\stackrel{}{q}^2+m^2)^{1/2}`$ the nucleon energy and $`\stackrel{}{q}_1^2=\stackrel{}{q}_2^2\stackrel{}{q}^2=((sM^2m^2)^24m^2M^2)/(4s)`$. $`t=(q_1q_2)^2`$ is the invariant momentum transfer squared and $`s`$ denotes the total cms energy squared. Furthermore, $`g^\pm (\omega ,t)`$ refers to the isoscalar/isovector non-spin-flip amplitude and $`h^\pm (\omega ,t)`$ to the isoscalar/isovector spin-flip amplitude. This form is most suitable for the HBCHPT calculation since it is already defined in a two–component framework.
### 3.2 Chiral expansion of the amplitudes
What we are after is the chiral expansion of the various amplitudes $`g^\pm ,h^\pm `$. These consist of essentially three pieces, which are the tree and counterterm parts of polynomial type as well as the unitarity corrections due to the pion loops. To be precise, we have
$$X=X^{\mathrm{tree}}+X^{\mathrm{ct}}+X^{\mathrm{loop}},X=g^\pm ,h^\pm ,$$
(3.2)
where the tree contribution subsumes all Born terms with fixed coefficients, the counterterm amplitude the ones proportional to the dimension two, three and four LECs. The last term in Eq.(3.2) is the complete one–loop amplitude consisting of terms of order $`q^3`$ and $`q^4`$. The latter is a complex–valued function and restores unitarity in the perturbative sense. Its various terms are all proportional to $`1/F^4`$. We remark that the topologies of the new loop graphs are not different from the ones already present at third order. The loops of fourth order have exactly one insertion from the dimension two Lagrangian. Note that we treat the chiral symmetry breaking scale $`\mathrm{\Lambda }_\chi 1`$GeV on the same footing as the nucleon mass. In principle, one could also organize the loop expansion, which proceeds in powers of $`1/F^2`$, and the $`1/m`$ expansion independently of each other (with some prescription for the mixed terms). These amplitudes are functions of two kinematical variables, which we choose to be the pion energy and the invariant momentum transfer squared, i.e. $`X=X(\omega ,t)`$. In what follows, we mostly suppress these arguments. The full one–loop amplitude to order $`q^4`$ is obtained after mass and coupling constant renormalization,
$$(\underset{A}{\overset{}{g}},\stackrel{}{m},F,M)(g_A,m,F_\pi ,M_\pi ).$$
(3.3)
To fourth order, these read (we also give the corresponding $`Z`$–factors for the pion and the nucleon)
$`Z_\pi `$ $`=`$ $`1{\displaystyle \frac{2M^2}{F^2}}\mathrm{}_4{\displaystyle \frac{\mathrm{\Delta }_\pi }{F^2}},`$ (3.4)
$`Z_N`$ $`=`$ $`1{\displaystyle \frac{g_A^2}{F^2}}{\displaystyle \frac{3M^2}{32\pi ^2}}+{\displaystyle \frac{g_A^2}{F^2}}{\displaystyle \frac{9M^3}{64\pi m}},`$ (3.5)
$`M_\pi ^2`$ $`=`$ $`M^2\left\{1+{\displaystyle \frac{2M^2}{F^2}}\mathrm{}_3+{\displaystyle \frac{\mathrm{\Delta }_\pi }{2F^2}}\right\},`$ (3.6)
$`m`$ $`=`$ $`\stackrel{}{m}4M^2c_1{\displaystyle \frac{3g_A^2M^3}{32\pi F^2}}`$ (3.7)
$`M^4(16\overline{e}_{38}+2\overline{e}_{115}+{\displaystyle \frac{1}{2}}\overline{e}_{116})+{\displaystyle \frac{3M^4c_2}{128\pi ^2F^2}}{\displaystyle \frac{3g_A^2M^4}{64\pi ^2mF^2}},`$
$`F_\pi `$ $`=`$ $`F\left\{1+{\displaystyle \frac{M^2}{F^2}}\mathrm{}_4{\displaystyle \frac{\mathrm{\Delta }_\pi }{F^2}}\right\},`$ (3.8)
$`{\displaystyle \frac{g_A}{F_\pi }}`$ $`=`$ $`{\displaystyle \frac{\underset{A}{\overset{}{g}}}{F}}\{1{\displaystyle \frac{M^2}{F^2}}\mathrm{}_4+{\displaystyle \frac{4M^2}{g_A}}d_{16}(\lambda )+{\displaystyle \frac{g_A^2}{4F^2}}(\mathrm{\Delta }_\pi {\displaystyle \frac{M^2}{4\pi ^2}})`$ (3.9)
$`{\displaystyle \frac{M^3}{6\pi F^2}}({\displaystyle \frac{1}{8m}}+c_3(2c_4+{\displaystyle \frac{1}{2m}}){\displaystyle \frac{3g_A^2}{4m}})\},`$
with
$$\mathrm{\Delta }_\pi =2M_\pi ^2\left(L+\frac{1}{16\pi ^2}\mathrm{ln}\frac{M_\pi }{\lambda }\right)+𝒪(d4)$$
(3.10)
and
$$L=\frac{\lambda ^{d4}}{16\pi ^2}\left(\frac{1}{d4}+\frac{1}{2}(\gamma _E1\mathrm{ln}4\pi )\right),$$
(3.11)
where Euler’s constant $`\gamma _E=0.557215`$ has been used, $`\lambda `$ is the scale of dimensional regularization and $`d`$ the number of space–time dimensions.
After these preliminaries, we give the final expressions for the tree, counterterm and loop graphs at fourth order in terms of the renormalized quantities:
Tree and counterterm graphs:
$`m^3F_\pi ^2g^+(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{64\omega ^4}}[32\omega ^4M_\pi ^2t+45M_\pi ^4t^211M_\pi ^2t^380M_\pi ^6t+52M_\pi ^8`$ (3.12)
$`+7\omega ^2t^3+110\omega ^2M_\pi ^4t49\omega ^2M_\pi ^2t^276\omega ^2M_\pi ^6+11\omega ^4t^2`$
$`+t^44\omega ^6t+28\omega ^4M^4]`$
$`+{\displaystyle \frac{1}{2}}M_\pi ^2(2\omega ^22M_\pi ^2+t)mc_1+8M_\pi ^4m^3c_1{\displaystyle \frac{\mathrm{}_3}{F_\pi ^2}}`$
$`+{\displaystyle \frac{1}{4}}(22\omega ^2M_\pi ^2+8M_\pi ^4+3\omega ^2t+14\omega ^44M_\pi ^2t)mc_2`$
$`+{\displaystyle \frac{1}{8}}(4\omega ^2M_\pi ^2+2\omega ^2t+4M_\pi ^44M_\pi ^2t+t^2)mc_316M_\pi ^2\omega ^2m^2c_1c_2`$
$`+\omega ^2tm^2(\overline{d}_{14}\overline{d}_{15})+{\displaystyle \frac{g_A}{4\omega ^2}}M_\pi ^2(4M_\pi ^4+t^2+4\omega ^2t4M_\pi ^2t)m^2\overline{d}_{18}`$
$`4(4M_\pi ^2t4M_\pi ^4+t^2)m^3\overline{e}_{14}8\omega ^2(2M_\pi ^2+t)m^3\overline{e}_{15}`$
$`+16\omega ^4m^3\overline{e}_{16}4M_\pi ^2(2M_\pi ^2+t)m^3(2\overline{e}_{19}\overline{e}_{22}\overline{e}_{36})`$
$`+16\omega ^2M_\pi ^2m^3(\overline{e}_{20}+\overline{e}_{35}{\displaystyle \frac{g_A\overline{d}_{16}}{8m}})+8M_\pi ^4m^3(\overline{e}_{22}4\overline{e}_{38}),`$
$`m^3F_\pi ^2h^+(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{32\omega ^4}}[5\omega ^2t^2+27M_\pi ^4t9M_\pi ^2t^225\omega ^2M_\pi ^2t28M_\pi ^6+t^3`$ (3.13)
$`+30\omega ^2M_\pi ^44\omega ^4M_\pi ^2+3\omega ^4t]`$
$`+M_\pi ^2mc_1{\displaystyle \frac{1}{2}}\omega ^2mc_2+{\displaystyle \frac{1}{4}}(2M_\pi ^2+t)mc_3`$
$`+{\displaystyle \frac{1}{2}}(8\omega ^24M_\pi ^2+t)m^2(\overline{d}_{14}\overline{d}_{15}){\displaystyle \frac{g_A}{2\omega ^2}}M_\pi ^2(4M_\pi ^2+t)m^2\overline{d}_{18},`$
$`m^3F_\pi ^2g^{}(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{64\omega ^4}}[32\omega ^4M_\pi ^2t+45M_\pi ^4t^211M_\pi ^2t^382M_\pi ^6t+56M_\pi ^8`$ (3.14)
$`+7\omega ^2t^3+110\omega ^2M_\pi ^4t49\omega ^2M_\pi ^2t^280\omega ^2M_\pi ^6+11\omega ^4t^2`$
$`+8\omega ^8+t^42\omega ^6t8\omega ^6M_\pi ^2+24\omega ^4M_\pi ^4]`$
$`+{\displaystyle \frac{1}{32\omega ^4}}\left[4\omega ^8+8\omega ^6M_\pi ^24\omega ^4M_\pi ^4\omega ^6t+\omega ^4M_\pi ^2t\right]`$
$`+{\displaystyle \frac{1}{16}}t(4\omega ^24M_\pi ^2+t)mc_4`$
$`{\displaystyle \frac{1}{2}}(8\omega ^2M_\pi ^2+4\omega ^2t+8M_\pi ^4+t^26M_\pi ^2t)m^2(\overline{d}_1+\overline{d}_2)`$
$`+3\omega ^2(4\omega ^24M_\pi ^2+t)m^2\overline{d}_3+2M_\pi ^2(4\omega ^24M_\pi ^2+t)m^2\overline{d}_5`$
$`{\displaystyle \frac{g_A}{4\omega ^2}}M_\pi ^2(t^2+2\omega ^2t8\omega ^4+8M_\pi ^46M_\pi ^2t)m^2\overline{d}_{18},`$
$`m^3F_\pi ^2h^{}(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{32\omega ^4}}[5\omega ^2t^2+27M_\pi ^4t9M_\pi ^2t^225\omega ^2M_\pi ^2t26M_\pi ^6+t^3`$ (3.15)
$`+30\omega ^2M_\pi ^44\omega ^4M_\pi ^2+3\omega ^4t2\omega ^6]{\displaystyle \frac{\omega ^4(\omega ^2M_\pi ^2)}{16\omega ^4}}`$
$`M_\pi ^2mc_1+{\displaystyle \frac{1}{8}}(2\omega ^22M_\pi ^2+t)mc_4+{\displaystyle \frac{g_A}{2\omega ^2}}M_\pi ^2(2\omega ^22M_\pi ^2+t)m^2\overline{d}_{18}`$
$`4(2M_\pi ^2+t)m^3\overline{e}_{17}+8\omega ^2m^3\overline{e}_{18}+4M_\pi ^2m^3(2\overline{e}_{21}\overline{e}_{37}).`$
Loop graphs:
$`mF_\pi ^4g^+(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{1}{12\omega ^3}}J_0(\omega )[2M_\pi ^2g_A^4(tM_\pi ^22\omega ^4+4\omega ^2M_\pi ^2t\omega ^22M_\pi ^4)`$ (3.16)
$`+\omega ^2g_A^2(12\omega ^2M_\pi ^2M_\pi ^2t+\omega ^2t+12\omega ^4)+6\omega ^4M_\pi ^2]`$
$`+{\displaystyle \frac{1}{12\omega ^3}}J_0(\omega )[2g_A^4(6M_\pi ^6+8\omega ^610\omega ^4M_\pi ^2+3t\omega ^2M_\pi ^2+M_\pi ^2t^2+2t\omega ^4`$
$`4\omega ^2M_\pi ^45tM_\pi ^4)\omega ^2g_A^2(M_\pi ^2t+4\omega ^2M_\pi ^2+\omega ^2t4\omega ^4)`$
$`+6\omega ^4(3M_\pi ^2+t+4\omega ^2)]`$
$`{\displaystyle \frac{1}{12\omega ^2}}{\displaystyle \frac{J_0}{\omega }}(\omega )[g_A^4(8\omega ^6+\omega ^2t^28M_\pi ^6M_\pi ^2t^2+24\omega ^2M_\pi ^4+6t\omega ^4+6tM_\pi ^4`$
$`12t\omega ^2M_\pi ^224\omega ^4M_\pi ^2)`$
$`+2\omega ^2g_A^2(4\omega ^4M_\pi ^2t8\omega ^2M_\pi ^2+\omega ^2t+4M_\pi ^4)`$
$`+3\omega ^4(4\omega ^24M_\pi ^2+t)]`$
$`+{\displaystyle \frac{g_A^2}{32}}{\displaystyle \frac{K_0}{\omega }}(0,t)(12M_\pi ^4t4M_\pi ^69M_\pi ^2t^2+2t^3)`$
$`{\displaystyle \frac{1}{24}}I_0(t)[3g_A^2t(2tM_\pi ^2)48M_\pi ^2mc_1(M_\pi ^22t)`$
$`+2mc_2(2t^2+4M_\pi ^49tM_\pi ^2)+12mc_3(5tM_\pi ^2+2t^2+2M_\pi ^4)]`$
$`{\displaystyle \frac{1}{1152\pi ^2\omega ^3}}[2g_A^4(52\omega ^5M_\pi ^2+24\omega ^7+\omega ^3t^2+10t\omega ^596\pi M_\pi ^7+96\pi M_\pi ^5\omega ^2`$
$`+72\pi M_\pi ^5t12\pi M_\pi ^3t^248\pi M_\pi ^3\omega ^2t12\omega M_\pi ^4t+6\omega M_\pi ^2t^2`$
$`+11\omega ^3M_\pi ^2t+46M_\pi ^4\omega ^3)`$
$`+g_A^2(18\omega ^3t^2288\omega ^5M_\pi ^2+60\omega ^3M_\pi ^4+33\omega ^3tM_\pi ^2+192\omega ^7`$
$`+64\omega ^5t)+4\omega ^3mc_2(2t^2+6M_\pi ^413M_\pi ^2t)],`$
$`mF_\pi ^4h^+(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{12\omega ^3}}J_0(\omega )(\omega ^2M_\pi ^2)\left[g_A^2M_\pi ^22\omega ^2+8m\omega ^2(c_3c_4)\right]`$ (3.17)
$`{\displaystyle \frac{g_A^2}{12\omega ^3}}J_0(w)[g_A^2(M_\pi ^2\omega ^2+M_\pi ^2t3M_\pi ^4+2\omega ^4)`$
$`+(\omega ^2M_\pi ^2)(2\omega ^28\omega ^2m(c_3c_4))]`$
$`+{\displaystyle \frac{g_A^4}{24\omega ^2}}{\displaystyle \frac{J_0}{\omega }}(\omega )(4\omega ^4M_\pi ^2t8\omega ^2M_\pi ^2+\omega ^2t+4M_\pi ^4)`$
$`{\displaystyle \frac{g_A^2}{32}}{\displaystyle \frac{K_0}{\omega }}(0,t)(9M_\pi ^2t+4M_\pi ^4+2t^2){\displaystyle \frac{g_A^2}{8}}I_0(t)(2tM_\pi ^2)`$
$`{\displaystyle \frac{g_A^2}{1152\omega ^3\pi ^2}}[2g_A^2(12\pi M_\pi ^56\pi M_\pi ^3\omega ^2+6\pi M_\pi ^3t+32\omega ^5+4\omega ^3t`$
$`16\omega ^3M_\pi ^2+12\omega M_\pi ^43\omega tM_\pi ^2)`$
$`+3\omega ^2(3\omega M_\pi ^26\omega t16\pi M_\pi ^3)+192\pi M_\pi ^3\omega ^2m(c_3c_4)],`$
$`mF_\pi ^4g^{}(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{1}{24\omega ^3}}J_0(\omega )[g_A^4(t\omega ^2M_\pi ^2+2\omega ^4M_\pi ^24\omega ^2M_\pi ^4tM_\pi ^4+2M_\pi ^6)`$ (3.18)
$`+g_A^2(6\omega ^2M_\pi ^4+12\omega ^4M_\pi ^2+3\omega ^2M_\pi ^2t3\omega ^4t12\omega ^6`$
$`+8\omega ^2m(c_3c_4)(2\omega ^44M_\pi ^2\omega ^2+2M_\pi ^4tM_\pi ^2+\omega ^2t))`$
$`+6\omega ^4(M_\pi ^216M_\pi ^2mc_1+8\omega ^2m(c_2+c_3))]`$
$`{\displaystyle \frac{1}{24\omega ^3}}J_0(\omega )[g_A^4(5tM_\pi ^410\omega ^4M_\pi ^24\omega ^2M_\pi ^4+3t\omega ^2M_\pi ^2+6M_\pi ^6`$
$`+8\omega ^6+M_\pi ^2t^2+2t\omega ^4)`$
$`+g_A^2\omega ^2(6M_\pi ^44\omega ^2M_\pi ^2+\omega ^2t3M_\pi ^2t+4\omega ^4`$
$`+8m(c_3c_4)(2M_\pi ^4+tM_\pi ^2\omega ^2t2\omega ^4+4\omega ^2M_\pi ^2))`$
$`+6\omega ^4(3M_\pi ^2+t+4\omega ^2+16M_\pi ^2mc_18\omega ^2m(c_2+c_3))]`$
$`+{\displaystyle \frac{1}{48\omega ^2}}{\displaystyle \frac{J_0}{\omega }}(\omega )[g_A^4(12t\omega ^2M_\pi ^224\omega ^4M_\pi ^2+6tM_\pi ^4+8\omega ^6+24\omega ^2M_\pi ^4`$
$`8M_\pi ^6+6t\omega ^4M_\pi ^2t^2+\omega ^2t^2)`$
$`+4\omega ^2g_A^2(4\omega ^4+4M_\pi ^4+\omega ^2t8\omega ^2M_\pi ^2M_\pi ^2t)`$
$`+6\omega ^4(4M_\pi ^2+4\omega ^2+t)]`$
$`+{\displaystyle \frac{g_A^2}{8}}K_0(0,t)\omega (4M_\pi ^4+t^24M_\pi ^2t)`$
$`{\displaystyle \frac{g_A^2}{64}}{\displaystyle \frac{K_0}{\omega }}(0,t)(10M_\pi ^2t^2+32\omega ^2M_\pi ^424\omega ^2M_\pi ^2t+4\omega ^2t^2+32M_\pi ^4t`$
$`32M_\pi ^6+t^3)`$
$`+{\displaystyle \frac{1}{48}}I_0(t)[2g_A^2(t^24M_\pi ^2\omega ^2+4M_\pi ^4+4\omega ^2t5M_\pi ^2t)`$
$`8M_\pi ^2t+16M_\pi ^4+t^216\omega ^2M_\pi ^2+4\omega ^2t]`$
$`+{\displaystyle \frac{1}{2304\omega ^3\pi ^2}}[g_A^4(48\pi M_\pi ^712\pi M_\pi ^3\omega ^2t24\pi M_\pi ^5\omega ^2+48\pi M_\pi ^5t12\pi M_\pi ^3t^2`$
$`+96\omega ^7+4\omega ^3t^2+40t\omega ^5+72\pi \omega ^4M_\pi ^336\omega M_\pi ^4t+6\omega M_\pi ^2t^2`$
$`12\omega ^3M_\pi ^2t+32M_\pi ^4\omega ^3+48\omega M_\pi ^6176\omega ^5M_\pi ^2)`$
$`+g_A^2(7\omega ^3t^2+72\pi M_\pi ^3\omega ^2t144\pi M_\pi ^5\omega ^2+72\pi \omega ^4tM_\pi `$
$`456\omega ^5M_\pi ^2+264\omega ^3M_\pi ^494\omega ^3tM_\pi ^2+192\omega ^7+76\omega ^5t`$
$`+216\pi M_\pi ^3\omega ^4+192\pi M_\pi ^3\omega ^2m(c_3c_4)(t2\omega ^2+2M_\pi ^2))`$
$`+2\omega ^3(10tM_\pi ^224\omega ^2M_\pi ^2+t^2+24M_\pi ^4+4\omega ^2t)],`$
$`mF_\pi ^4h^{}(\omega ,t)`$ $`=`$ $`{\displaystyle \frac{g_A^2}{12\omega ^3}}J_0(\omega )(\omega ^2M_\pi ^2)(2M_\pi ^2g_A^2+3\omega ^2+8\omega ^2mc_4)`$ (3.19)
$`+{\displaystyle \frac{g_A^2}{12\omega ^3}}J_0(\omega )[2g_A^2(3M_\pi ^4+2\omega ^4+M_\pi ^2t+M_\pi ^2\omega ^2)`$
$`+\omega ^2(\omega ^23M_\pi ^2+8mc_4(\omega ^2M_\pi ^2))]`$
$`{\displaystyle \frac{g_A^2}{12\omega ^2}}{\displaystyle \frac{J_0}{\omega }}(\omega )[g_A^2(8M_\pi ^2\omega ^2M_\pi ^2t+\omega ^2t+4\omega ^4+4M_\pi ^4)`$
$`+2\omega ^2(\omega ^2M_\pi ^2)]`$
$`{\displaystyle \frac{g_A^2}{8}}K_0(0,t)\omega (4M_\pi ^2+t)+{\displaystyle \frac{g_A^2}{32}}{\displaystyle \frac{K_0}{\omega }}(0,t)(8M_\pi ^46M_\pi ^2t+t^2)`$
$`{\displaystyle \frac{1}{24}}I_0(t)\left[2g_A^2(5M_\pi ^2+2t)+(4M_\pi ^2t)(1+4mc_4)\right]`$
$`{\displaystyle \frac{1}{1152\omega ^3\pi ^2}}[g_A^4(96\pi M_\pi ^564\omega ^3M_\pi ^2+12\omega tM_\pi ^2+8\omega ^3t24\pi M_\pi ^3t`$
$`48\pi M_\pi ^3\omega ^2+64\omega ^5)`$
$`+g_A^2(18\omega ^3M_\pi ^2+48\omega ^5+36\pi \omega ^4M_\pi +11\omega ^3t`$
$`+16\omega ^3mc_4(15M_\pi ^24\omega ^2))`$
$`+2\omega ^3(6M_\pi ^2t+4mc_4(6M_\pi ^2t))].`$
We have used the loop functions of ref.. The $`\overline{e}_i`$ are scale-independent (using the same procedure to eliminate the chiral logarithms as detailed in (I) for the $`\overline{d}_i`$). It is important to stress that we have obtained a more precise representation of the imaginary parts as compared to (I) since in that paper, only the leading terms were included. Here, we also have the next–to–leading order corrections of the unitarity corrections. Clearly, unitarity is perturbatively fulfilled, i.e. $`\mathrm{Im}T^{(4)}(\mathrm{Re}T^{(2)})^2`$ in a highly symbolic notation, where $`T^{(n)}`$ refers to the chiral representation of the $`\pi `$N amplitude to $`n^{\mathrm{th}}`$ order. It goes without saying that we also expect the corresponding real parts to be given more accurately.
### 3.3 Counterterm combinations
In the counterterm and loop contributions given above, a set of LECs appears. Most of these only enter in certain combinations and some only lead to quark mass renormalizations of the dimension two LECs. Therefore, it is instructive to work out how many independent local contact terms can contribute to $`\pi `$N scattering to fourth order. This can be most easily done based on a dispersive analysis by counting the number of possible subtractions. For that, it is most appropriate to describe the pertinent T–matrix in terms of the standard invariant amplitudes $`A`$ and $`B`$,
$$T_{\pi N}^\pm =A^\pm +\overline{)𝑞}B^\pm ,$$
(3.20)
in a highly symbolic notation. The invariant amplitudes are functions of two variables, which one can choose to be $`\nu `$ and $`t`$; these count as $`𝒪(q)`$ and $`𝒪(q^2)`$, respectively. The most general polynom for the four amplitudes $`A^\pm ,B^\pm `$ to fourth order commensurate with crossing and the other symmetries thus takes the form
$`A_{\mathrm{pol}}^+`$ $`=`$ $`a_1^++a_2^+t+a_3^+\nu ^2+a_4^+t^2+a_5^+t\nu ^2+a_6^+\nu ^4,`$
$`A_{\mathrm{pol}}^{}`$ $`=`$ $`\nu (a_1^{}+a_2^{}t+a_3^{}\nu ^2),`$
$`B_{\mathrm{pol}}^+`$ $`=`$ $`b_1^+\nu ,`$
$`B_{\mathrm{pol}}^{}`$ $`=`$ $`b_1^{}+b_2^{}t+b_3^{}\nu ^2.`$ (3.21)
Certain combinations of dimension two, three and four LECs are related to the subtraction constants $`(a_1^+,\mathrm{},b_3^{})`$. We refrain from giving the precise relationship here since it is not needed in what follows. Therefore, in total we have 14 LECs since at third order there is one more related to the Goldberger–Treiman discrepancy, i.e. a local term with a LEC which allows to express the axial–vector coupling $`g_A`$ in terms of the pion–nucleon coupling $`g_{\pi N}`$, i.e.
$$g_{\pi N}=\frac{g_Am}{F_\pi }\left(1\frac{2M_\pi ^2\overline{d}_{18}}{g_A}\right).$$
(3.22)
This term is important if one wants to properly account for the Born terms expressed as a function of the pion–nucleon coupling constant. If one then calculates to orders $`q^2`$, $`q^3`$ and $`q^4`$, one has to pin down 4, 9 and 14 LECs, respectively. This pattern is quite different from the total number of terms in the Lagrangian allowed at the various orders (7, 23, and 118, respectively); it is a general rule that simple processes do not involve an exorbitant number of LECs. Indeed, the terms proportional to the dimension four LECs $`\overline{e}_i`$ ($`i=19,20,21,22,35,36,37,38`$) only amount to quark mass renormalizations of the dimension two LECs $`c_i`$ ($`i=1,2,3,4`$) via
$`\stackrel{~}{c}_1`$ $`=`$ $`c_12M^2(\overline{e}_{22}4\overline{e}_{38}),`$
$`\stackrel{~}{c}_2`$ $`=`$ $`c_2+8M^2\left(\overline{e}_{20}+\overline{e}_{35}{\displaystyle \frac{g_A\overline{d}_{16}}{8m}}\right),`$
$`\stackrel{~}{c}_3`$ $`=`$ $`c_3+4M^2(2\overline{e}_{19}\overline{e}_{22}\overline{e}_{36}),`$
$`\stackrel{~}{c}_4`$ $`=`$ $`c_4+4M^2(2\overline{e}_{21}\overline{e}_{37}).`$ (3.23)
We have also used these parameters in the one–loop graphs of order $`q^4`$, although this induces some higher order contributions. This is a very general phenomenon of CHPT calculations at higher orders (for a discussion, see e.g. ref.). There are different ways of determining the LECs. As in (I), we use data from the physical region for doing so. Our first strategy is to fit the renormalized $`c_i`$, called $`\stackrel{~}{c}_i`$ here, together with the four (combinations of) dimension three LECs $`\overline{d}_1+\overline{d}_2,\overline{d}_3,\overline{d}_5,\overline{d}_{14}\overline{d}_{15}`$ and $`\overline{d}_{18}`$ and the genuine dimension four LECs $`\overline{e}_{14},\overline{e}_{15},\overline{e}_{16},\overline{e}_{17},\overline{e}_{18}`$. As enumerated before, we thus have 14 free parameters. In such a fit, we cannot disentangle the $`\stackrel{~}{c}_i`$ into their quark mass dependent and independent pieces without further information from other processes. This defines our best fit. To study the convergence compared to the lower order calculations, we also show the best fit from (I) and a best second order fit based on tree diagrams with the dimension two insertions $`c_i`$. One can argue that the second order contribution is given by the amplitude up to second order, with the $`c`$’s taking on their values as given by the best fit at that order. By including the amplitude at third order, the values of the $`c`$’s will change, but these changes are considered to be effects of third order. (The same is valid of course for the $`c`$’d and $`d`$’s, when going from third to fourth order.) One might also be interested in how big the contribution from genuine second and third order terms are (terms really proportioanl to $`q^2`$, respectively to $`q^3`$). In order to address this question, we consider an alternative strategy, in which we fix the $`c_i`$ ($`i=1,2,3,4`$) to their values determined from the best fit up to third order and use the four combinations of dimension four LECs appearing in eq.(3.3) as fit parameters. Of course, this leads to the same number of free parameters, but this second method allows for a clean separation of the contributions from the various orders. Clearly, physical observables do not depend on this reshuffling of fit parameters (modulo higher order corrections effectively included when using the $`\stackrel{~}{c}_i`$ in the loop graphs).
## 4 Results
### 4.1 The fitting procedure
There are various possibilities to fix the LECs, a general discussion is given in (I). We proceed here along the same lines as in (I), namely we fit to the phase shifts given by three different partial wave analyses in the low energy region. As input we use the phase shifts of the Karlsruhe (KA85) group , from the analysis of Matsinos (EM98) and the solution called SP98 from the VPI/GW group <sup>#6</sup><sup>#6</sup>#6In the meantime, novel solutions like SM99 have appeared. Since these are not very different from SP98 and we want to have a direct comparison with the results of (I), we use SP98 here. We come back to this later. In contrast to what was done in (I), we do not assign a common error of 3% to the Karlsruhe and VPI phases, but rather mimic the uncertainties of the Matsinos analysis in all cases, which is 1.5% for $`S_{31}`$, 0.5% for $`S_{11}`$, 1% for $`P_{33}`$ and 3.5% for the other P–waves. This assignment gives more weight to the better determined larger partial waves and is more natural than one common global error.<sup>#7</sup><sup>#7</sup>#7Of course, this might induce some mismatch in the sense that real errors associated to the KA and VPI/GW phases are different from the ones of the Matsinos analysis. We believe, however, that this procedure is preferable to the one using common global errors. The LEC $`\overline{d}_{18}`$ is fixed by means of the Goldberger–Treiman discrepancy, i.e. by the value for the pion–nucleon coupling constant extracted in the various analyses. The actual values of $`g_{\pi N}`$ are $`g_{\pi N}=13.4\pm 0.1,\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}13.18}\pm 0.12,\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}13.13}\pm 0.03,`$ for KA85, EM98 and SP98. Throughout, we use $`g_A=1.26`$, $`F_\pi =92.4`$MeV, $`m=938.27`$MeV and $`M_\pi =139.57`$MeV. Finally, we remark that we do not use the value of the pion–nucleon $`\sigma `$–term in the fitting procedure. This has two reasons: First, as noted before, we only want to use information from the physical region to pin down the LECs and second, it is known that the convergence of the chiral series for this quantity is slow . Before presenting the results of the actual fits, we already anticipate that the EM98 data basis will lead to the smallest $`\chi ^2`$ for the following reasons. First, this data base is only covering the low–energy region of pion–nucleon scattering. Also, the representation is available on a denser grid of points in momentum transfer. In contrast, the KA85 and SP98 analyses span a much larger range of energy and thus uncertainties also from higher energies will play a role in the energy range considered here. Furthermore, in (I) we had already discussed that the extraction of the threshold parameters from the SP98 analysis is not unproblematic. Note, however, that the model underlying the EM98 analysis should not be used in the unphysical region, quite in contrast to the dispersion theoretical approach on which the KA85 phase shifts are based.
### 4.2 Phase shifts and threshold parameters
After the remarks of the preceding paragraph, we can now present results. For the KA85 case, we have fitted the data up to 100 MeV pion lab momentum (i.e. 4 points per partial wave at $`q_\pi =40,60,79,97`$ MeV). For the analysis of Matsinos, we use 17 points for each partial wave in the range of $`q_\pi =41.496.3`$ MeV. For the VPI SP98 solution, we use the 5 data points in the range between 60 and 100 MeV, which give a stable fit. Of course, we could now extend the fits to higher energies than it was done in (I), but for a better comparison we do not show the results of these extended fits here. As discussed in section 3.3, we have two options for pinning down the LECs. Using strategy one, i.e. working with the $`\stackrel{~}{c}_i`$, we call the fits corresponding to the Karlsruhe, Matsinos and VPI analysis, fit 1, 2 and 3, in order. The resulting LECs are given in table 1.
Note that the error on the LECs is purely the one given by the fitting routine and is certainly underestimated. We remark that the $`\stackrel{~}{c}_i`$ and $`\overline{d}_i`$ (or combinations thereof) are mostly of natural size, whereas some of the $`\overline{e}_i`$ come out fairly large. Also, there is some sizeable variation in the actual values of most LECs among the different fits. The resulting S– and P–wave phase shifts are shown in figs. 1 (fit 1), 2 (fit 2) and 3 (fit 3), in order. The corresponding $`\chi ^2`$/dof is 0.40, 0.008 and 0.44 for fits 1,2 and 3, respectively. The description of the phase shifts is excellent for fit 2. For fits 1 and 3, the $`S_{31}`$ and $`P_{11}`$ partial waves at pion momenta above 150 MeV are somewhat off. In all cases, the description of the $`P_{33}`$ partial wave is improved as compared to the third order calculation. Since we do not fit data below $`q_\pi =40,41,60`$MeV (fit 1,2,3) , the threshold parameters are now predictions. These are shown for the various fits in table 2, in comparison to the empirical values of the various analyses. First, we observe that the numbers resulting from the one–loop calculation are consistent with the “empirical” ones.
The agreement with the threshold parameters based on the chiral amplitude with the ones based on the approaches underlying the various partial wave analyses is best for fit 2, slightly worse for fit 1 and clearly problematic for some of the parameters of fit 3. The reason for this was already spelled out in (I). The bands for the S–wave scattering lengths $`a_{0+}^+`$ and $`a_{0+}^{}`$ are as in (I) since the uncertainties extracted there are mostly due to the input and not to the theory. They agree with the recent determinations from the shift and width of pionic hydrogen and deuterium, cf. Fig.2 in ref.. For comparison, we translate our bands on the isoscalar and isovector scattering lengths into the physical ones,
$`a_{\pi ^{}p\pi ^0n}`$ $`=`$ $`0.131\mathrm{}0.117M_\pi ^1[(0.128\pm 0.006)M_\pi ^1],`$
$`a_{\pi ^{}p\pi ^{}p}`$ $`=`$ $`0.073\mathrm{}0.093M_\pi ^1[(0.0883\pm 0.0008)M_\pi ^1],`$ (4.1)
where the experimental numbers (in the square brackets) are taken from ref.. Note, however, that recent progress in calculating $`\pi ^{}p`$ atoms in effective field theory lets one expect that the uncertainty due to electromagnetic corrections for the band derived from the hydrogen shift has been underestimated, see e.g. ref.. Furthermore, only recently deuteron wave functions have been obtained precisely enough in an EFT approach to readdress the question of the deuterium shift constraining the elementary $`\pi N`$ amplitudes. It would also be worthwhile to repeat the EFT calculation of pion–deuteron scattering using our fourth order $`\pi N`$ amplitudes as input. As argued before, we can study the convergence of the chiral expansion. In figs.1–3, the dot–dashed, dotted, dashed and solid lines refer to the best fits up to first, second, third and fourth order, respectively. Since we used the errors of the Matsinos analysis, it is best to consider fit 2 shown in fig. 2. In most cases, the fourth order corrections are smaller than the third order ones, indicating convergence. This could not be concluded from the third order calculation, compare the discussion in (I) and ref.. Note also that in some partial waves the second order result is close to the data. The resulting values of the $`c_i`$ are very different from the ones given in table 4. The second order best fit based on the (KA85, EM98, SP98) analysis leads to $`c_1=(0.81,0.77,1.06)`$, $`c_2=(2.47,2.69,2.36)`$, $`c_3=(3.78,3.96,4.04)`$ and $`c_4=(2.49,2.64,2.35)`$ (all in GeV<sup>-1</sup>).<sup>#8</sup><sup>#8</sup>#8Note that for such a second order fit the differentiation between the $`c_i`$ and $`\stackrel{~}{c}_i`$ becomes meaningless. That these values are very different from the ones based on a one–loop third order amplitude fit was already pointed out in ref.. It is of particular interest to study the convergence of the S–wave scattering lengths, which has been already discussed in ref. estimating LECs from resonance saturation. Our results are summarized in table 3. Although it was already shown in ref. that there are no fourth order corrections to $`a_{0+}^{}`$, the readjustment of the LECs when going from third to fourth order leads to a small difference. That this difference is so small is also in agreement with ref., where it was argued that the dominant correction to the Weinberg–Tomozawa low–energy theorem is a pion loop effect. For fits 1 and 2, the fourth order correction to the isoscalar S–wave scattering length is fairly small, even for fit 3 the dominant correction is the one from second to third order.
The second option is to keep the dimension two LECs fixed to their value determined from the third order fits and fit the additional four dimension four combinations. This allows for a clean discussion of the various contributions to the chiral expansion. The results for the LECs are shown in table 4.<sup>#9</sup><sup>#9</sup>#9The slight differences for the values of the $`c_i`$ as compared to the ones given in (I) stem from the fact that we use different error bars for the KA85 and SP98 partial waves as explained before.
The stability of the values for most of the LECs is better than in the previous case and the fourth order LECs $`\overline{e}_i`$, $`i=14,\mathrm{},18`$ smaller (more natural). Some of the additional combinations of dimension four LECs are fairly large and vary considerably for the various fits. The resulting S– and P–wave phase shifts are shown in figs. 4 (fit 1\*), 5 (fit 2\*) and 6 (fit 3\*), in order. The corresponding $`\chi ^2`$/dof is 0.50, 0.14 and 0.58 for fits 1\*,2\* and 3\*, respectively. In these plots a different way of looking at the convergence properties of the amplitude is adopted: all four curves are based on the same fit and are thus obtained with the same set of LECs. The dot–dashed, dotted, dashed and solid lines show the contributions from the amplitude up to first, second, third and fourth order, respectively. We remark that the fourth order contributions are mostly small, with the exception of the $`P_{11}`$ and $`P_{13}`$ partial waves. The threshold parameters determined from these fits come out very close to the values given before and we thus refrain from adding another table here. Similar remarks hold for the convergence of the S–wave scattering lengths, only that in this way of fixing the LECs there is indeed no contribution to the isovector one from fourth order.
### 4.3 Sigma term, subthreshold parameters and further comments
In this section, we briefly discuss the so–called subthreshold parameters and the sigma term. Already in (I) we noted that the representation of the chiral amplitude, when pinned down by scattering data, is not very precise in the unphysical region. In particular, the small isoscalar amplitudes are obtained from various contributions, which are individually much larger than their sum. Consequently, this fine balance which is enforced through the fit in the physical region down to the scattering lengths is disturbed because the strict $`1/m`$ expansion performed here does not properly account for all cuts appearing in the $`\pi `$N amplitude. In fact, using our fourth order representation, we do not find an improvement of the subthreshold parameters as given in (I), in some cases even a clear disimprovement. This problem could e.g. be circumvented in the formulation of ref.. Another option is to pin down the LECs inside the Mandelstam triangle , which will lead to an improved representation in the unphysical region. This is also reflected in the prediction for the sigma term, which came out rather large in the fits shown in (I), but was considerably different (and consistent with the result from dispersion theory) based on the method used in .
We now consider the sigma term, which is the matrix element of the explicit chiral symmetry breaking part of the QCD Lagrangian sandwiched between proton states at zero momentum transfer. While at third order we can directly give the sigma term, $`\sigma (0)`$, in terms of the LEC $`c_1`$, this can no longer be done at fourth order due to the appearance of the LEC combination $`2e_{115}+e_{116}/2`$. These operators contribute to the nucleon mass shift and the sigma term (scalar form factor) as noted before. These contact interactions have no pion matrix–elements and therefore can not appear in the scattering amplitude, even not in higher order loop graphs. We therefore use a more indirect method to determine the sigma term. For that, we consider $`\mathrm{\Sigma }=F_\pi ^2\overline{D}^+(\nu =0,t=2M_\pi ^2)`$ which can be related to $`\sigma (0)`$ by the venerable low–energy theorem of ref.. There exists a whole family of relations between $`\mathrm{\Sigma }`$ and certain combinations of threshold parameters, as detailed in ref.. These relations have been worked out to third order and should be generalized to fourth order. We will use here the version given in ref.,
$$\mathrm{\Sigma }=\pi F_\pi ^2[(4+2\mu +\mu ^2)a_{0+}^+4M_\pi ^2b_{0+}^++12\mu M_\pi ^2a_{1+}^+]+\mathrm{\Sigma }_0,$$
(4.2)
with $`\mathrm{\Sigma }_0=12.6`$MeV and $`\mu =M_\pi /m1/7`$. Using the pertinent threshold parameters from the fourth (third) order representation, we find $`\mathrm{\Sigma }=65(62)`$MeV, $`73(79)`$MeV and $`90(82)`$MeV for fits 1, 2 and 3, respectively. A special variant, which also contains some fourth order pieces, has recently been given by Olsson ,
$`\mathrm{\Sigma }`$ $`=`$ $`[F_\pi ^2F(2M_\pi ^2)],`$ (4.3)
$`F(2M_\pi ^2)`$ $`=`$ $`14.5a_{0+}^+5.06(a_{0+}^{1/2})^210.13(a_{0+}^{3/2})^216.65b_{0+}^+0.06a_1^++5.70a_{1+}^+0.05,`$
with the quantities on the right–hand–side being given in units of the pion mass. This leads to $`\mathrm{\Sigma }=73(69)`$MeV, $`85(91)`$MeV and $`104(93)`$MeV for fits 1, 2 and 3, respectively. We consider the differences between the results based on eqs.(4.2) and (4.3) (and also using the fourth order results for the threshold parameters in the third order representation, eq.(4.2)) as an indication of the size of the fourth order terms. We note that the values we find for the Karlsruhe analysis are consistent with the direct determination based on hyperbolic dispersion relations , whereas the results based on the SP98 partial waves lead to a sizeably larger value than advocated by the VPI/GW group .
## Acknowledgements
We are grateful to Thomas Becher and Bastian Kubis for some clarifying discussions. One of us (N.F.) would like to thank all members of the Kellogg Radiation Lab for the hospitality extended to her during a stay while this work was completed.
## Appendix A Threshold parameters
In this appendix, we give the analytical expressions for the threshold parameters up to fourth order. These read:
$`a_{0+}^+`$ $`=`$ $`{\displaystyle \frac{M_\pi ^2[g_A^2+8m(2c_1+c_2+c_3)]}{16\pi (m+M_\pi )F_\pi ^2}}`$ (A.1)
$`+{\displaystyle \frac{3g_A^2mM_\pi ^3}{256\pi ^2(m+M_\pi )F_\pi ^4}}`$
$`{\displaystyle \frac{g_A^2M_\pi ^4}{64\pi (m+M_\pi )m^2F_\pi ^2}}{\displaystyle \frac{4M_\pi ^4c_1c_2}{\pi (m+M_\pi )F_\pi ^2}}+{\displaystyle \frac{2mM_\pi ^4c_1\mathrm{}_3}{\pi (m+M_\pi )F_\pi ^4}}{\displaystyle \frac{g_AM_\pi ^4(2\overline{d}_{16}\overline{d}_{18})}{4\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{2M_\pi ^4m(2\overline{e}_{14}+2\overline{e}_{15}+2\overline{e}_{16}+2\overline{e}_{19}+2\overline{e}_{20}+2\overline{e}_{35}\overline{e}_{36}4\overline{e}_{38})}{\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^4[83g_A^2+2g_A^4+4m(2c_1c_3)]}{256\pi ^3(m+M_\pi )F_\pi ^4}},`$
$`a_{0+}^{}`$ $`=`$ $`{\displaystyle \frac{mM_\pi }{8\pi (m+M_\pi )F_\pi ^2}}`$ (A.2)
$`+{\displaystyle \frac{M_\pi ^3(g_A^2+32m^2(\overline{d}_1+\overline{d}_2+\overline{d}_3+2\overline{d}_5))}{32\pi m(m+M_\pi )F_\pi ^2}}+{\displaystyle \frac{M_\pi ^3m}{64\pi ^3(m+M_\pi )F_\pi ^4}},`$
$`b_{0+}^+`$ $`=`$ $`{\displaystyle \frac{g_A^2(4m^2+2mM_\pi M_\pi ^2)}{64\pi m^2(m+M_\pi )F_\pi ^2}}+{\displaystyle \frac{2c_1(2mM_\pi M_\pi ^2)+(c_2+c_3)(4m^22mM_\pi +M_\pi ^2)}{8\pi m(m+M_\pi )F_\pi ^2}}`$ (A.3)
$`+{\displaystyle \frac{M_\pi (g_A^2+8mc_2)}{16\pi m(m+M_\pi )F_\pi ^2}}+{\displaystyle \frac{g_A^2M_\pi (154m^218mM_\pi +9M_\pi ^2)}{3072\pi ^2m(m+M_\pi )F_\pi ^4}}`$
$`{\displaystyle \frac{g_A^2M_\pi ^2(16m^22mM_\pi +M_\pi ^2)}{256\pi (m+M_\pi )m^4F_\pi ^2}}+{\displaystyle \frac{M_\pi ^2c_2}{2\pi (m+M_\pi )mF_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2c_1c_2(4m^22mM_\pi +M_\pi ^2)}{\pi (m+M_\pi )m^2F_\pi ^2}}{\displaystyle \frac{M_\pi ^3c_1\mathrm{}_3(2mM_\pi )}{2\pi (m+M_\pi )mF_\pi ^4}}`$
$`{\displaystyle \frac{M_\pi ^2[8m^2(\overline{d}_{14}\overline{d}_{15})g_A\overline{d}_{18}(M_\pi ^22mM_\pi +4m^2)]}{16\pi (m+M_\pi )m^2F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2(\overline{e}_{14}+\overline{e}_{15}+\overline{e}_{16})(8m^2+2mM_\pi M_\pi ^2)}{\pi (m+M_\pi )mF_\pi ^2}}{\displaystyle \frac{M_\pi ^2(2mM_\pi M_\pi ^2)(\overline{e}_{22}4\overline{e}_{38})}{2\pi (m+M_\pi )mF_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2[2(\overline{e}_{20}+\overline{e}_{35}\frac{g_A\overline{d}_{16}}{8m})+(2\overline{e}_{19}\overline{e}_{22}\overline{e}_{36})](4m^2+2mM_\pi M_\pi ^2)}{2\pi (m+M_\pi )mF_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2}{9216\pi ^3(m+M_\pi )m^2F_\pi ^4}}[432m^2144mM_\pi +72M_\pi ^2+g_A^2(44m^2+54mM_\pi 27M_\pi ^2)`$
$`+g_A^4(m^2(88+192\pi )36mM_\pi +18M_\pi ^2)+mc_1(1248m^2144mM_\pi +72M_\pi ^2)`$
$`24m^3c_2+mc_3(768m^2+72mM_\pi 36M_\pi ^2)],`$
$`b_{0+}^{}`$ $`=`$ $`{\displaystyle \frac{2m^22mM_\pi +M_\pi ^2}{32\pi mM_\pi (m+M_\pi )F_\pi ^2}}`$ (A.4)
$`+{\displaystyle \frac{12g_A^2}{16\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi g_A^2(10m^22mM_\pi +M_\pi ^2)}{128\pi m^3(m+M_\pi )F_\pi ^2}}{\displaystyle \frac{M_\pi c_4}{4\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi [(\overline{d}_1+\overline{d}_2+\overline{d}_3)(6M_\pi ^22mM_\pi +M_\pi ^2)+\overline{d}_5(4m^24mM_\pi +2M_\pi ^2)]}{4\pi m(m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi (4m^2+6mM_\pi 3M_\pi ^214g_A^2m^2)}{768\pi ^3m(m+M_\pi )F_\pi ^4}}`$
$`{\displaystyle \frac{3g_A^2M_\pi ^2}{64\pi (m+M_\pi )m^2F_\pi ^2}}+{\displaystyle \frac{M_\pi ^2(\overline{d}_1+\overline{d}_2+3\overline{d}_3+2\overline{d}_5+g_A\overline{d}_{18})}{2\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2(18+69\pi g_A^2+(412\pi )g_A^4)}{2304\pi ^3(m+M_\pi )F_\pi ^4}},`$
$`a_1^{}`$ $`=`$ $`{\displaystyle \frac{g_A^2m}{24\pi M_\pi (m+M_\pi )F_\pi ^2}}`$ (A.5)
$`+{\displaystyle \frac{32g_A^2+8mc_4}{48\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi [33g_A^2+24mc_432m^2(\overline{d}_1+\overline{d}_2)+16m^2g_A\overline{d}_{18}]}{96\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi m^2[3+g_A^2(21+36\pi )+g_A^4(2+24\pi )]}{3456\pi ^3(m+M_\pi )F_\pi ^4}}`$
$`{\displaystyle \frac{g_A^2M_\pi ^2}{192\pi (m+M_\pi )m^2F_\pi ^2}}{\displaystyle \frac{M_\pi ^2c_1}{6\pi (m+M_\pi )mF_\pi ^2}}+{\displaystyle \frac{M_\pi ^2(\overline{d}_1+\overline{d}_2+3\overline{d}_3+2\overline{d}_5+g_A\overline{d}_{18})}{6\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{2mM_\pi ^2(2\overline{e}_{17}+2\overline{e}_{18}+2\overline{e}_{21}\overline{e}_{37})}{3\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2}{6912\pi ^3(m+M_\pi )F_\pi ^4}}[18+g_A^2(7233\pi )+g_A^4(4+30\pi )+96\pi g_A^2mc_3`$
$`+(17696\pi )g_A^2mc_4],`$
$`a_1^+`$ $`=`$ $`{\displaystyle \frac{g_A^2m}{12\pi M_\pi (m+M_\pi )F_\pi ^2}}`$ (A.6)
$`{\displaystyle \frac{g_A^2+2mc_3}{12\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi [c_2+2m(\overline{d}_{14}\overline{d}_{15}+g_A\overline{d}_{18})]}{6\pi (m+M_\pi )F_\pi ^2}}{\displaystyle \frac{mM_\pi g_A^2(231\pi +g_A^2(112+96\pi ))}{13824\pi ^3(m+M_\pi )F_\pi ^4}}`$
$`+{\displaystyle \frac{g_A^2M_\pi ^2}{192\pi (m+M_\pi )m^2F_\pi ^2}}+{\displaystyle \frac{M_\pi ^2(2c_1c_2c_3)}{8\pi (m+M_\pi )mF_\pi ^2}}+{\displaystyle \frac{M_\pi ^2(3(\overline{d}_{14}\overline{d}_{15})+2g_A\overline{d}_{18})}{6\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{2mM_\pi ^2(4\overline{e}_{14}2\overline{e}_{15}2\overline{e}_{19}+\overline{e}_{22}+\overline{e}_{36})}{3\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2}{6912\pi ^3(m+M_\pi )F_\pi ^4}}[36+g_A^2(13348\pi )+g_A^4(74+12\pi )`$
$`6m(52c_1c_232c_3(1+g_A^2\pi )+32\pi c_4)],`$
$`a_{1+}^{}`$ $`=`$ $`{\displaystyle \frac{mg_A^2}{24\pi M_\pi (m+M_\pi )F_\pi ^2}}`$ (A.7)
$`{\displaystyle \frac{g_A^2+2mc_4}{24\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi [3g_A^2+32m^2(\overline{d}_1+\overline{d}_2)16m^2g_A\overline{d}_{18}]}{96\pi m(m+M_\pi )F_\pi ^2}}{\displaystyle \frac{M_\pi m[3+g_A^2(2118\pi )+g_A^4(212\pi )]}{3456\pi ^3(m+M_\pi )F_\pi ^4}}`$
$`{\displaystyle \frac{g_A^2M_\pi ^2}{48\pi (m+M_\pi )m^2F_\pi ^2}}+{\displaystyle \frac{M_\pi ^2c_1}{12\pi (m+M_\pi )mF_\pi ^2}}+{\displaystyle \frac{M_\pi ^2(\overline{d}_1+\overline{d}_2+3\overline{d}_3+2\overline{d}_5+g_A\overline{d}_{18})}{6\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{mM_\pi ^2(2\overline{e}_{17}2\overline{e}_{18}2\overline{e}_{21}+\overline{e}_{37})}{3\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi ^2}{6912\pi ^3(m+M_\pi )F_\pi ^4}}[18+g_A^2(36+141\pi )+g_A^4(4+42\pi )`$
$`+8mg_A^2(12\pi c_3+(12\pi +11)c_4)],`$
$`a_{1+}^+`$ $`=`$ $`{\displaystyle \frac{g_A^2m}{24\pi M_\pi (m+M_\pi )F_\pi ^2}}`$ (A.8)
$`+{\displaystyle \frac{g_A^24mc_3}{24\pi (m+M_\pi )F_\pi ^2}}`$
$`+{\displaystyle \frac{M_\pi [3g_A^2+16mc_216m^2(\overline{d}_{14}\overline{d}_{15}+g_A\overline{d}_{18})]}{96\pi m(m+M_\pi )F_\pi ^2}}{\displaystyle \frac{M_\pi mg_A^2(231\pi +g_A^2(56+96\pi ))}{13824\pi ^2(m+M_\pi )F_\pi ^4}}`$
$`+{\displaystyle \frac{g_A^2M_\pi ^2}{48\pi (m+M_\pi )m^2F_\pi ^2}}{\displaystyle \frac{g_AM_\pi ^2\overline{d}_{18}}{6\pi (m+M_\pi )F_\pi ^2}}+{\displaystyle \frac{2mM_\pi ^2(4\overline{e}_{14}2\overline{e}_{15}2\overline{e}_{19}+\overline{e}_{22}+\overline{e}_{36})}{3\pi (m+M_\pi )F_\pi ^2}}`$
$`{\displaystyle \frac{M_\pi ^2}{6912\pi ^3(m+M_\pi )F_\pi ^4}}[36+g_A^2(133+24\pi )+g_A^4(10+66\pi )`$
$`+6m(52c_1+c_2+(3216\pi g_A^2)c_3+16\pi g_A^2c_4)].`$
Figures
|
warning/0002/quant-ph0002039.html
|
ar5iv
|
text
|
# Methodology for quantum logic gate construction
## I Introduction
Practical realization of quantum information processing requires specific types of quantum operations that may be difficult to construct. In particular, to perform quantum computation robustly in the presence of noise, one needs fault-tolerant implementation of quantum gates acting on states that are block-encoded using quantum error correcting codes . Fault-tolerant quantum gates must prevent propagation of single qubit errors to multiple qubits within any code block so that small correctable errors will not grow to exceed the correction capability of the code. This requirement greatly restricts the types of unitary operations that can be performed on the encoded qubits. Certain fault-tolerant operations can be implemented easily by performing direct transversal operations on the encoded qubits, in which each qubit in a block interacts only with one corresponding qubit, either in another block or in a specialized ancilla. Unfortunately, for a given code, only a few useful operations can be done transversally, and these are not universal in that they cannot be composed to approximate an arbitrary quantum circuit. To obtain a universal set of gates, additional gates have to be constructed using ancilla states and fault-tolerant measurement. Although these additional gates have been constructed successfully , their ad-hoc construction is complicated and is not easily generalized.
Another kind of application in which we are challenged to construct useful quantum operations from a limited set of primitives is in distributed quantum information processing. In this problem, certain kinds of communication between different parties are constrained or prohibited, but prior distribution of standard states may be allowed. For example, quantum teleportation demonstrates how an unknown quantum state can be sent between two parties without sending any quantum information, using only classical communication and prior entanglement. Protocols for distributed state preparation and computation are also known , but again, they have been largely constructed by hand and offer neither an explanation of why a particular ancilla state is required nor a systematic path for generalization.
A general framework for addressing such problems has been presented in ; it uses quantum teleportation as a basic primitive to enable construction of quantum operations that cannot be directly performed through unitary operations. This framework provides systematic and generalizable construction for an infinite family of fault-tolerant gates, including the $`\pi /8`$ and Toffoli gates. It does not, however, lead to circuits equivalent to (or as simple as) prior ad-hoc construction for the same gates.
In this paper, we provide an extension to the teleportation method of gate construction with a similar but simpler primitive, which we call “one-bit teleportation” because it uses one qubit instead of two as ancilla. This method simplifies the construction of and, furthermore, provides strikingly unified construction of the $`\pi /8`$, controlled-phase, and Toffoli gates. An infinite hierarchy of gates, including the controlled rotations $`\text{diag}(1,1,1,e^{i2\pi /2^k})`$ used in the quantum factoring algorithm , can be constructed with the present scheme.
The structure of the paper is as follows. First, in Section II, we define one-bit teleportation, and describe its properties and various guises. Its application to fault-tolerant gate construction is presented in Section III, which is followed in Section IV with specific circuits for the $`\pi /8`$, controlled-phase, and Toffoli gates. In Section V, we describe the use of one-bit teleportation to derive the two-bit quantum teleportation protocol and to construct a remote quantum gate. We summarize our results in Section VI.
## II One-bit teleportation
In standard quantum teleportation, Alice performs a joint measurement of the unknown qubit and some ancilla, and sends the classical measurement outcome to Bob, who subsequently reconstructs the unknown state. No quantum operation is performed jointly by Alice and Bob, but they need a certain two-qubit entangled ancilla state. (We refer to any state other than the original unknown qubit state as ancilla state.) The same objective, communicating a qubit, can be accomplished in a simpler manner if Alice and Bob are allowed to perform a quantum gate (such as a controlled-not gate, a cnot) between their respective qubits. In this case, only a single qubit ancilla in Bob’s possession is required. We call such a quantum circuit one-bit teleportation, which can be derived using the following facts:
* Fact 1: An unknown qubit state $`|\psi `$ can be swapped with the state $`|0`$ using only two cnot gates, as shown in the following circuit:
(1)
Note that in all circuits we show, time proceeds from left to right as is usual, and conventions are as in . Throughout this section, the first and second qubits refer to the registers with respective initial states $`|0`$ and $`|\psi `$.
* Fact 2: $`X=HZH`$, where $`X`$ and $`Z`$ are Pauli operators, and $`H`$ is the Hadamard gate defined as
$$H=\frac{1}{\sqrt{2}}\left[\begin{array}{cc}1& 1\\ 1& 1\end{array}\right].$$
(2)
Then Eq. (1) is equivalent to the following circuit:
(3)
* Fact 3: A quantum-controlled gate can be replaced by a classically-controlled operation when the control qubit is measured.
(4)
The meter represents the measurement of $`Z`$, which projects the measured state onto $`|0`$ or $`|1`$. The double line coming out of the meter carries the classical measurement result, and $`U`$ is performed if the measurement result is $`|1`$.
In Eq. (3), the two qubits are disentangled before the second Hadamard gate. Therefore, the second qubit can be measured before the second Hadamard gate without affecting the unknown state in the first qubit. Applying fact 3 to Eq. (3) results in the following circuit:
(5)
The circuit in Eq. (5) uses a cnot and only one qubit for the ancilla. Therefore it is a one-bit teleportation circuit, which we refer to as “$`Z`$-teleportation” because a classically-controlled-$`Z`$ is applied after the measurement.
Using $`Z`$-teleportation we can derive other one-bit teleportation circuits. For instance, the following circuit first teleports the state $`H|\psi `$ using $`Z`$-teleportation, and then applies $`H^{}=H`$ to the teleported state $`H|\psi `$ to obtain the original state $`|\psi `$:
(6)
This circuit can be simplified to
(7)
which we refer to as “$`X`$-teleportation”. Similarly, we can derive other one-bit teleportation circuits as discussed in Appendix A. We will focus on $`X`$ and $`Z`$-teleportation circuits because they are sufficient for our construction in this paper.
$`X`$ and $`Z`$-teleportation circuits can both be represented using the same general structure:
(8)
where the first qubit (the ancilla qubit) is initially in the $`|0`$ state. For $`Z`$-teleportation, $`A=I`$ ($`I`$ is the $`2\times 2`$ identity operator), $`B=H,D=Z`$, and $`E`$ is a cnot with the first qubit as its target. For $`X`$-teleportation, $`A=H,B=I,D=X`$, and $`E`$ is a cnot with the first qubit as its control.
## III Fault-tolerant gate construction using one-bit teleportation
In this section, we develop a general method for fault-tolerant gate construction using one-bit teleportation as a basic primitive. We will confine our attention to the Calderbank-Shor-Steane (CSS) codes that are doubly even and self-dual , although the results can be extended to any other stabilizer codes .
### A Fault-tolerant gate hierarchy
We first summarize the fault-tolerant gate hierarchy introduced in . Let $`C_1`$ denote the Pauli group. Then for $`k2`$, we can recursively define $`C_k`$ as
$$C_k\{U|UC_1U^{}C_{k1}\}.$$
(9)
For every $`k`$, $`C_kC_{k1}`$, and the set difference $`C_k\backslash C_{k1}`$ is nonempty. For instance, $`\text{diag}(1,e^{i2\pi /2^k})C_k\backslash C_{k1}`$.
$`C_2`$ is a group called the Clifford group , which is the set of operators that conjugate Pauli operators into Pauli operators. Besides the Pauli operators, $`C_2`$ also contains other important gates, such as the cnot, $`H`$, and the phase gate $`S`$ (defined by $`S|x=i^x|x`$ for $`x\{0,1\}`$). For doubly even and self-dual CSS codes, any encoded $`C_2`$ gate has transversal unitary implementation , which is fault-tolerant.
$`C_2`$ gates alone, however, are not sufficient for universal quantum computation . An additional gate outside $`C_2`$ is necessary and sufficient to complete universality . In particular, adding any one of the following gates in $`C_3\backslash C_2`$ to the Clifford group results in a universal set of unitary operations: the $`\pi /8`$ gate $`T`$($`T|x=e^{i\pi x/4}|x`$ for $`x\{0,1\}`$), the controlled-phase gate $`\mathrm{\Lambda }_1(S)`$ ($`\mathrm{\Lambda }_1(S)|xy=i^{xy}|xy`$ for $`x,y=\{0,1\}`$), and the Toffoli gate (controlled-controlled-not.
The construction of an encoded operation in $`C_3\backslash C_2`$ is much more complicated than that of an encoded operation in $`C_2`$, and requires quantum measurement and a particular ancilla state. But applying a $`C_3\backslash C_2`$ gate to certain known states can be replaced by direct preparation of the final states, which can be relatively easier as stated in the following:
Theorem 1: Let $`U`$ be an $`n`$-qubit gate in $`C_3`$. Then the encoded state $`U(|0^n)`$ can be prepared fault-tolerantly by applying and measuring $`C_2`$ operators.
Proof: See Appendix B.
Since $`C_3`$ is closed under multiplication by elements in $`C_2`$ , Theorem 1 is also applicable when $`|0^n`$ is replaced by $`V(|0^n)`$ for $`VC_2`$, because $`U|\psi =UV(|0^n)`$ with $`UVC_3`$. We will use Theorem 1 in our fault-tolerant logic gate construction.
### B $`C_3`$ gate construction using one-bit teleportation
We now consider a general method of constructing fault-tolerant gates in $`C_3`$ using the one-bit teleportation scheme as a primitive. The basic idea is the following. To apply the encoded operation $`U`$ to an encoded state $`|\psi `$, we can first teleport $`|\psi `$ by either $`X`$ or $`Z`$-teleportation, and apply $`U`$ to the reconstructed $`|\psi `$. The extra teleportation step can be done fault-tolerantly because both $`X`$ and $`Z`$-teleportation use $`C_2`$ gates only. It further reduces the problem of fault-tolerant construction of a quantum logic gate to fault-tolerant preparation of a particular ancilla state. The reason for the reduction is that $`U`$ is applied to the ancilla, which is originally in the known state $`|0`$. If $`U`$ can be commuted backwards until it is applied to a known state without introducing more complicated gates, we can prepare the resulting known state, without applying $`U`$ directly, as the input ancilla. Using such an ancilla, the reconstructed state after the modified one-bit teleportation circuit will be $`U|\psi `$. That is, the encoded $`U`$ has been applied to the encoded $`|\psi `$ fault-tolerantly.
We now detail the formal construction. Let $`UC_3`$ be an $`n`$-qubit gate to be applied to $`|\psi `$, an encoded quantum state with $`n`$ logical qubits. We first teleport each logical qubit using either $`X`$ or $`Z`$-teleportation such that $`|\psi `$ is reconstructed in the ancilla, which is initially in the $`|0^n`$ state. We then apply $`U`$ to the reconstructed $`|\psi `$ to obtain $`U|\psi `$. This is described by the following quantum circuit:
(10)
In Eq. (10), a register (wire) with the symbol “/<sup>n</sup>” represents a bundle of $`n`$ logical qubits. $`A`$ is a bitwise operation, $`A=A_1\mathrm{}A_n`$, where $`A_i`$ acts on the $`i^{th}`$ logical qubit only. $`B`$ is a bitwise operation similar to $`A`$. $`E`$ is a tensor product such that $`E=E_i\mathrm{}E_n`$, where each $`E_i`$ is a cnot between the $`i^{th}`$ logical qubits of $`|\psi `$ and the known ancilla. The measurement box measures $`Z`$ bitwise and the double line represents the $`n`$-bit classical outcome. The $`i^{th}`$ classical bit controls whether an operator $`D_i`$ is performed on the $`i`$<sup>th</sup> logical state in the first register. This is denoted by $`D`$ for the sake of simplicity.
According to Sec. II, if $`Z`$-teleportation is applied to the $`i`$<sup>th</sup> logical qubit, $`A_i=I,B_i=H,D_i=Z`$, and $`E_i`$ is a cnot with the first qubit as its target; if $`X`$-teleportation is applied instead, $`A_i=H,B_i=I,D_i=X`$, and $`E_i`$ is a cnot with the first qubit as its control.
We now commute $`U`$ backwards in time. Commuting $`U`$ with the classically-controlled operation $`D`$ changes $`D`$ to $`UDU^{}`$. As $`DC_1`$ and $`UC_3`$, $`UDU^{}C_2`$ can still be performed transversally. Likewise, commuting $`U`$ with $`E`$ changes $`E`$ to $`UEU^{}`$ . As cnot $`C_1`$, the resulting operation $`UEU^{}`$ may not be in $`C_2`$ for an arbitrary $`UC_3`$. To ensure $`UEU^{}C_2`$, we only consider $`U`$ that commutes with $`E`$ such that $`UEU^{}=EC_2`$. Then Eq. (10) becomes
(11)
All the circuit elements outside the dotted box can be performed fault-tolerantly. Therefore, if we can prepare the input ancilla in the state $`UA(|0^n)`$, we can apply $`UC_3`$ to any encoded state $`|\psi `$ fault-tolerantly. As $`AC_2`$, $`UA`$ is also a $`C_3`$ operation. By Theorem 1, the ancilla state $`UA(|0^n)`$ can be created fault-tolerantly. The stabilizers of such an ancilla state, which will be measured in preparing the state, can be easily derived. Recall that when $`A_i=I`$, $`D_i=Z_i`$, and when $`A_i=H`$, $`D_i=X_i`$. Therefore, $`A_iZ_iA_i^{}=D_i`$ is always true , and the stabilizers of $`UA(|0^n)`$ are $`UA_iZ_iA_i^{}U^{}=UD_iU^{}C_2`$.
Using the above method, we can systematically construct interesting gates in $`C_3\backslash C_2`$, including the $`\pi /8`$, controlled-phase, and Toffoli gate, as will be shown in Section IV.
Finally, we remark that the $`C_3`$ gates commuting with $`E`$ are not the only gates that can be performed by the one-bit teleportation scheme. Any $`C_3`$ gate of the form $`U=G_bVG_a`$ for $`V`$ commuting with $`E`$ and $`G_a,G_bC_2`$ can be performed using the generalized one-bit teleportation circuits. We discuss this in Appendix A.
### C Recursive construction
In this section, we extend our discussion to the gates in $`C_k`$ and characterize a class of gates that can be recursively constructed with one-bit teleportation as a basic primitive.
We prove by induction that the diagonal subset of $`C_k`$, defined by $`F_k=\{UC_k\text{ and }U\text{ is diagonal}\}`$, can be recursively constructed.
First, when $`UF_k`$, we choose to apply $`X`$-teleportation to each logical qubit. In this case, each $`E_i`$ is a cnot taking the $`i^{th}`$ logical qubit in the ancilla as the control bit. Therefore, $`E`$ commutes with $`U`$ and Eq. (11) holds with $`A_i=H,B_i=I`$ and $`D_i=X_i`$ for $`i=1,\mathrm{},n`$. Second, since for $`UF_k`$ and $`PC_1`$, $`UPU^{}=\stackrel{~}{U}P`$ for some $`\stackrel{~}{U}F_{k1}`$ , $`UD_iU^{}=UX_iU^{}=U_xX_i`$ for some $`U_xF_{k1}`$. Therefore, if the gates in $`F_{k1}`$ can be performed, the classically-controlled operation $`UD_iU^{}`$ for $`UF_k`$ can also be performed. Third, the required ancilla $`UH^n(|0^n)`$ can be prepared fault-tolerantly with recursive construction as shown in Appendix B. Finally, the gates in $`F_2C_2`$ have transversal implementation. By induction, all the gates in $`F_k`$ can be performed fault-tolerantly with recursive application of the one-bit teleportation scheme.
The sets $`F_k`$ contain many interesting gates, such as $`V^k=\text{diag}(1,e^{i\pi /2^k})`$, which are the single qubit $`\pi /2^k`$ rotations, and $`\mathrm{\Lambda }_1(V^{k1})=\text{diag}(1,1,1,e^{i\pi /2^{k1}})`$, which are the controlled rotations used in the quantum Fourier transform circuit essential to Shor’s factoring algorithm . $`F_k`$ also includes the multiple-qubit gates $`\mathrm{\Lambda }_n(V^l)`$ for $`n+lk`$ , where $`\mathrm{\Lambda }_n(V^l)`$ applies $`V^l`$ to the $`(n+1)^{th}`$ qubit if and only if the first $`n`$ qubits are all in the state $`|1`$. By the closure property of $`F_k`$ , all products of $`\mathrm{\Lambda }_n(V^l)`$ for $`n+lk`$ are in $`F_k`$. To perform gates in $`F_k`$ for $`k`$ small, the recursive construction we have described can be more efficient than approximating these gates to the same accuracy using a universal set of fault-tolerant quantum logic gates.
The gates in $`F_k`$ are not the only ones that can be constructed using the one-bit teleportation scheme. For instance, if $`UC_k`$ is related to an element in $`F_k`$ by conjugation with Hadamard gates in the $`i_1^{th}`$, $`\mathrm{}`$, $`i_l^{th}`$ qubits, $`E`$ can be made to commute with $`U`$ by applying $`Z`$ teleportation to the $`i_1^{th}`$, $`\mathrm{}`$, $`i_l^{th}`$ qubits and $`X`$-teleportation to the rest. The Toffoli gate is an example. More generally, any gate that is a product of $`C_2`$ gates and a single $`F_k`$ gate can be constructed recursively.
## IV Examples
In this section, we systematically construct three important fault-tolerant gates in $`C_3\backslash C_2`$ using the general method described in Sec. III. Any one of these gates, together with the Clifford group, forms a universal set of gates. For each of the construction, we will derive the required circuit and the ancilla. The ancilla can always be prepared fault-tolerantly (see Appendix B).
### A The $`\pi /8`$ gate
The $`\pi /8`$ gate, $`T`$, has the following matrix representation:
$$T=\left[\begin{array}{cc}1& 0\\ 0& e^{i\pi /4}\end{array}\right].$$
(12)
As $`T`$ is diagonal, following the recipe in Sec. III, we choose to apply $`X`$-teleportation to $`|\psi `$ and apply $`T`$ to the teleported $`|\psi `$:
(13)
We commute $`T`$ backwards using two facts. First, $`TXT^{}=e^{i\pi /4}SX`$, where the phase gate $`S`$ is defined in Sec. III A. Second, $`T`$ commutes with the $`\mathrm{cnot}`$ by construction. Thus, we obtain a circuit to implement the $`\pi /8`$ gate (where an irrelevant overall phase has been ignored):
(14)
All the circuit elements outside the dotted box can be performed fault-tolerantly. The dotted box, then, can be replaced by an ancilla in the state
$$|\varphi _+=TH|0=\frac{|0+e^{i\pi /4}|1}{\sqrt{2}},$$
(15)
which can be prepared fault-tolerantly as described in Appendix B. Thus, we have derived a circuit and the corresponding ancilla for performing the fault-tolerant $`\pi /8`$ gate. We note that this re-derives the same circuit and ancilla state used in .
### B The controlled–phase gate
The controlled-phase gate $`\mathrm{\Lambda }_1(S)`$ (defined in Sec. III A) is in $`C_3`$, and forms a universal set of gates together with $`H`$ and cnot.
We use the following circuit symbol for $`\mathrm{\Lambda }_1(S)`$:
(16)
$`\mathrm{\Lambda }_1(S)`$ commutes with $`Z_i`$, and conjugates $`X_i`$ ($`i=1,2`$) as follows:
(17)
(18)
where the controlled-$`Z`$ operation, $`\mathrm{\Lambda }_1(Z)`$, acts on basis states as $`\mathrm{\Lambda }_1(Z)|x|y=(1)^{xy}|x|y`$. To construct $`\mathrm{\Lambda }_1(S)`$, we first teleport the two-qubit state $`|\psi `$ and apply $`\mathrm{\Lambda }_1(S)`$. This linear transformation preserves phase coherence, and thus, it suffices to consider its action on the basis states $`|xy`$. Since $`\mathrm{\Lambda }_1(S)`$ is diagonal, we apply $`X`$-teleportation to both qubit states such that the cnots in the circuit commute with $`\mathrm{\Lambda }_1(S)`$.
(19)
Commuting $`\mathrm{\Lambda }_1(S)`$ backwards using the commutation rules in Eqs. (17)-(18), we obtain a circuit for the controlled-phase gate:
(20)
where the double lines control all the operations in the corresponding boxes. All the circuit elements in Eq. (20), except those in the dotted box, can be performed fault-tolerantly. Finally, we can replace the dotted box by an input ancilla in the following state:
$`|\varphi _+`$ $`=`$ $`\mathrm{\Lambda }_1(S)(H_1H_2)|00`$ (21)
$`=`$ $`{\displaystyle \frac{1}{2}}(|00+|01+|10+i|11),`$ (22)
which can be prepared fault-tolerantly. This completes the requirement for performing the controlled-phase gate fault-tolerantly.
### C The Toffoli gate
To construct the Toffoli gate (controlled-controlled-not), we begin with some useful commutation rules:
(23)
(24)
As in the controlled-phase gate construction, we demonstrate the construction on basis states $`|xyz`$ for three qubits. We first teleport $`|xyz`$ and then apply a Toffoli gate. Since the Toffoli gate is diagonalized by a Hadamard transform on the target qubit, the choice of $`X`$-teleportation for the control qubits and $`Z`$-teleportation for the target qubit ensures that the three cnots commute with the Toffoli gate.
(25)
Commuting the Toffoli gate backwards to the far left using Eqs. (23)-(24), Eq. (25) is equivalent to
(26)
All the circuit elements except those in the dotted box can be performed fault-tolerantly. It remains to prepare the state created in the dotted box,
$`|\varphi _+`$ $`=`$ $`U(H_1H_2)|000`$ (27)
$`=`$ $`{\displaystyle \frac{1}{2}}(|000+|010+|100+|111),`$ (28)
where $`U`$ denotes the Toffoli gate. Again this ancilla state can be prepared fault-tolerantly, as described in Appendix B.
The ancilla and the quantum circuit derived here are the same as those in Shor’s original construction . The one-bit teleportation scheme elucidates the choice of the ancilla state and the procedure in .
## V Remote gate construction using one-bit teleportation
The one-bit teleportation scheme, in addition to being useful for fault-tolerant gate construction, can also be used to design a variety of remote quantum operations. Constructing remote quantum operations is related to constructing fault-tolerant gates in that both require a particular ancilla state to replace a prohibited operation. In this section, we use one-bit teleportation as a basic primitive to derive the quantum circuits and the required ancilla states for the two-bit quantum teleportation and the remote cnot.
### A Two-bit teleportation
Suppose Alice needs to send a qubit state $`|\psi `$ to Bob. Direct quantum communication is not allowed, but Alice and Bob can share some ancilla state. The question is, how can Alice send $`|\psi `$ to Bob? A well-known solution to this problem is quantum teleportation , which uses an EPR state and classical communication. Using one-bit teleportation, we give an alternative derivation of the required (entangled) ancilla and the required teleportation circuit.
We first construct a circuit to send the unknown state with a prohibited operation. Then we remove the requirement of such a prohibited operation. Let $`|\psi `$ be the state to be communicated from Alice to Bob. Alice can send $`|\psi `$ to Bob by applying one-bit teleportation twice. Step 1: Alice swaps $`|\psi `$ with an ancilla $`|0`$ using $`X`$-teleportation. Step 2: Alice sends the teleported $`|\psi `$ to Bob using $`Z`$-teleportation (with a prohibited cnot in this step). The circuit representation for the process is
(29)
The prohibited operation (cnot), which is marked by an asterisk, can be commuted backwards using the commutation relation:
(30)
This leads to the usual quantum teleportation circuit
(31)
In Eq. (31), the prohibited cnot acts on the known state inside the dotted box, which can be replaced by the following state it creates:
$$|\varphi =\mathrm{\Lambda }_1(X)H_1|00=\frac{1}{\sqrt{2}}(|00+|11)$$
(32)
In other words, if Alice shares this entangled state with Bob, the state $`|\psi `$ can be sent to Bob without quantum communication.
Note that the classically-controlled-$`X`$ on the second register only affects its overall sign, and can be omitted since the second register is subsequently measured.
An alternative circuit, which accomplishes the same task, can be derived when $`Z`$ and $`X`$-teleportation are used for the two steps instead. We start with the following circuit:
(33)
Using the commutation rule
(34)
we can commute the prohibited cnot backwards to obtain an equivalent quantum teleportation circuit
(35)
The disallowed element in the dotted box can be replaced by the EPR state of Eq. (32). The irrelevant classically-controlled-$`Z`$ on the second register can be omitted.
The two-bit teleportation circuits of Eqs. (31) and (35) are equivalent to that in , but as mentioned above, they are derived differently.
### B Remote cnot
Suppose Alice and Bob have in their possession quantum states $`|\alpha `$ and $`|\beta `$, respectively. How can they perform a simple distributed computation, a cnot from $`|\alpha `$ to $`|\beta `$, without communicating any quantum information between them, but perhaps with the aid of some initially shared standard quantum state? A solution to this problem is given in . The ad-hoc method employed, however, does not suggest a systematic technique for deriving the solution, or solutions to generalized versions of this problem. Here, we use one-bit teleportation to present a general technique and derive a different circuit that accomplishes the same task.
Alice and Bob first swap their states with their respective ancilla state $`|0`$ by one-bit teleportation, and then apply a prohibited cnot. The quantum circuit is chosen so that Alice uses $`X`$-teleportation and Bob uses $`Z`$-teleportation:
(36)
The prohibited cnot can be commuted backwards to obtain a remote cnot circuit:
(37)
The prohibited operation in the dotted box is applied to a known state, and can be replaced by the EPR state of Eq. (32). Provided such a shared entangled state is initially available to Alice and Bob, they can perform a remote cnot operation using two bits of classical communication.
Note that a remote cnot can also be constructed by using two-bit teleportation twice in an obvious way: Bob first sends his qubit $`|\beta `$ to Alice with two-bit teleportation, and then Alice applies cnot to $`|\alpha |\beta `$ and sends the qubit $`|\alpha \beta `$ to Bob with two-bit teleportation. Such construction, however, requires two pairs of maximally entangled state and four bits of classical communication, which is twice that required for the one-bit teleportation scheme.
Our remote cnot construction in Eq. (37) is different from that in , which can also be derived using the one-bit teleportation scheme, as described in Appendix C.
Finally, we remark that the two examples of constructing remote operations strengthen the concept of teleporting quantum logic gates with one-bit teleportation, as we have shown that if the input ancilla is a special state related to the cnot gate, the reconstructed state is the one to which a cnot gate has been applied.
## VI Conclusion
We have presented a systematic technique to construct a variety of quantum operations, by using a primitive one-bit teleportation scheme. Such a scheme reduces the problem of constructing a quantum logic gate to preparing an ancilla state created by the same gate applied to a known state. The usefulness of this technique is particularly manifest for two kinds of application: fault-tolerant quantum computation and remote quantum computation, as demonstrated in our construction of the $`\pi /8`$, controlled-phase, and Toffoli gates, and the remote-cnot. With recursive application of the one-bit teleportation scheme, we can also construct an infinite hierarchy of gates fault-tolerantly.
The idea of teleporting quantum logic gates has been used in , with two-bit teleportation as a primitive, to perform universal quantum computation. The two-bit teleportation scheme allows all $`C_3`$ gates to be teleported fault-tolerantly, and all $`C_k`$ gates to be teleported with recursive application of the scheme. For one-bit teleportation, however, we can only provide sufficient conditions for gates in $`C_3`$ to be teleportable, namely, any $`C_3`$ gate that can be written as a product of $`C_2`$ gates and a single $`C_3`$ gate that commutes with cnot. It is not known if this includes all the $`C_3`$ gates. The difficulty in describing the exact set of one-bit teleportable $`C_3`$ gates arises from the requirement for a $`C_2\backslash C_1`$ gate in the one-bit teleportation circuit. Such a $`C_2\backslash C_1`$ gate may be conjugated outside $`C_2`$ by a $`C_3`$ gate, and therefore cannot be directly performed fault-tolerantly. This places further constraint on the teleportable $`UC_k`$ for $`k>3`$. Because of our present lack of understanding of the general structure and nature of $`C_k`$ gates, the distinction between the ultimate capabilities of the one and two-bit teleportation schemes remains an interesting and difficult open question.
Nevertheless, as we have shown, one-bit teleportation can provide much simpler protocols than two-bit teleportation in constructing quantum logic gates. This is because one-bit teleportation only requires projective measurement of $`Z`$ and as many ancilla qubits as the state to be transformed; two-bit teleportation, however, requires Bell measurement and twice as many ancilla qubits as the original state.
At a very general level, the logical gate teleportation schemes reduce the difficulty of constructing quantum logic gates by using special ancilla states. This can be useful not only for simplifying hardware requirements, but also for designing and optimizing computation and communication protocols . Even more intriguing, perhaps, is that this result gives us a first glimpse at what might someday be a standard architecture for a quantum computer: a simple assembly of one-bit teleportation primitives, capable of universal quantum computation on quantum data, given the assistance of standard quantum states that are obtained as commercial resources. The definition of such a stored-program architecture could be pivotal in the development of this field, much as the von Neumann or Harvard architecture were important in classical computation.
## VII Acknowledgments
The relation between fault-tolerant quantum logic gate construction and teleportation is first alluded to by Shor . The $`X`$ and $`Z`$ teleportation circuits presented in this paper are due to Charles Bennett and Daniel Gottesman (unpublished). We are grateful to Daniel Gottesman for introducing us to the interesting subject of the $`C_k`$ hierarchy, and for enlightening discussions. We thank Professor James Harris and Yoshihisa Yamamoto for support and encouragement. This work was supported by the DARPA Ultra-scale Program under the NMRQC initiative, contract DAAG55-97-1-0341, administered by the Army Research Office. D.L. acknowledges support from the IBM Fellowship program and Nippon Telegraph and Telephone Corporation (NTT).
## A Generalizations of the one-bit teleportation circuits
The one-bit teleportation circuit used in fault-tolerant gate construction has three components: a particular input ancilla, a sequence of $`C_2`$ gates, and finally the measurement and classically-controlled operation. The teleportability of one-bit teleportation is governed by the sequence of $`C_2`$ gates before the measurement. Using the $`X`$ and $`Z`$-teleportation circuits of Eq. (11), any $`UC_3`$ that commutes with $`E`$ can be teleported. In this appendix, we derive other one-bit teleportation circuits, which use different $`C_2`$ gates, and then discuss their application in constructing fault-tolerant gates.
By teleporting $`G|\psi `$ using $`X`$-teleportation and applying $`G^{}`$ to the teleported $`G|\psi `$, we obtain the following generalized one-bit telelportation circuit:
(A1)
When $`G=I^n`$ and $`H^n`$, Eq. (A1) reduces to the $`X`$ and $`Z`$-teleportation circuits.
In Sec. III, we showed that all the operations in $`F_3`$ can be performed fault-tolerantly using $`X`$-teleportation. Here, we generalize the result to show that, if $`UC_3`$ and $`U=G_bVG_a`$, where $`VF_3`$ and $`G_a,G_bC_2`$, then $`U`$ can be performed fault-tolerantly using the general one-bit teleportation scheme by the following procedure:
Step 1: Using the circuit of Eq. (A1) with $`G=G_a`$, we first teleport the state $`|\psi `$ to the ancilla initialized in the state $`|0^n`$, and then apply $`U`$ to the ancilla. This can be represented by
(A2)
Step 2: Commuting $`U`$ backwards, one obtains
(A3)
Note that the new classically-controlled operation is $`G_bVX^nV^{}G_b^{}`$, which is in $`C_2`$ because $`VX^nV^{}C_2`$. Therefore, all the circuit elements can be performed fault-tolerantly, except those in the dotted box, which can be replaced by an ancilla in the state $`VH^n(|0^n)`$.
There are $`C_3`$ gates that cannot be constructed using $`X`$ and $`Z`$-teleportation directly, but can be constructed using other one-bit teleportation circuits. For instance, the controlled-Hadamard gate $`\mathrm{\Lambda }_1(H_2)C_3\backslash C_2`$ does not commute with $`E`$ in Eq. (10) for all possible combinations of $`X`$ and $`Z`$-teleportation circuits, but $`\mathrm{\Lambda }_1(H_2)`$ can be written as $`G_bVG_a`$ with $`G_a=Q_2^{}`$, $`G_b=\mathrm{\Lambda }_1(X_2)Q_2`$ and $`V=T_1\mathrm{\Lambda }_1(S_2^{})`$, where $`Q=S^{}HSC_2`$. Thus, $`\mathrm{\Lambda }_1(H_2)`$ can still be performed using the general one-bit teleportation scheme.
We remark that a $`C_3`$ gate $`U=G_bVG_a`$ with $`G_a,G_bC_2`$ and $`VF_3`$ can be performed indirectly by applying $`G_a,V`$ and $`G_b`$ in sequence, where $`V`$ is applied by $`X`$-teleportation. If the operations in the generalized one-bit teleportation circuit, $`G_b,\mathrm{\Lambda }_1(X)`$, and $`G_a`$ of Eq. (A3), are also considered, the total requirements to perform $`U`$ by such indirect implementation and by direct one-bit teleportation are almost the same. But if we are given different one-bit teleportation circuits as primitives, we can use them to directly teleport different sets of $`C_3`$ gates. In other words, if we are given the circuit of Eq. (11), using an input ancilla in the state $`UA(|0^n)`$, we can teleport $`UC_3`$ that commutes with $`E`$; if we are given the circuit of Eq. (A3), using an input ancilla in the state $`VH^n(|0^n)`$, we can teleport $`U`$ in the form of $`G_bVG_a`$. In this sense, then, the generalized one-bit teleportation circuits are interesting and allow more gates in $`C_3`$ to be teleported directly.
## B Fault-tolerant state preparation
In this section, we first prove Theorem 1 in Sec. III A by construction. We then show how to create the three ancilla states in Sec. IV fault-tolerantly. Finally, we explain how to prepare a class of encoded quantum states fault-tolerantly by recursive application of the one-bit teleportation scheme.
### 1 Fault-tolerant preparation of quantum states
A stabilizer of a quantum state is a quantum operator that transforms the state to itself. Let $`𝒞`$ be the codeword space corresponding to an $`[[m,n]]`$ stabilizer code, which encodes $`n`$ logical qubits using $`m`$ physical qubits. The stabilizer $`S`$ of $`𝒞`$ is an Abelian subgroup of the Pauli group, or $`C_1`$, such that $`|\psi 𝒞`$ if and only if $`MS,M|\psi =|\psi `$. By performing error correction for the stabilizer code, we can project an arbitrary state onto an encoded state in $`𝒞`$ .
The stabilizer $`S`$ has $`2^{mn}`$ elements generated by $`mn`$ independent operators in $`C_1`$, and defines a quantum code of dimension $`2^n`$. Each encoded state is, then, determined by $`n`$ extra independent stabilizers. (In the following, we will restrict our discussion to the codeword space and exclude the stabilizers of the code from the stabilizers of an encoded state.) For instance, the encoded $`|0^n`$ is determined by $`Z_i`$ for $`i=1,\mathrm{},n`$, where $`Z_i`$ is the encoded $`Z`$ on each logic qubit. In general, stabilizers need not commute with one another and need not square to the identity. But an independent set of stabilizers can always be chosen to be a mutually commuting set of elements that square to the identity. This is because $`|\psi =U(|0^n)`$ for some encoded $`U`$, leading to a possible choice $`\text{St}(|\psi )\{UZ_iU^{},i=1,\mathrm{},n\}`$ with the desired properties. We restate the above as a lemma:
Lemma 1: For any $`|\psi `$, $`\text{St}(|\psi )`$ can be chosen such that $`M,N\text{St}(|\psi )`$, (a) $`M^2=I`$ and (b) $`[M,N]=MNNM=0`$.
Note that the elements in $`\text{St}(|\psi )`$ are all valid encoded operations, and their actions preserve the codeword space.
As a quantum state is the simultaneous $`+1`$ eigenstate of its stabilizers, the state can be prepared by projecting an arbitrary encoded state onto the simultaneous $`+1`$ eigenstate of its stabilizers. In the following we will show how to create a class of quantum states fault-tolerantly by measuring their stabilizers.
Given a quantum state $`|\psi `$ encoded with an $`[[m,n]]`$ stabilizer code, the operator $`MC_2`$ with $`M^2=I`$ can be measured fault-tolerantly on $`|\psi `$ as follows. First, we prepare a cat state
$$|\text{cat}\frac{1}{\sqrt{2}}(|\overline{0}+|\overline{1}),$$
(B1)
where $`|\overline{i}`$ consists of $`m`$ physical qubits in the state $`|i`$ ($`i=0,1`$). (The cat state cannot be created fault-tolerantly, but it can always be verified .) For the doubly even and self-dual CSS codes, the encoded $`MC_2`$ can be written as $`M=M^1\mathrm{}M^m`$, where $`M^j`$ acts only on the $`j^{th}`$ physical qubit of each block of the encoded state $`|\psi `$. For each $`j`$, we perform controlled-$`M^j`$ with the $`j^{th}`$ qubit of the cat state as the control bit and the $`j^{th}`$ qubit of $`|\psi `$ as the target qubit. Effectively, a cat-state-controlled-$`M`$ is applied to the state $`|\text{cat}|\psi `$ with transversal operations leading to the state
$`{\displaystyle \frac{1}{\sqrt{2}}}|\overline{0}|\psi +{\displaystyle \frac{1}{\sqrt{2}}}|\overline{1}M|\psi `$ (B2)
$`=`$ $`{\displaystyle \frac{1}{2}}(|\overline{0}+|\overline{1})(I+M)|\psi +{\displaystyle \frac{1}{2}}(|\overline{0}|\overline{1})(IM)|\psi .`$ (B3)
Note that as $`M^2=I`$, $`(I\pm M)|\psi `$ are $`\pm 1`$ eigenstates of $`M`$ for any $`|\psi `$.
We can measure the cat state fault-tolerantly using the procedure described in to distinguish $`|\overline{0}+|\overline{1}`$ from $`|\overline{0}|\overline{1}`$. (We omit the unimportant normalization factors.) If we obtain $`|\overline{0}+|\overline{1}`$, the encoded state is projected onto $`(I+M)|\psi `$, the $`+1`$ eigenstate of $`M`$; otherwise the resulting encoded state is $`(IM)|\psi `$, the $`1`$ eigenstate of $`M`$, which may be transformed to a $`+1`$ eigenstate of $`M`$ by the following Lemma.
Lemma 2: If $`MC_2,M^2=I`$, and there exists $`QC_2`$ such that $`\{M,Q\}=MQ+QM=0`$, then we can always transform an arbitrary encoded state $`|\psi `$ onto a $`+1`$ eigenstate of $`M`$ using fault-tolerant operations. The resulting $`+1`$ eigenstate is either $`(I+M)|\psi `$ or $`(I+M)Q|\psi `$, which can be written jointly as $`(I+M)Q^a|\psi `$ for $`a=0`$ or $`1`$.
Proof: We have shown that we can project an arbitrary encoded state $`|\psi `$ onto $`(I\pm M)|\psi `$, the $`\pm 1`$ eigenstate of $`M`$. If the resulting state is $`(I+M)|\psi `$, we are done; otherwise, we apply $`QC_2`$ fault-tolerantly to $`(IM)|\psi `$. Since $`Q`$ anticommutes with $`M`$, it transforms the $`1`$ eigenstate of $`M`$ to a $`+1`$ eigenstate of $`M`$ as follows: $`Q(IM)|\psi =(I+M)(Q|\psi )`$. Thus, we can always obtain a $`+1`$ eigenstate of $`M`$, which is $`(I+M)Q^a|\psi `$ for $`a=0`$ or $`1`$.
Next we will show that a special class of quantum states can be created fault-tolerantly.
Lemma 3: If $`\text{St}(|\psi )=\{M_1,\mathrm{},M_n\}C_2`$ and $`M_i\text{St}(|\psi )`$ there exists $`Q_iC_2`$ such that $`\{M_i,Q_i\}=0`$, and $`[M_i,Q_j]=0`$ for $`ij`$, then $`|\psi `$ can be created fault-tolerantly by measuring the elements in $`\text{St}(|\psi )`$ fault-tolerantly.
Proof: By Lemma 1, $`i`$, $`M_i^2=I`$. Starting from any encoded state $`|\varphi `$, we measure $`M_1,\mathrm{},M_n`$ sequentially, and after each measurement we apply the corresponding operation $`Q_i`$ if the projected state is the $`1`$ eigenstate of $`M_i`$. By Lemma 2, the resulting state is
$`|\psi `$ $`=`$ $`(I+M_n)Q_n^{a_n}\mathrm{}(I+M_1)Q_1^{a_1}|\varphi `$ (B4)
$`=`$ $`(I+M_n)\mathrm{}(I+M_1)Q_n^{a_n}\mathrm{}Q_1^{a_1}|\varphi ,`$ (B5)
where $`a_i=0`$ or $`1`$, and we have used the fact that $`[M_i,Q_j]=0`$ for $`ij`$. As $`[M_i,M_j]=0`$, it is easily verified that $`i,M_i|\psi =|\psi `$, and $`|\psi `$ is the desired state that has been created fault-tolerantly.
Theorem 1 in Sec III A immediately follows:
Theorem 1: $`UC_3`$, $`U`$ can be applied to the encoded $`|0^n`$ state using $`C_2`$ operators and fault-tolerant measurement of $`C_2`$ operators.
Proof: Applying $`UC_3`$ to the encoded $`|0^n`$ state is equivalent to preparing the state $`|\psi =U(|0^n)`$, which has stabilizers $`M_i=UZ_iU^{}`$ for $`i=1,\mathrm{},n`$. Define $`Q_iUX_iU^{}C_2`$ for each $`i`$. Then $`\{Z_i,X_i\}=0`$ implies $`\{M_i,Q_i\}=U\{Z_i,X_i\}U^{}=0`$, and for $`ij`$, $`[Z_i,X_j]=0`$ implies $`[M_i,Q_j]=U[Z_i,X_j]U^{}=0`$. Thus by Lemma 3, the state $`|\psi `$ can be created fault-tolerantly.
### 2 Examples
To prepare a specific encoded state from an unknown encoded state we need to measure all its independent stabilizers. When the initial state is a known encoded state related to the desired state, we may not have to measure all the independent stabilizers. For instance, given two encoded states $`|\varphi `$ and $`|\varphi ^{}`$ with $`\text{St}(|\varphi )=\{M_1,\mathrm{},M_k,M_{k+1},\mathrm{},M_n\}`$ and $`\text{St}(|\varphi ^{})=\{M_1,\mathrm{},M_k,M_{k+1}^{},\mathrm{},M_n^{}\}`$, the following state
$$(I+M_n)\mathrm{}(I+M_{k+1})Q_n^{a_n}\mathrm{}Q_{k+1}^{a_{k+1}}|\varphi ^{}$$
(B6)
is the simultaneous $`+1`$ eigenstate of $`M_i`$ for $`i=1,\mathrm{},n`$. Thus, starting from $`|\varphi ^{}`$, we can prepare the encoded state $`|\varphi `$ by measuring only the $`nk`$ different stabilizers. In the following, we will construct an initial state, with which, the desired state can be obtained by measuring only a single stabilizer.
Assume we want to prepare the encoded state $`|\psi _+=U(|0^n)`$ for $`UC_3`$. Define $`M_i`$ and $`Q_i`$ for $`i=1,\mathrm{},n`$ as in the proof of Theorem 1. Then $`Q_i|\psi _+`$ is a $`1`$ eigenstate of $`M_i`$ such that $`\psi _+|Q_i|\psi _+=0`$, and the following state
$$|\psi =\frac{1}{\sqrt{2}}(|\psi _++Q_i|\psi _+)$$
(B7)
is different from $`|\psi _+`$ by only one independent stabilizer: $`Q_i`$ has replaced $`M_i`$. Therefore, the state $`|\psi `$ also satisfies the conditions of Lemma 3, and can be prepared fault-tolerantly. It follows that to obtain the state $`|\psi _+`$, we only need to measure the single stabilizer $`M_i`$ on $`|\psi `$.
To prepare an encoded state $`|\psi _+`$ by preparing $`|\psi `$ first can be simpler than directly preparing $`|\psi _+`$ from an arbitrary encoded state if $`|\psi `$ itself can be prepared easily. For instance, when $`|\psi `$ is a product state, it can be prepared by measuring only single qubit operators. We will describe how to prepare the required ancilla states for the three gates in Sec. IV. When the required ancilla $`|\psi _+`$ is an entangled state with multiple-qubit stabilizers, we will construct it by preparing an easier state $`|\psi `$ first.
#### a Fault-tolerant preparation of the ancilla required for $`T`$ gate
The required ancilla for constructing the $`\pi /8`$ gate, $`T`$, is
$$|\psi _+=TH|0=\frac{|0+e^{i\pi /4}|1}{\sqrt{2}},$$
(B8)
with stabilizer
$$M=(TH)Z(TH)^{}=e^{i\pi /4}SX,$$
(B9)
which anticommutes with $`(TH)X(TH)^{}=Z`$. Then starting from any encoded state, we can measure $`M`$, and apply $`Z`$ if the projected state is the $`1`$ eigenstate, to create the state $`|\psi _+`$ fault-tolerantly.
#### b Fault-tolerant preparation of the ancilla required for controlled-phase gate
The required ancilla for constructing the controlled phase gate is
$`|\psi _+`$ $`=`$ $`\mathrm{\Lambda }_1(S)(H_1H_2)|00`$ (B10)
$`=`$ $`{\displaystyle \frac{1}{2}}(|00+|01+|10+i|11),`$ (B11)
with stabilizers $`M_i=\mathrm{\Lambda }_1(S)(H_1H_2)Z_i(H_1H_2)\mathrm{\Lambda }_1(S^{})`$ for $`i=1,2`$. Using Eqs. (17)-(18),
$`M_1`$ $`=`$ $`(X_1S_2)\mathrm{\Lambda }(Z),`$ (B12)
$`M_2`$ $`=`$ $`(S_1X_2)\mathrm{\Lambda }(Z).`$ (B13)
The corresponding operator that anticommutes with $`M_i`$ is $`Q_i=\mathrm{\Lambda }_1(S)(H_1H_2)X_i(H_1H_2)\mathrm{\Lambda }_1(S^{})=Z_i`$ for $`i=1,2`$. $`|\psi _+`$ is an entangled state, and both of $`M_1`$ and $`M_2`$ are two-qubit operators. But the following state
$`|\psi `$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|\psi _++Q_1|\psi _+)`$ (B14)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}|0(|0+|1)`$ (B15)
is a product of single qubit states and has stabilizers $`Z_1`$ and $`X_2`$. Thus, we can first prepare $`|\psi `$ fault-tolerantly by measuring $`Z_1`$ and $`X_2`$, and then measure $`M_1`$ alone to get the state $`|\psi _+`$. Equivalently, we can also first prepare the state $`\frac{1}{\sqrt{2}}(|\psi _++Q_2|\psi _+)=\frac{1}{\sqrt{2}}(|0+|1)|0`$, which has stabilizers $`X_2`$ and $`Z_1`$, and measure $`M_2`$ to obtain the state $`|\psi _+`$.
#### c Fault-tolerant preparation of the required ancilla for the Toffoli gate
The required ancilla for the Toffoli gate construction is
$`|\psi _+`$ $`=`$ $`U(H_1H_2)|000`$ (B16)
$`=`$ $`{\displaystyle \frac{1}{2}}(|000+|010+|100+|111),`$ (B17)
where $`U`$ is the Toffoli gate. The stabilizer of this state is $`M_i=U(H_1H_2)Z_i(H_1H_2)U^{}`$ for $`i=1,2,`$ and $`3`$. Using Eqs. (23)-(24),
$`M_1`$ $`=`$ $`X_1\mathrm{cnot}_{23},`$ (B18)
$`M_2`$ $`=`$ $`X_2\mathrm{cnot}_{13},`$ (B19)
$`M_3`$ $`=`$ $`Z_3\mathrm{cz}_{12},`$ (B20)
where cz represents a controlled-$`Z`$, and the ordered subscripts for cnot and cz specifies the control and target bits. The operator that anticommutes with $`M_i`$ is $`Q_i=U(H_1H_2)X_i(H_1H_2)U^{}`$, or $`Z_1,Z_2`$ and $`X_3`$ for $`i=1,2,`$ and $`3`$, respectively. Again, each of $`M_i`$ is a two-qubit operator, but the following state
$$|\psi =\frac{1}{\sqrt{2}}(|\psi _++Q_1|\psi _+)=\frac{1}{\sqrt{2}}|0(|0+|1)|0$$
(B21)
can be prepared easily by measuring its stabilizers $`Z_1,X_2`$ and $`Z_3`$. Then we only need to measure a single two-qubit operator $`M_1`$ on $`|\psi `$ to obtain $`|\psi _+`$. Equivalently, we can also first prepare the state $`|\psi =\frac{1}{\sqrt{2}}(I+Q_2)|\psi _+`$ with stabilizers $`X_1,Z_2`$ and $`Z_3`$ or the state $`|\psi =\frac{1}{\sqrt{2}}(I+Q_3)|\psi _+`$ with stabilizers $`X_1,X_2`$ and $`X_3`$, and measure the corresponding single stabilizer to obtain $`|\psi _+`$.
### 3 Recursive preparation
In this subsection, we will prove the following Theorem, which is used in Sec. III C:
Theorem 2: The encoded state $`|\psi =UH^n(|0^n)`$ for $`UF_k`$ can be prepared fault-tolerantly by recursive application of one-bit teleportation.
First we have the following Lemma, which is a generalization of Lemma 3.
Lemma 4: $`|\psi `$ can be created fault-tolerantly if given $`\text{St}(|\psi )=\{M_1,\mathrm{},M_n\}`$, $`i,j`$
(1) the cat-state-controlled-$`M_i`$ can be performed fault-tolerantly;
(2) there exists $`Q_i`$ such that $`Q_i`$ can be performed fault-tolerantly, $`\{M_i,Q_i\}=0`$, and for $`ij,[M_i,Q_j]=0`$.
Proof: Since $`M_i^2=I`$, by applying the cat-state-controlled-$`M_i`$ and measuring the cat state fault-tolerantly as before, we can project any encoded state onto $`\pm 1`$ eigenstate of $`M_i`$. Then apply $`Q_i`$ if a $`1`$ eigenstate is obtained. Using the same argument as in the proof of Lemma 3, we can fault-tolerantly prepare the state $`|\psi `$.
Lemma 5: If operations in $`F_{k1}`$ and cat-state-controlled-$`V`$ for any $`VF_{k2}`$ can be performed fault-tolerantly using the one-bit teleportation scheme, then $`UH^n(|0^n)`$ for $`UF_k`$ can be created fault-tolerantly.
Proof: The stabilizers of $`|\psi =UH^n(|0^n)`$ are $`M_i=UH^nZ_i(UH^n)^{}=UX_iU^{}=U_xX_i`$ for some $`U_xF_{k1}`$. Define $`Q_iUZ_iU^{}=Z_i`$. $`Q_i`$ satisfies condition (2) of Lemma 4.
Since $`M_i=U_xX_i`$, the cat-state-controlled-$`M_i`$ is the product of cat-state-controlled-$`U_x`$ and cat-state-controlled-$`X_i`$. The cat-state-controlled-$`X_i`$ is easily performed fault-tolerantly. Thus it remains to show how to perform the cat-state-controlled-$`U_x`$ for $`U_xF_{k1}`$ fault-tolerantly.
As $`U_xF_{k1}`$ is constructed with one-bit teleportation scheme using the circuit of Eq. (11), where $`A_i=H,B_i=I`$ and $`D_i=X_i`$, to perform cat-state-controlled-$`U_x`$, we need to perform cat-state-controlled-$`E`$, cat-state-controlled-$`U_xX_iU_x^{}`$, and to prepare the ancilla $`U_xH^n(|0^n)`$ fault-tolerantly. As $`EC_2`$ and $`U_xX_iU_x^{}=U_x^{}X_i`$ with $`U_x^{}F_{k2}`$, both of cat-state-controlled-$`E`$ and cat-state-controlled-$`U_xD_iU_x^{}`$ can be performed fault-tolerantly. Next, the state $`U_xH^n|0^n`$ has stabilizers $`M_i^{}=U_xX_iU_x^{}=U_x^{}X_i`$ with $`U_x^{}F_{k2}`$, which satisfies both conditions of Lemma 4 and can therefore be prepared fault-tolerantly. Thus cat-state-controlled-$`U_x`$ can be performed fault-tolerantly. This completes the proof of Lemma 5.
In fact, what we have shown in the proof of Lemma 5 is that if operations in $`F_{k1}`$ and the cat-state-controlled-$`V`$ for $`VF_{k2}`$ can be performed fault-tolerantly, then the cat-state-controlled-$`U`$ for $`UF_{k1}`$ and operations in $`F_k`$ can be performed fault-tolerantly. This is because according to Sec. III C, fault-tolerant construction of $`F_k`$ gates only require fault-tolerant $`F_{k1}`$ gates and an ancilla $`UH^n(|0^n)`$ for $`UF_k`$.
Since both the operations in $`F_2`$ and the cat-state-controlled-$`U`$ for $`UF_1`$ can be performed fault-tolerantly, by induction, operations in $`F_k`$ and the cat-state-controlled-$`U`$ for $`UF_{k1}`$ can be constructed fault-tolerantly, with which we can fault-tolerantly prepare the encoded state $`UH^n(|0^n)`$ for $`UF_k`$.
## C Alternative remote cnot circuit
In this section, we re-derive the remote cnot construction, given in , using one-bit teleportation. A remote cnot between the states $`|\alpha `$ and $`|\beta `$ belonging to Alice and Bob, respectively, can be performed by a four step procedure: (1) Alice swaps her state $`|\alpha `$ with an ancilla $`|0`$, (2) Alice sends the teleported $`|\alpha `$ to Bob using $`X`$-teleportation, (3) Bob applies cnot from $`|\alpha `$ to $`|\beta `$, and (4) Bob teleports $`|\alpha `$ back to Alice using $`Z`$-teleportation. Steps (2) and (4) involve prohibited operations. Here is a circuit representation:
(C1)
The two prohibited cnots are labelled with asterisks. They can be commuted backwards to obtain the equivalent circuit:
(C2)
which again reduces prohibited operations to some specific shared entangled state.
|
warning/0002/hep-th0002175.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The way one usually relates field-theories to branes is to take the low-energy limit. Thus, taking $`M_s\mathrm{}`$ for $`N`$ coincident D3-branes leaves the $`U(N)`$ $`𝒩=4`$ SYM degrees of freedom only and taking the $`M_p\mathrm{}`$ limit for $`N`$ coincident M5-branes leaves the $`(2,0)`$ degrees of freedom . One can obtain field theories with less supersymmetry by placing the branes at singularities -. A larger class of theories is possible if one relaxes the condition of Lorentz invariance. One can realize the Lorentz noninvariant theories by placing the branes in backgrounds that break Lorentz invariance. The most studied example is Yang-Mills theories on noncommutative spaces obtained by an appropriate scaling limit of branes in backgrounds with an NSNS flux -.
The purpose of this paper is to study more configurations of branes at backgrounds that break Lorentz invariance and have an interesting low-energy limit.
The construction that we will use is as follows. Consider a smooth 4D Taub-NUT space (in either M-theory or type-II string theories) that at infinity behaves as a circle fibration over the sphere $`S^2`$ with first Chern class $`c_1=1`$. To be concrete, let us take M-theory. The Taub-NUT space is homogeneous in 6+1 directions. We can turn on a constant 3-form $`C`$-flux along the circle at infinity and two of the homogeneous directions, such that the 4-form field-strength $`dC`$ is zero at infinity. Because the Taub-NUT circle shrinks to a point at the origin the 4-form field strength cannot remain zero throughout the interior of the Taub-NUT space. In the classical approximation, a solution with this particular boundary conditions forces a nonzero field-strength and therefore also affects the metric. The metric is changed in such a way that a brane that is transverse to the Taub-NUT space (i.e. parallel to the 6+1 directions) would have a lower tension if it is at the center of the space.
By tuning the external parameters (the $`C`$ flux at infinity and the radius of the Taub-NUT circle at infinity) we can decouple gravity. In fact, we will suggest two possible limits that decouple gravity. In the first limit the flux is small and the low-energy theory appears to be a new kind of a 5+1D theory with $`𝒩=(1,0)`$ supersymmetry that can roughly be described as the $`(2,0)`$ theory with a massive hypermultiplet. There is no contradiction with chirality of the hypermultiplet in 5+1D because we believe the theory (and, in particular, the mass term) breaks Lorentz invariance explicitly. In the second limit the flux is kept finite and the decoupling argument is of the same nature as the dynamical argument presented in .
Similar constructions can be repeated with the M2-brane and D-branes.
The paper is organized as follows. In section (2) we will describe the setting for the constructions and review the geometry of the Taub-NUT space. In section (3) we will study the spectrum of BPS excitations of the pinned branes and the various energy scales involved. We will show that the spectrum includes a particle with a very low mass. In section (4) we will interpret some of the results of section (2) from the low-energy supergravity solutions. In section (5) we will present the limits of the external parameters that decouples gravity. In section (6) we discuss the low-energy description of the theories and resolve a puzzle about fermions. In section (7) we remove the M5-branes and study the Taub-NUT space with 3-form flux, $`C_3`$ on its own. We analyze the spectrum of BPS states and show that the large $`C_3`$ limit can be accompanied with a rescaling of coordinates that makes the energies of the low-lying BPS states finite.
## 2 The setting
### 2.1 Review of the Taub-NUT geometry
The metric of a KK-monopole is the Taub-NUT metric:
$$ds^2=R^2U(dyA_idx^i)^2+U^1(d\stackrel{}{x})^2,i=1\mathrm{}3,0y2\pi .$$
(1)
where,
$$U=\left(1+\frac{R}{|\stackrel{}{x}|}\right)^1,$$
and $`A_i`$ is the gauge field of a monopole centered at the origin. This metric has a few properties that we will utilize.
* It is a circle fibration over $`𝐑^3`$ with the origin excluded.
* The radius of the fiber shrinks to zero as we approach the origin and becomes a constant $`R`$ as we approach infinity.
* If we restrict to $`|\stackrel{}{x}|=r`$ with constant $`r>0`$ the circle fibration is equivalent to the Hopf fibration of $`S^1`$ over $`S^2`$.
* There is a $`U(1)`$ isometry $`yy+ϵ`$. It has one fixed point at the origin.
* The $`U(1)`$ isometry acts nontrivially on the tangent space to the point at the origin.
### 2.2 Turning on a flux at infinity
Now suppose that we have a theory of gravity coupled to a vector field $`A_\mu `$ and we are looking for a solution to the equations of motion with boundary conditions such that at infinity we have a circle fibration over $`S^2`$ with $`c_1=1`$ and there is a constant Wilson line $`A_y𝑑y=w`$ on the circle at infinity. If $`w0`$, we cannot have $`F_{\mu \nu }=0`$ throughout space because this will force the holonomy $`A_y𝑑y`$ to be constant contradicting the fact that the circle shrinks to zero at $`\stackrel{}{x}=0`$. Thus, we expect that the solution with the above boundary conditions will have a nonzero field-strength near the center of the Taub-NUT space. We also expect the Taub-NUT metric to change, as a consequence.
Now let us turn to the setting in our case. We take a Taub-NUT space in M-theory or one of the type-II string theories and we turn on a tensor field at infinity. In M-theory we take the Taub-NUT circle direction to be the $`7^{th}`$ and let $`0\mathrm{}6`$ be directions perpendicular to the Taub-NUT space. We can then turn on $`C_{167}`$ at infinity. In a low energy limit, we will see that there exists a solution with this kind of boundary condition (i.e. being a Hopf fibration at infinity and having the constant $`C_{167}`$ flux). It is also very plausible that for any value of $`C_{167}`$ there exists a background of M-theory with these boundary conditions. Next, we add M5-branes along directions $`0\mathrm{}5`$ and ask whether there is a limit of $`R`$ and $`C_{167}`$ for which we obtain a theory that is decoupled from gravity.
## 3 BPS states
We wish to study the dynamics of $`N`$ M5-branes at the center of a Taub-NUT space with a 3-form field that is constant at infinity and has one direction along the Taub-NUT circle. In order to understand the dynamics and the relevant energy scales of the system we will study various BPS states and fluxes in this theory.
### 3.1 Embedding in M-theory on $`T^7`$
We would like to use the formula for masses of BPS particles in M-theory on $`T^7`$ (see ). We will start with all fluxes turned off and take $`T^7`$ in the form of a product of circles of radii $`R_1\mathrm{}R_7`$. We will denote by $`M_p`$ the 11-dimensional Planck scale. We take the Taub-NUT circle to be $`R_7`$. We let the Taub-NUT wrap directions $`1\mathrm{}6`$. Its mass is:
$$M_{TN}=M_p^9VR_7,VR_1\mathrm{}R_7.$$
The mass of the M5-brane is:
$$M_{M5}=M_p^6R_1R_2R_3R_4R_5=M_p^6VR_6^1R_7^1.$$
There are $`56`$ $`U(1)`$ charges in the low-energy description of M-theory on $`T^7`$. We will denote them by:
$$Q^i,\stackrel{~}{Q}_i,Q_{ij},\stackrel{~}{Q}^{ij}.$$
They correspond to BPS particles with masses:
$$R_i^1,M_p^9VR_i,M_p^3R_iR_j,M_p^6VR_i^1R_j^1.$$
These are KK-particles, KK-monopoles, M2-branes and M5-branes respectively. Now let us turn on some flux $`C_{mnp}`$ ($`1m<n<p7`$) and let us assume, for simplicity, that $`C_{mnp}0`$ for only one set of indices $`m,n,p`$. Let us consider a BPS-state (that preserves some fraction of the SUSY) with integer charges $`Q^i,\stackrel{~}{Q}_i,Q_{ij},\stackrel{~}{Q}^{ij}`$. From this vector one can construct an $`8\times 8`$ complex anti-symmetric central charge matrix $`Z_{ab}=Z_{ba}`$ ($`a,b=1\mathrm{}8`$) that is linear in the $`Q`$’s. The procedure is as follows . Let us define the periodic variable:
$$\varphi ^{mnp}C_{mnp}R_mR_nR_p,\varphi ^{mnp}\varphi ^{mnp}+1.$$
Next, one defines:<sup>1</sup><sup>1</sup>1We wish to thank the anonymous referee for pointing out a typo in the first line. This corrects two formulas in section (7).
$`q^i`$ $``$ $`Q^i+\varphi ^{ijp}Q_{jp},`$
$`\stackrel{~}{q}_i`$ $``$ $`\stackrel{~}{Q}_i,`$
$`\stackrel{~}{q}^{ij}`$ $``$ $`\stackrel{~}{Q}^{ij}+\varphi ^{ijp}\stackrel{~}{Q}_p,`$
$`q_{ij}`$ $``$ $`Q_{ij}ϵ_{ijklmnp}\stackrel{~}{Q}^{kl}\varphi ^{mnp}.`$
Here we have used the assumption that $`\varphi _{mnp}0`$ for only one set of indices $`m,n,p`$ (that we are going to take to be $`1,6,7`$ later on). Otherwise, we will also have terms that are quadratic and cubic in $`\varphi `$. One way to think about these equations is that the presence of the fractional $`\varphi _{mnp}`$ creates effective fractional charges. For example, there is an effective fractional membrane charge if there is an M5-brane with a $`C`$-field turned on.
The central charge matrix is now given by:
$$Z_{ab}=\underset{1m<n7}{}(M_p^3R_mR_nq_{mn}+iM_p^6VR_m^1R_n^1\stackrel{~}{q}^{mn})\mathrm{\Gamma }_{ab}^{mn}+\underset{m=1}{\overset{7}{}}(M_p^9VR_m\stackrel{~}{q}_m+iR_m^1q^m)\mathrm{\Gamma }_{ab}^{m8}.$$
Here we have used the anti-symmetric
$$\mathrm{\Gamma }_{ab}^{pq}=\mathrm{\Gamma }_{ab}^{qp}=\mathrm{\Gamma }_{ba}^{pq},p,q=1\mathrm{}8,a,b=1\mathrm{}8$$
which are the generators of $`SO(8)`$ in the spinor representation $`\underset{¯}{\mathrm{𝟖}}_s`$. The BPS bound from the matrix $`Z`$ is that the mass squared of a state with given charge should be at least the maximal eigenvalue of $`Z^{}Z`$.
We shall now apply this formula to various BPS states in the theory.
### 3.2 The 5-brane tension
How much energy does it cost to separate the 5-brane from the Taub-NUT space?
We set $`C_{167}`$ to a nonzero value and set $`\stackrel{~}{Q}_7=1`$ and $`\stackrel{~}{Q}^{67}=N`$. We also set $`\varphi ^{167}=C_{167}R_1R_6R_7`$. The central charge matrix is given by:
$$Z=M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}.$$
The maximal eigenvalue of this matrix is:
$$M_p^6V\sqrt{(M_p^3R_7+NR_6^1R_7^1)^2+C_{167}^2R_7^2}$$
We compare this to the separate masses of the Taub-NUT alone and the M5-brane alone. The sum of the masses is:
$$M_p^6V\left(\sqrt{M_p^6R_7^2+C_{167}^2R_7^2}+NR_6^1R_7^1\right)$$
Thus, the bound state energy is:
$$M=M_p^6V\left(\sqrt{M_p^6R_7^2+C_{167}^2R_7^2}+NR_6^1R_7^1\sqrt{(M_p^3R_7+NR_6^1R_7^1)^2+C_{167}^2R_7^2}\right)$$
Let us take the limit $`M_p^3R_6R_7^2\mathrm{}`$. We obtain ($`CM_p^3C_{167}`$):
$$MNM_p^6VR_6^1R_7^1\left(1\frac{1}{\sqrt{1+C^2}}\right).$$
Thus, the tension of each 5-brane effectively decreases by $`\sqrt{1+C^2}`$ when it is bound to the Taub-NUT.
### 3.3 Momentum States
The result above can be interpreted simply as a rescaling of the metric $`g_{11}`$ in the $`1^{st}`$ direction. To see this, let us compare the energy of a KK-particle in the $`1^{st}`$ direction to the energy of a KK-particle with momentum in the $`2^{nd}`$ direction.
For momentum in the $`1^{st}`$ direction we find:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}+ikR_1^1\mathrm{\Gamma }^{18}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+iv\mathrm{\Gamma }^{18}.`$
The mass is:
$$\sqrt{(x+y+v)^2+z^2}\sqrt{(x+y)^2+z^2}\frac{x}{\sqrt{x^2+z^2}}v$$
The last result is in the limit $`x,zyv`$. Thus the energy of a massless particle with $`k`$ units of momentum in the $`1^{st}`$ direction is:
$$\frac{k}{\sqrt{1+C^2}}R_1^1.$$
This suggests that we define:
$$\stackrel{~}{R}_1\sqrt{1+C^2}R_1$$
The KK-mass is then $`\stackrel{~}{R}_1^1`$. For particles with momentum in the $`2^{nd}`$ direction we find:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}+ikR_2^1\mathrm{\Gamma }^{28}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+iv\mathrm{\Gamma }^{28}.`$
The mass is then just $`kR_2^1`$.
### 3.4 Massive Particles
As we have seen, the $`N`$ M5-branes are stuck at the center of the Taub-NUT space where their tension is minimal. Intuitively, this suggests that small fluctuations of the world-volume of the M5-branes are described by a massive field. A Taub-NUT space has a $`U(1)`$ isometry. When $`C_{167}=0`$, a fluctuation of the M5-brane in the Taub-NUT directions is charged under that $`U(1)`$ because at the center of the Taub-NUT geometry the $`U(1)`$ is embedded in the local $`SO(4)`$ isometry of the tangent space.
Thus, we should check what is the mass of a BPS state with $`U(1)`$ charge. We therefore set $`Q^7=k`$ and calculate:
$$Z=(M_p^9VR_7+ikR_7^1)\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iM_p^6C_{167}VR_7\mathrm{\Gamma }^{16}.$$
Now the maximal eigenvalue is:
$$M_p^6V\sqrt{(M_p^3R_7+NR_6^1R_7^1)^2+(C_{167}R_7+kM_p^6V^1R_7^1)^2}$$
The energy of the excitation is therefore:
$$M_p^6V\left(\sqrt{(M_p^3R_7+NR_6^1R_7^1)^2+(C_{167}R_7+kM_p^6V^1R_7^1)^2}\sqrt{(M_p^3R_7+NR_6^1R_7^1)^2+C_{167}^2R_7^2}\right).$$
In the limit $`M_p^3R_6R_7^2\mathrm{}`$ this becomes:
$$k\frac{C}{\sqrt{1+C^2}}R_7^1.$$
### 3.5 Tensor fluxes
We can also calculate the energy of fluxes of the anti-self-dual 2-form field that is part of the low-energy tensor multiplet of an M5-brane. Because $`C_{167}`$ explicitly breaks Lorentz invariance, we should discuss fluxes in various directions separately. We set $`Q_{ij}=N_{ij}`$ for $`N_{ij}`$ units of tensor flux in the direction $`i,j`$. If $`T_{ijk}`$ is the anti-self-dual 3-form field-strength on the M5-brane then $`N_{ij}=2\pi T_{ij0}R_iR_j`$ for a single M5-brane.
Let us first set only $`N_{23}0`$. The central charge matrix is:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}+M_p^3N_{23}R_2R_3\mathrm{\Gamma }^{23}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+u\mathrm{\Gamma }^{23}.`$
The BPS bound is:
$$\sqrt{\left(x+\sqrt{u^2+y^2}\right)^2+z^2}$$
In the limit $`M_p^3R_6R_7^2\mathrm{}`$, $`x,zy,u`$ we obtain the energy of the flux:
$`E_{23}`$ $`=`$ $`\sqrt{\left(x+\sqrt{u^2+y^2}\right)^2+z^2}\sqrt{(x+y)^2+z^2}={\displaystyle \frac{\sqrt{u^2+y^2}u}{\sqrt{x^2+z^2}}}`$
$`=`$ $`{\displaystyle \frac{\sqrt{1+M_p^6N^2N_{23}^2R_1^2R_4^2R_5^2}1}{\sqrt{1+C^2}}}M_p^6NR_1R_2R_3R_4R_5`$
$``$ $`{\displaystyle \frac{N_{23}^2R_2R_3}{2\sqrt{1+C^2}NR_1R_4R_5}}={\displaystyle \frac{N_{23}^2R_2R_3}{2N\stackrel{~}{R}_1R_4R_5}}.`$
The last line is in the limit $`M_pR_i1`$ for $`i=1,4,5`$. Next we check a flux with one index in the direction of $`C_{167}`$. Let us take $`N_{12}0`$. We find:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}+M_p^3N_{12}R_1R_2\mathrm{\Gamma }^{12}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+v\mathrm{\Gamma }^{12}.`$
$`E_{12}`$ $`=`$ $`\sqrt{v^2+x^2+y^2+z^2+2\sqrt{v^2x^2+x^2y^2+v^2z^2}}\sqrt{(x+y)^2+z^2}`$
$``$ $`{\displaystyle \frac{\sqrt{v^2(1+C^2)+y^2}y}{1+C^2}}_{yv}{\displaystyle \frac{v^2\sqrt{1+C^2}}{2y}}`$
$`=`$ $`{\displaystyle \frac{N_{12}^2\sqrt{1+C^2}R_1R_2}{2NR_3R_4R_5}}={\displaystyle \frac{N_{12}^2\stackrel{~}{R}_1R_2}{2NR_3R_4R_5}}.`$
We see that these results support the claim that we have to rescale the first coordinate by a factor of $`\sqrt{1+C^2}`$.
### 3.6 Strings
So far we described the excitations of a possibly free theory. Now we would like to take the number of M5-branes to be $`N=2`$ and separate the M5-branes along the $`6^{th}`$ direction. Let the separation be $`\varphi R_6`$. We expect to find strings, made by M2-branes stretched between the M5-branes, with a tension proportional to $`\varphi `$. We also have to establish the coefficient of the kinetic term, $`(\varphi )^2`$, in the low energy effective action. We can do that by taking $`R_2R_6`$ and calculate the energy of and M5-brane that wraps the diagonal of the $`26`$ directions. This is described by $`\varphi (x_2)=lx_2R_2^1R_6`$. The corresponding central charge matrix is:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+ilM_p^6VR_2^1R_7^1\mathrm{\Gamma }^{27}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+iw\mathrm{\Gamma }^{27}.`$
The energy is:
$$E=\sqrt{x^2+y^2+z^2+w^2+2\sqrt{x^2y^2+x^2w^2+w^2z^2}}\sqrt{(x+y)^2+z^2}\frac{\sqrt{x^2+z^2}}{2xy}w^2$$
This is:
$$\sqrt{1+C^2}\frac{M_p^6VR_6}{2NR_7}l^2=\frac{M_p^6\stackrel{~}{R}_1R_2R_3R_4R_5}{2N}\left(\frac{R_6l}{R_2}\right)^2.$$
We compare this to:
$$\frac{1}{2N}\stackrel{~}{R}_1\mathrm{}R_5(\varphi )^2$$
and find that:
$$\phi M_p^3\varphi $$
has the normalized kinetic energy.
To calculate the tension of the string we gradually change $`\varphi `$ from $`0`$ to $`2\pi `$. When it is $`2\pi `$ we calculate the mass of $`k`$ strings stretched on the $`2^{nd}`$ direction:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+kM_p^3R_2R_6\mathrm{\Gamma }^{26}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+iz\mathrm{\Gamma }^{16}+w\mathrm{\Gamma }^{26}.`$
We find $`E=w`$. The energy of the string in the limit $`R_6\mathrm{}`$ is thus:
$$M_p^3R_2\varphi .$$
The tension of the string wrapped in a direction not including the $`1^{st}`$ is therefore $`T\phi `$.
For the mass of strings stretched in the $`1^{st}`$ direction we calculate:
$`Z`$ $`=`$ $`M_p^9VR_7\mathrm{\Gamma }^{78}+iNM_p^6VR_6^1R_7^1\mathrm{\Gamma }^{67}+kM_p^3R_1R_6\mathrm{\Gamma }^{16}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}`$
$``$ $`x\mathrm{\Gamma }^{78}+iy\mathrm{\Gamma }^{67}+(w+iz)\mathrm{\Gamma }^{16}.`$
$$E=\sqrt{(x+y+w)^2+z^2}\sqrt{(x+y)^2+z^2}\frac{x+y}{\sqrt{(x+y)^2+z^2}}w$$
The energy of the string in the limit $`R_6\mathrm{}`$ is thus:
$$\frac{1}{\sqrt{1+C^2}}M_p^3R_1\varphi =\frac{1}{1+C^2}M_p^3\stackrel{~}{R}_1\varphi .$$
We note that in these calculations we assume that the metric on the $`\phi `$-moduli space is constant. In 5+1D, supersymmetry implies that the metric on the tensor-multiplet moduli space is flat. However, in our case we can only use the $`SO(4,1)`$ subgroup of the $`SO(5,1)`$ Lorentz group. So, in principle there can be a nontrivial metric on the moduli space .
### 3.7 Summary
Based on the BPS analysis we found the following facts.
* We have to rescale the metric in the $`1^{st}`$ direction so that $`\stackrel{~}{R}_1=\sqrt{1+C^2}R_1`$ where $`C=M_p^3C_{167}`$. This way the energy of massless particles with momentum $`p`$ is given by the Lorentz invariant expression $`|p|`$.
* The tension of each M5-brane is smaller by a factor of $`\sqrt{1+C^2}`$ when it is at the center of the Taub-NUT (relative to infinity).
* There appears to be a massive particle in the spectrum with mass
$$m_0\frac{C}{\sqrt{1+C^2}}R_7^1.$$
(2)
* The energy of tensor fluxes is as it should be if expressed in terms of $`\stackrel{~}{R}_1`$. Tensor fluxes in direction $`1,2`$ have energy:
$$E_{12}=\frac{N_{12}^2\stackrel{~}{R}_1R_2}{2NR_3R_4R_5},$$
while tensor fluxes in direction $`2,3`$ have energy:
$$E_{23}=\frac{N_{23}^2R_2R_3}{2N\stackrel{~}{R}_1R_4R_5}.$$
* When the M5-branes are separated there appear to be strings in the spectrum. The tension of the strings is proportional to the separation. The tension seems to be smaller by a factor of $`(1+C^2)`$ for strings stretched in the $`1^{st}`$ direction relative to strings stretched in the directions $`2\mathrm{}5`$.
## 4 Gravity solutions
We have seen in section (3) that a Taub-NUT space with nonzero boundary conditions for the 3-form field along the circle at infinity creates a potential that pins M5-branes to the origin. We have also seen that some excitations of the M5-brane become massive. In this section we will explain the mechanism that is responsible for these effects. For that purpose we will describe the classical supergravity solution that corresponds to this Taub-NUT space. The solution is a good approximation when the curvature and field strength are small and that is true for $`M_pR_71`$ and $`M_p^3C_{167}1`$. We will also describe the solution of the Taub-NUT space with $`N`$ M5-branes at the center. This solution is a good approximation either when $`N`$ is large or $`N=0`$.
### 4.1 The solution
We start with a configuration of $`Q_5`$ D5-branes and $`Q_3`$ D3-branes in type-IIB theory oriented along ($`x_0,x_2,x_3,x_4,x_5,x_6`$) and ($`x_0,x_2,x_3,x_7`$) respectively. We have kept $`x_1`$ to be the $`11^{th}`$ direction so as to remain consistent with the notations of the previous sections.
Under an S-duality this will become a system of $`Q_5`$ NS5-branes and $`Q_3`$ D3-branes. We define the following terms:
$$H_3=1+Q_3/r,H_5=1+Q_5/r$$
and $`r=\sqrt{(x^8)^2+(x^9)^2+(x^{10})^2}`$. $`Q_i`$ depends on the number $`N_i`$ of D-branes and also on $`M_p`$. We will use the explicit form of the $`Q_i`$ later. The metric for the NS5-D3 configuration is (see ):
$$ds^2=H_3^{1/2}ds_{023}^2+H_3^{1/2}ds_{456}^2+H_5H_3^{1/2}ds_7^2+H_5H_3^{1/2}ds_{89,10}^2$$
After a series of T-dualities in the $`4^{th}`$ and $`5^{th}`$ directions, we get a configuration of $`Q_5`$ NS5-branes and $`Q_3`$ D5 oriented along ($`x_0,x_2,x_3,x_4,x_5,x_6`$) and ($`x_0,x_2,x_3,x_4,x_5,x_7`$) respectively with metric:
$$ds^2=H_3^{1/2}ds_{02345}^2+H_3^{1/2}ds_6^2+H_5H_3^{1/2}ds_7^2+H_5H_3^{1/2}ds_{89,10}^2.$$
Basically this is our starting configuration. The whole chain of dualities was done only to calculate the metric for this configuration. The directions $`x^6,x^7`$ are on a square torus. To go to an inclined torus we make the following transformations:
$`x^6`$ $`=`$ $`y^6\mathrm{sec}\theta +y^7\mathrm{sin}\theta ,`$
$`x^7`$ $`=`$ $`y^7\mathrm{cos}\theta .`$
$`y^i`$ are the new coordinates. Observe that $`x^j=y^j`$ for $`j6,7`$. Therefore only the 6,7 part of the metric will undergo some change, and it will look like:
$`ds^2`$ $`=`$ $`H_3^{1/2}(dy^6\mathrm{sec}\theta +dy^7\mathrm{sin}\theta )^2+H_5H_3^{1/2}(dy^7\mathrm{cos}\theta )^2`$
$`=`$ $`H_3^{1/2}(dy^6\mathrm{sec}\theta )^2+(H_3^{1/2}\mathrm{sin}^2\theta +H_5H_3^{1/2}\mathrm{cos}^2\theta )(dy^7)^2+2H_3^{1/2}\mathrm{tan}\theta dy^6dy^7`$
The rest of the components of the metric will remain the same. Now under a T-duality along $`y^7`$ we get a ($`Q_5`$-centered) Taub-NUT space and $`Q_3`$ D4-branes. The Taub-NUT space has a nontrivial metric along ($`y_7,y_8,y_9,y_{10}`$) and the D4-branes are oriented along ($`y_2,y_3,y_4,y_5`$).
The metric for this configuration is:
$$ds^2=H_3^{1/2}ds_{02345}^2+hH_5(dy^6)^2+h(dy^7+B_{7i}dy^i)^2+H_5H_3^{1/2}ds_{89,10}^2$$
where $`h^1=H_3^{1/2}\mathrm{sin}^2\theta +H_5H_3^{1/2}\mathrm{cos}^2\theta `$ and $`i=8,9,10`$. Observe that the KK gauge field $`A_i=B_{7i}`$ and $`B_{7i}`$ comes from the NS5 brane. The Taub-NUT circle is along $`y_7`$.
There is also an antisymmetric two form background coming from the inclination of the torus. It is given by:
$$B^{(NS)}=hH_3^{1/2}\mathrm{tan}\theta dy^6(dy^7+B_{7i}dy^i)$$
The coefficient goes to a constant $`T=\mathrm{tan}\theta `$ at infinity. The dilaton behaves as:
$$e^{2\varphi }=hH_5/H_3.$$
The string coupling constant, $`g`$, has been set to one.
We now lift this configuration to M-theory. The M-theory direction is $`x^1`$. The various components of the metric are ($`a,b`$ are 10-dimensional indices):
$$G_{ab}=(hH_5/H_3)^{1/3}g_{ab},G_{a1}=0,G_{11}=(hH_5/H_3)^{2/3}$$
and the three form background is:
$$C_3=M_p^3hH_3^{1/2}\mathrm{tan}\theta dx_1dy^6(dy^7+B_{7i}dy^i)$$
At infinity, the value of the 3-form flux becomes $`M_p^3CM_p^3\mathrm{tan}\theta `$.
We can still subtract from $`C_3`$ a constant because this does not affect the field-strength. We do it so as to fix the boundary condition $`C_3=0`$ at $`r=0`$. This is because the radius of the circle in the $`7^{th}`$ direction shrinks to zero and and if $`C_3(0)0`$ we would get a singularity in the field-strength at the origin. We calculate the constant piece to be:
$$\frac{M_p^3Q_3\mathrm{tan}\theta }{Q_3\mathrm{sin}^2\theta +Q_5\mathrm{cos}^2\theta }.$$
Now let us study the background geometry alone, setting the number $`Q_3`$ of M5-branes, to zero. We obtain the metric:
$`ds^2`$ $`=`$ $`(hH_5/H_3)^{2/3}dx_1^2+(hH_5)^{1/3}H_3^{1/6}(dx_0^2+dx_2^2+dx_3^2+dx_4^2+dx_5^2)+(hH_5)^{2/3}H_3^{1/3}(dy^6)^2`$
$`+h^{2/3}(H_5/H_3)^{1/3}(dx_7+B_{7i}dx_i)^2+h^{1/3}H_5^{2/3}H_3^{5/6}(dx_8^2+dx_9^2+dx_{10}^2)`$
$`=`$ $`(hH_5)^{2/3}dx_1^2+(hH_5)^{1/3}(dx_0^2+dx_2^2+dx_3^2+dx_4^2+dx_5^2)`$
$`+(hH_5)^{2/3}(dy^6)^2+h^{2/3}H_5^{1/3}(dx_7+B_{7i}dx_i)^2+h^{1/3}H_5^{2/3}(dx_8^2+dx_9^2+dx_{10}^2)`$
where:
$$H_5=1+\frac{R}{r},h^1=\mathrm{sin}^2\theta +H_5\mathrm{cos}^2\theta =1+\frac{R\mathrm{cos}^2\theta }{r}.$$
We also find:
$`M_p^3C_3`$ $`=`$ $`h(H_3)^{1/2}\mathrm{tan}\theta dx_1dy^6(dx_7+B_{7i}dx_i)`$ (3)
$`=`$ $`\left(1+{\displaystyle \frac{R\mathrm{cos}^2\theta }{r}}\right)^1\mathrm{tan}\theta dx_1dy^6(dx_7+B_{7i}dx_i)`$
as $`r\mathrm{}`$ we see that
$$M_p^3C_3(\mathrm{})=\mathrm{tan}\theta dx_1dy^6(dx_7+B_{7i}dx_i)Cdx_1dy^6(dx_7+B_{7i}dx_i)$$
### 4.2 Small $`r`$
For $`r0`$ we can still trust our supergravity solution as long as the background value of the three form potential is sufficiently small. In this limit $`h^1=H_5/(1+C^2)`$ in the absence of M5 brane. $`H_5`$, on the other hand, goes as $`Rr^1`$ for small $`r`$. The metric as seen by the M5 brane is nonsingular and behaves as follows:
$`ds^2`$ $`=`$ $`(1+C^2)dx_1^2+dx_{02345}^2+(1+C^2)dx_6^2`$ (4)
$`+`$ $`(1+C^2){\displaystyle \frac{r}{R}}(dy_7+A_idy_i)^2+{\displaystyle \frac{R}{r}}(dr^2+r^2d\mathrm{\Omega }_{8910}^2)`$
We have scaled the coordinates by $`(1+C^2)^{1/6}`$ to get the above metric. Here $`y_7`$ and the angular variables $`\mathrm{\Omega }_{8,9,10}`$ can be taken to parameterize an $`𝐒^3`$. If we change variables to $`r=u^2`$ we see that the $`r,x_7,\mathrm{\Omega }_{8,9,10}`$ parameterize a smooth 4-dimensional point at $`r=0`$, as in the ordinary Taub-NUT space. As $`r0`$, the field $`C_3`$ behaves as:
$$C_3C(1+C^2)M_p^3(r/R)dx_1dy^6(dx_7+B_{7i}dx_i)$$
(5)
It is easy to check that the field strength, $`F_4=dC_3`$, has a finite magnitude as $`r0`$.
### 4.3 The pinning potential
To calculate the potential we have to compute $`\sqrt{detG}`$ along the M5-brane directions. This is given by the following expression:
$$detG=G_{00}G_{11}\mathrm{}G_{55}=H_3^{3/2}H_5^1(H_3^{1/2}\mathrm{sin}^2\theta +H_5H_3^{1/2}\mathrm{cos}^2\theta )$$
Now to calculate the potential as seen by the M5 brane we put $`H_3=1`$. This reduces to
$$\sqrt{detG}=(H_5^1\mathrm{sin}^2\theta +\mathrm{cos}^2\theta )^{1/2}.$$
(6)
As discussed earlier one can trust supergravity solution if $`C1`$. Therefore we have the following limits:
* For $`r\mathrm{}`$ all the harmonic functions become $`1`$ and so $`\sqrt{detG}=1`$.
* Near $`r0`$, $`H_5R/r`$ and we can neglect $`\mathrm{sin}^2\theta `$. This gives
$$\sqrt{detG}=\frac{1}{\sqrt{1+C^2}}.$$
This is precisely the reduction expected from the BPS analysis in the previous sections.
### 4.4 Lorentz invariance
The origin $`r=0`$ is a smooth point for the metric (4). If we rescale:
$$\stackrel{~}{x}_1\sqrt{1+C^2}x_1,$$
then at the vicinity of $`r=0`$ the $`SO(5,1)`$ Lorentz invariance is restored. This is the same rescaling found in section (3). As we will see in section (5), the low-energy theory that describes the M5-branes breaks $`SO(5,1)`$ Lorentz invariance, but at lowest order the breaking term involves only the fermions.
Let us also mention that even though the metric is smooth, the 4-form field strength, $`dC_3`$, is discontinuous at $`r=0`$. It is nevertheless finite and that will play an important role in generating the Lorentz-breaking fermion term (see section (6)).
### 4.5 Small fluctuations
We can expand the potential (6) in small $`r`$. To leading order,
$$\sqrt{detG}=\frac{1}{\sqrt{1+C^2}}+\frac{C^2}{2(1+C^2)^{1/2}}\frac{r}{R}+O(r)^2.$$
If we change coordinates to $`u=r^{1/2}`$ and rescale $`x_1`$ by the factor $`\sqrt{1+C^2}`$, we see that fluctuations of the position of the M5-brane have an effective potential of
$$\frac{C^2}{2(1+C^2)}M_pu^2.$$
We conclude that the fluctuations become massive with the same mass as predicted by the BPS calculation (2).
## 5 Decoupling limits
We now wish to study limits of the previous constructions where gravity can be decoupled. To be concrete, we will concentrate on the example of $`N`$ M5-branes at the center of a Taub-NUT space with $`C_{167}`$-flux turned on.
There are two kinds of decoupling arguments that we can utilize. The first is a low-energy argument where we set to infinity the scale of all the excitations that we wish to discard. Thus, we say that $`𝒩=4`$ SYM theory describes $`N`$ coincident D3-branes when we set the string-scale to infinity, thereby decoupling the massive string states . We will call this type of argument “kinematical”. Sometimes we are forced to keep the scale of excitations, that we wish to decouple, finite. Another type of decoupling argument is possible if we can set the coupling constant between those excitations and the excitations of our theory to zero. This type of argument was introduced in for the decoupling of bulk string states from the little-string theory. The bulk states have masses of order the string scale $`M_s`$ which is the same scale as that of the little-string theory. The decoupling is argued to occur in the limit of zero string coupling constant. We will call this a “dynamical” argument.
### 5.1 Kinematical decoupling
We have seen in section (3) that an M5-brane at the center of a Taub-NUT space with the field $`C_{167}`$ turned on has massive BPS states. Let us make a list of the various energy scales that we found in section (3). In the limit of $`C0`$, we have:
* $`M_p`$ is the Planck scale.
* $`R_7^1`$ is the scale of KK-excitations far away from the center.
* $`M_pC^{1/3}`$ is the energy-scale of the binding energy per unit volume of the 5-brane.
* $`CR_7^1`$ is the energy scale of excitations of the 5-brane.
In a low-energy decoupling limit we must set the first three scales to infinity. The Planck scale, $`M_p`$ must be set to infinity in order to decouple gravity. The scale $`M_pC^{1/3}`$ must be set to infinity so that we will not have to consider local fluctuations where a small portion of the M5-brane escapes to infinity. Finally, the scale $`R_7^1`$ will be sent to infinity so as to keep the scale of Kaluza-Klein particles that are far from the center small. We will keep $`CR_7^1`$ finite. Finally, we should also take $`M_pR_7\mathrm{}`$. The reason is as follows. If we reduce from M-theory to type-IIA along $`R_7`$, the Taub-NUT space becomes a D6-brane and the M5-branes becomes NS5-branes. The string scale is $`M_s^2=M_p^3R_7`$ and this must be set to infinity.<sup>2</sup><sup>2</sup>2We thank S. Sethi for pointing this out. Moreover, we want the scale that is set by the tension of the D6-brane to be much higher than the scale set by the tension of the M5-branes, otherwise we could not decouple the 6+1D $`U(1)`$ gauge field on the D6-brane. The implies that the string coupling constant should be large, and hence $`M_pR_7\mathrm{}`$.
The decoupled theory that we obtain seems to be a new type of 5+1D theory that has not been encountered before.
### 5.2 Dynamical decoupling
In this limit we take $`M_p\mathrm{}`$ but we wish to keep $`C`$ finite. Since we wish to keep $`CR_7^1`$ finite as well, we are forced to keep the scale $`R_7^1`$ itself finite. This means that we must find another argument for the decoupling of Kaluza-Klein excitations that are far from the center. Such an argument has to be dynamical, namely that the coupling constant between these states and the states of the 5+1D theory is proportional to inverse powers of $`M_p`$. This argument is similar in spirit to the argument made in for the decoupling of bulk string states from the little-string theory. The bulk states have masses of order the string scale $`M_s`$ which is the same scale as that of the little-string theory. The decoupling is argued to occur in the limit of zero string coupling constant.
In our case, in the limit of keeping both $`R_7`$ and $`C`$ finite, we saw in section (3) that the metric in the $`1^{st}`$ direction has to be rescaled by a factor of $`\sqrt{1+C^2}`$ in order to preserve Lorentz invariance, at least to leading order. By this we mean that the energy of massless particles with low momentum $`p`$, will be $`E=|p|`$ no matter what the direction of $`p`$ is. We have also seen that with this rescaling the energy of tensor fluxes in directions $`2,3`$ (i.e. not including the direction of the $`C`$-flux) is:
$$E_{23}=\frac{N_{23}^2R_2R_3}{2N\stackrel{~}{R}_1R_4R_5},$$
where $`N_{23}`$ is an integer. Fluxes in directions $`1,2`$ have energy:
$$E_{12}=\frac{N_{12}^2\stackrel{~}{R}_1R_2}{2NR_3R_4R_5}.$$
which also respects the invariance under interchange of directions $`1`$ and $`2`$.
At higher orders in the momentum expansion, we expect Lorentz invariance to be broken. As we shall see in section (6), it is the fermions that first exhibit the breakdown of Lorentz invariance.
## 6 Low-energy description
What is the low-energy description of these 5+1D theories? A regular M5-brane is described, at low-energies, by a tensor multiplet of $`𝒩=(2,0)`$ supersymmetry. It contains an anti-self-dual tensor field and 5 scalars. In our case, we have argued that the M5-brane becomes pinned in 4 out of the 5 transverse directions. Thus, it is reasonable to assume that 4 scalars become massive and the low-energy description is an anti-self-dual tensor field and a single scalar. Together with fermions, that would make up a single tensor multiplet of $`𝒩=(1,0)`$ supersymmetry and the equations of motion would be Lorentz invariant at the lowest order in the derivative expansion.
### 6.1 The fermions: a puzzle
What about the fermions? At first sight, there seems to be a puzzle. We have seen that 4 scalars become massive. In terms of $`𝒩=(1,0)`$ the 4 scalars are part of a hyper-multiplet. However, a hypermultiplet in 5+1D contains chiral fermions and these cannot be made massive. Furthermore, imagine that we turn on $`C`$ gradually from $`0`$ to its present value. The theory at $`C=0`$ has a (massless) hypermultiplet that contributes to the gravitational anomaly. How did this part of the anomaly disappear at $`C0`$?
The resolution of both puzzles is that the theory at $`C0`$ is not Lorentz invariant. Thus, we should only consider the $`SO(4,1)SO(5,1)`$ subgroup of Lorentz-invariance and therefore only 4+1D supersymmetry. In 4+1D, a hypermultiplet can be given a mass.
Let $`\lambda `$ be the fermion that lives on an M5-brane. It is in the representation $`(\underset{¯}{\mathrm{𝟒}},\underset{¯}{\mathrm{𝟒}})`$ of $`Spin(4,1)\times Spin(5)`$ where $`Spin(5)`$ is the rotation of the transverse directions.
$`Spin(4,1)`$ invariance in directions $`0,2\mathrm{}5`$ suggests that the effective coupling would be:
$$\overline{\lambda }\mathrm{\Gamma }^1\mathrm{\Omega }\lambda .$$
(7)
Here $`\mathrm{\Gamma }^1`$ is one of the 5+1D Dirac matrices and $`\mathrm{\Omega }`$ acts on the $`Spin(5)`$ R-symmetry indices. Furthermore, $`Spin(3)`$ invariance in directions $`8,9,10`$ suggests that $`\mathrm{\Omega }`$ should be a constant $`Spin(3)Spin(5)`$ invariant matrix. After dimensional reduction to 4+1D the term must reduce to an ordinary mass term for the fermions of the hypermultiplet because the theory would be Lorentz invariant. This means that $`\mathrm{\Gamma }^1\mathrm{\Omega }`$ must be the identity on the fermions of the hypermultiplet. Starting with a vector multiplet with 16 supersymmetries in 4+1D we can consider an $`𝒩=1`$ subgroup ($`8`$ supersymmetries) of the SUSY algebra. The fermions that go into the hypermultiplet can be separated from those that go into the vector multiplet (in 4+1D) by their transformation under the $`Spin(4)`$ that, in the original setting, corresponds to rotations in directions $`7,8,9,10`$. The fermions of the hypermultiplet are invariant and this determines $`\mathrm{\Omega }=\mathrm{\Gamma }^6`$ – the 11D Dirac matrix in the $`6^{th}`$ direction. Note that we cannot set $`\mathrm{\Omega }`$ to the identity because then the effective action will be just like a coupling to a constant gauge field that can be gauged away.
### 6.2 The supergravity mechanism
What is the explicit mechanism by which our fermions get a mass? If one places an M5-brane in a region where the 4-form field strength, $`F`$, of M-theory is nonzero but still small, the fermions, $`\lambda `$, on the world-volume of an M5-brane (that are part of the tensor multiplet) couple to it schematically as $`\lambda \lambda F`$. One way to see this is by expanding the action with the fermionic zero-modes of the M5-brane solution. Recall that M-theory has a term in the effective action that is of the form $`\overline{\psi }^M\mathrm{\Gamma }^{PQ}\psi ^NF_{MPQN}`$, where $`M,P,Q,N=0\mathrm{}10`$, $`\psi ^M`$ is the gravitino field, $`F_{MPQN}`$ is the 4-form field-strength and $`\mathrm{\Gamma }^{PQ}`$ is an anti-symmetric product of two Dirac matrices. After reduction to the zero-modes on the M5-brane solution we can find the effective coupling between the 5+1D fermions and the external 4-form field-strength. (Note that the M5-brane solution has strong curvature but we can still use it for this discussion since supersymmetry determines that term uniquely.)
In our case, let $`\mu =0\mathrm{}5`$ a direction parallel to the M5-brane. Let $`A=6\mathrm{}10`$ be a direction orthogonal to the M5-brane. We are interested in the coupling between the components $`F_{\mu ABC}`$ of the field-strength and the fermions $`\lambda `$ from the $`(2,0)`$ tensor-multiplet. This coupling has to be of the form $`F_{\mu ABC}\overline{\lambda }\mathrm{\Gamma }^\mu \mathrm{\Gamma }^{ABC}\lambda `$, where $`\mathrm{\Gamma }^{ABC}`$ acts on the $`Sp(2)`$ R-symmetry indices only and $`\mathrm{\Gamma }^\mu `$ acts on the space-time $`Spin(5,1)`$ spinor indices only.
We have seen in section (4.2) that the 4-form field-strength, $`F_4`$, approaches a constant magnitude as $`r0`$ but it is still discontinous because its direction depends on the path along the transverse directions ($`7,8,9,10`$) in which we take the limit $`r0`$. At first sight this would seem to suggest that the $`F\lambda \lambda `$ term has a scalar-field dependent coefficient. We believe, however, that the correct procedure is to expand the gravitino fields in the presence of the M5-brane as $`\psi ^M\lambda \psi _0^M`$, where $`\psi _0^M`$ are the gravitino zero-modes and the $`\lambda `$’s depend only on directions $`0\mathrm{}5`$ (along the M5-brane). Then, we have to plug this back into the M-theory coupling, $`\overline{\psi }^M\mathrm{\Gamma }^{PQ}\psi ^NF_{MPQN}`$, with $`F_{MNPQ}`$ taken from (5). We believe that this will produce the term suggested in (7).
We could be a bit more precise here. Observe that the value of $`F_{r167}`$ near $`r0`$ is a constant given by $`C(1+C^2)R^1`$. Therefore we expect the fermions to pick up a mass proportional to the value of $`F`$ at the origin. However this is not the case. It is of course true that the $`SO(4,1)`$-invariant mass involves $`F`$ but there is also a contribution from the zero modes of the gravitino in the picture. In the presence of the background $`C`$ field, the normalisable zero modes also pick up contributions from the $`C`$ field in such a way that the zero modes are actually suppressed by inverse powers of $`(1+C^2)`$. To see this, consider the D5-NS5 brane configuration. Let us assume that the system supports a normalisable gravitino zero mode $`\psi _0^7(r)`$. We now go to the slanted torus by the transformation given in section (4.1) and then make a T-duality along $`x_7`$. Using the T-duality rules the combined effect now gives a zero mode suppressed by $`(1+C^2)^{1/2}`$. Now integrating out the zero modes using the $`11`$-dimensional term (with $`x_1`$ being scaled by a factor of $`\sqrt{1+C^2}`$) we see that after dimensional reduction to 4+1D the fermions get the same mass as the bosons.
## 7 Various BPS states in 6+1D
We will now drop the M5-branes from the story and concentrate only on the Taub-NUT space with the C-flux. For the purposes of the discussion it is convenient to think of it as a 6+1D theory, although we do not necessarily claim that it is decoupled from gravity. We will return to that question in section (7.1).
We can calculate the tension of various BPS states of this 6+1D “theory”. We can do this using the same techniques as described in section (3). We can even check some of the statements for small $`C`$ using the solution in section (4.1). As in section (3) we will assume that the theory is compactified on $`𝐓^6`$ with radii $`R_1,\mathrm{},R_6`$. We will express the results in terms of the rescaled radii:
$$\stackrel{~}{R}_1\sqrt{1+C^2}R_1,\stackrel{~}{R}_6\sqrt{1+C^2}R_6.$$
We have calculated the energy of the following BPS objects:
* Kaluza-Klein particles: They have energy:
$$\sqrt{\frac{k_1^2}{\stackrel{~}{R}_1^2}+\frac{k_2^2}{R_2^2}+\mathrm{}+\frac{k_5^2}{R_5^2}+\frac{k_6^2}{\stackrel{~}{R}_6^2}}.$$
* M2-branes: M2-branes that are stretched in directions $`IJ`$ have the following masses, according to the dimension of the intersection of the plane of the membrane with the plane of the C-flux.
+ For $`I,J=2,3,4,5`$ we have the mass:
$$\frac{1}{\sqrt{1+C^2}}M_p^3R_IR_J.$$
+ For $`I,J`$ with $`I=1,6`$ and $`J=2,3,4,5`$ we find the mass:
$$\frac{1}{\sqrt{1+C^2}}M_p^3\stackrel{~}{R}_IR_J$$
+ The BPS formula for M2-branes in direction $`1,6`$ gives:
$$\frac{1}{\sqrt{1+C^2}}M_p^3\stackrel{~}{R}_1\stackrel{~}{R}_6,$$
which again agrees with the supergravity calculation as in (4.1), for small $`C`$.
* M5-branes: This again depends on whether the M5-brane hyper-plane contains both the $`1^{st}`$ and $`6^{th}`$ directions or just one of them.
+ For M5-branes in direction $`1\mathrm{}5`$ we find the mass:
$$\frac{1}{1+C^2}M_p^6\stackrel{~}{R}_1R_2R_3R_4R_5.$$
+ For M5-branes in direction $`1,3\mathrm{}6`$ we find the mass:
$$\frac{1}{1+C^2}M_p^6\stackrel{~}{R}_1R_3R_4R_5\stackrel{~}{R}_6.$$
* Electric fluxes: For this purpose we can think of the Taub-NUT as a D6-brane (after reduction on the $`7^{th}`$ direction).
+ For electric flux in the $`1^{st}`$ direction we find the central charge:
$$Z=M_p^9VR_7\mathrm{\Gamma }^{78}iC_{167}R_1R_7\mathrm{\Gamma }^{68}+M_p^3R_1R_7\mathrm{\Gamma }^{17}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}$$
and the energy is:
$$\frac{\sqrt{1+C^2}\stackrel{~}{R}_1}{2M_p^3R_2R_3R_4R_5\stackrel{~}{R}_6}.$$
+ For flux in the $`2^{nd}`$ direction we find:
$$Z=M_p^9VR_7\mathrm{\Gamma }^{78}+M_p^3R_2R_7\mathrm{\Gamma }^{27}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16}$$
The energy is:
$$\frac{\sqrt{1+C^2}R_2}{2M_p^3\stackrel{~}{R}_1R_3R_4R_5\stackrel{~}{R}_6}.$$
* Magnetic fluxes: We take the magnetic flux to be in direction $`I,J`$. We will distinguish three cases:
+ For $`I,J=2,\mathrm{}5`$, for example $`I,J=2,3`$, we find that the central charge is:
$$Z=M_p^9VR_7\mathrm{\Gamma }^{78}+iM_p^3R_1R_4R_5R_6R_7\mathrm{\Gamma }^{23}C_{167}R_1R_4R_5R_6R_7\mathrm{\Gamma }^{45}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16},$$
and the energy is:
$$\frac{M_p^3\stackrel{~}{R}_1R_4R_5\stackrel{~}{R}_6}{2\sqrt{1+C^2}R_2R_3}.$$
+ For $`I=2,\mathrm{}5`$ and $`J=1,6`$, for example, $`I=2`$ and $`J=1`$, we find:
$$Z=M_p^9VR_7\mathrm{\Gamma }^{78}+iM_p^3R_3R_4R_5R_6R_7\mathrm{\Gamma }^{12}+iC_{167}M_p^6VR_7\mathrm{\Gamma }^{16},$$
and the energy is:
$$\frac{M_p^3R_3R_4R_5\stackrel{~}{R}_6}{2\sqrt{1+C^2}\stackrel{~}{R}_1R_2}$$
+ For $`I,J=1,6`$, we find the energy:
$$\frac{C}{\sqrt{1+C^2}}M_p^6R_2R_3R_4R_5R_7.$$
### 7.1 The large $`C`$ limit
Perhaps the most interesting limit to study is that of $`C\mathrm{}`$.<sup>3</sup><sup>3</sup>3We wish to thank S. Sethi for pointing this limit out. Specifically, let us consider the following limit:
$$M_p\mathrm{},C\mathrm{},M_p^3C^1\text{fixed}.$$
(8)
This limit has been studied in . They argued that a D6-brane in the limit of noncommutative geometry - is described by a decoupled 6+1D theory. Keeping $`M_p^3C^1`$ finite makes sures that the effective Yang-Mills coupling constant $`g_{YM}^2=M_p^3C`$ is fixed. The proposal of a decoupled theory is also related to a previous suggestion of that a non-commutative version of the 6+1D theory that should have been the M(atrix)-model of M-theory on $`T^6`$ could perhaps avert the problems explained in and actually be a decoupled theory. We will not address the issue of decoupling in this paper. (Following a correspondence with O. Aharony, we tend to believe that the 6+1D theory is decoupled, as suggested in , but has a continuum of states like a 10+1D theory.)
Nevertheless, we will point out that all the energies of the BPS states studied in the previous section, except the magnetic flux in directions $`1,6`$, have a finite limit if $`M_p^3C^1`$ is kept fixed and $`R_7`$ is also kept finite.
The mass of the M2-branes is proportional to $`1/g_{YM}^2`$, the mass of the M5-branes is proportional to $`1/g_{YM}^4`$ and the energy of electric-fluxes in directions orthogonal to $`1`$ and $`6`$ is proportional to $`g_{YM}^2`$. It is interesting to note that the M2-branes and M5-branes are becoming light when $`g_{YM}^2`$ is large (compared to the radii $`\stackrel{~}{R}_1,R_2,\mathrm{},R_5,\stackrel{~}{R}_6`$).
## 8 Discussion
We have seen that the dynamics near the origin of a Taub-NUT space with a $`C`$-field turned on at $`\mathrm{}`$ can be used to construct various decoupled theories. We have suggested two types of theories. These theories can be obtained by placing M5-branes as probes and taking either a low-energy limit or a less understood dynamical decoupling limit. In all these examples, the dynamics depends only on the type of singularity at $`r=0`$. The virtue of using a Taub-NUT space is that we can easily use BPS arguments as in section (3).
We can repeat the discussions of the previous sections with D-branes of various dimensions instead of M5-branes or with M2-branes. When we discuss D-branes we can turn on various RR-fields or NSNS 2-form fields at infinity. These fields can have various numbers of indices along the direction of the branes. Thus, we will obtain theories that are either Lorentz invariant or have a Lorentz-breaking term that is characterized by a vector or a tensor.
Alternatively, we can compactify the 5+1D theory that we found on the M5-branes on $`𝐓^d`$ and look for a low-energy description of the resulting $`(6d)`$-dimensional theory. Since the Lorentz-breaking direction (the $`1^{st}`$ direction in the notation of section (3)) can be either compactified or not, we obtain low-energy descriptions that are either Lorentz-invariant or have a vector Lorentz-breaking term.
The two questions, one regarding replacing M5-branes with D-branes or M2-branes and the other regarding the low-energy description of compactified theories, are overlapping. For example, if we compactify the theory of ordinary M5-branes, i.e. the $`(2,0)`$-theory, on $`𝐓^2`$ we obtain the theory of D3-branes, i.e. $`U(N)`$ $`𝒩=4`$ SYM at low-energy. If we further compactify the theory of D3-branes on $`𝐒^1`$ we obtain the theory of D2-branes that at the IR limit flows to the $`Spin(8)`$ theory of M2-branes at a certain point in moduli space -.
These theories seem to provide a mechanism for fixing the position of branes in an ambient space. It would be interesting to generalize these constructions to cases, with probably less supersymmetry, in which more than 4 coordinates of the position of a D-brane are fixed. This might have applications for brane-world scenarios as in -.
Let us also mention that our models are related to the models studied in . Both models are U-dual to the elliptic brane configurations of .
Another question that we hope to address in a later paper is the large $`N`$ limit of this theory, along the lines of the AdS/CFT correspondence .
It would also be interesting to study the limit of large $`C`$ as discussed in (7.1).
## Acknowledgments
It is a pleasure to thank M. Berkooz, M.B. Green, A. Kapustin, S. Mukhi, A. Sen and S. Sethi for helpful discussions. We are also grateful to O. Aharony and T. Banks for helpful correspondence. G.R. gratefully acknowledges the hospitality of the Mathematical Institute at the University of Oxford, where part of this work was done.
The research of K.D. is supported by Department of Energy grant No. DE-FG02-90ER4054442. The research of O.J.G was supported by National Science Foundation grant No. PHY98-02484. The research of G.R. is supported by NSF grant number NSF DMS-9627351.
|
warning/0002/quant-ph0002075.html
|
ar5iv
|
text
|
# Operator monotones, the reduction criterion and the relative entropy.
## Abstract
We introduce the theory of operator monotone functions and employ it to derive a new inequality relating the quantum relative entropy and the quantum conditional entropy. We present applications of this new inequality and in particular we prove a new lower bound on the relative entropy of entanglement and other properties of entanglement measures.
Recently the entanglement of finite systems has received considerable attention when it was realized that the theory of majorization provides a simple mathematical framework in which the theory can be formulated . In general the well developed theory of matrix analysis provides many techniques and ideas that may be useful for the study of entanglement. However, the restriction to finite entanglement, while justified from an experimental point of view, places an additional constraint on the system which may cloud some of the truly fundamental aspects of entanglement. Therefore the study of the asymptotic limit, i.e. a situation in which large numbers of entangled pairs can be manipulated simultaneously, is of interest from a fundamental point of view. A substantial body of work has been developed in recent years, beginning with the case of pure entangled states and an extensive study of different ways to quantify the amount of entanglement in mixed bipartite states. Some interesting examples are the entanglement of formation , the entanglement of distillation, and the relative entropy of entanglement . With the notable exception of the entanglement of formation , these entanglement measures are very difficult to compute analytically even in the qubit case. Therefore it is of great interest to obtain upper and lower bounds for them. To further our understanding of entanglement and our ability to manipulate it locally, it is of interest to try to establish connections with other ideas such as distinguishability , and thermodynamical considerations . In these contexts one mathematical function emerges as a central quantity, namely the relative entropy which is defined as
$$S(\sigma ||\rho )=tr\{\sigma \mathrm{log}\sigma \sigma \mathrm{log}\rho \}.$$
(1)
It has a number of remarkable properties and is closely related to the problem of the quantification of entanglement , the distinguishability of quantum states and to thermodynamical ideas (see for example ). Any new inequality relating the relative entropy to other entropic quantities is therefore expected to lead to potentially important new insights into any of these topics and is potentially an important contribution.
In general one would attempt to formulate inequalities that are valid for any density operator. For the study of entanglement, however, a particular special type of inequality would be of great interest. These are inequalities that are only valid when at least some of the density operators that are involved are non-distillable or separable, but may be violated for distillable states. These inequalities naturally lead to sufficient criteria for the distillability of states and they are, as we will demonstrate here, very useful for example in the study of entanglement measures.
In this paper we combine the ideas of positive maps with the concept of operator monotones which has been developed originally in matrix analysis to derive such a new inequality relating the relative entropy and the entropy. We present some lemmas and corollaries to this inequality to demonstrate its usefulness. In particular we derive a new lower bound on the relative entropy of entanglement and a much simplified proof that for pure states the relative entropy of entanglement coincides with the entropy of entanglement.
Let us briefly introduce the idea of operator monotone function as this is a concept which is not very familiar to quantum information theory. Much more material can be found for example in . First we begin with
###### Definition 1
Given two operators $`A`$ and $`B`$, we say that $`AB`$ if the operator $`AB`$ is a non-negative operator, i.e. $`AB`$ if for all $`|\psi `$ we have $`\psi |AB|\psi 0`$.
This definition allows us to compare operators and in particular we are now able to introduce the idea of operator monotones. Given a real valued function $`f:`$ we canonically extend it to a function on Hermitean operators . Then we make the following
###### Definition 2
A function $`f`$ is called operator monotone, if for all pairs of Hermitean operators satisfying $`AB`$ we have $`f(A)f(B)`$.
It should be noted that ordinary monotonicity of a function and operator monotonicity are two entirely different concepts. An example is the function $`f(x)=x^2`$ on the interval $`[0,\mathrm{}]`$, which is not an operator monotone function although it is clearly a monotone function in the ordinary sense ! In physics, and in particular in thermodynamics and the theory of entanglement the entropy and therefore the logarithm plays a central role. It is therefore important to note that
###### Lemma 3
The function $`f(x)=log(x)`$ is operator monotone!
The complicated proof of this theorem can be found in . Lemma 3 is one of the key ingredients in the proof of our new inequality.
The other major input comes from the theory of positive but not completely positive maps, whose application to quantum entanglement of mixed states was pioneered by the Horodeckis and further developed for example in . Positive maps can be used to detect the non-separability of mixed states and a number of important positive maps have been found, amongst them the well known partial transposition . Here we employ a different positive map which has been introduced in . This map, the reduction map $`\mathrm{\Lambda }`$, is defined as
$$\mathrm{\Lambda }(\rho ):=\mathrm{𝟣}\text{ }\text{ }tr\rho \rho .$$
(2)
The reduction map is evidently positive, but not completely positive as the map $`\mathrm{𝟣}\text{ }\text{ }\mathrm{\Lambda }`$ is not positive, i.e. it can transform a positive operator into a non-positive operator. The corresponding reduction criterion is then given by
$$\rho \text{ is non-distillable}\rho _A\mathrm{𝟣}\text{ }\text{ }\rho _{AB}.$$
(3)
The reduction criterion is remarkable as its violation implies distillability of the density operator $`\rho _{AB}`$ while this is not known to be the case for the partial transposition.
Now we use the two key properties of operator monotonicity of the logarithm (Lemma 3) and the reduction criterion Eq. (3) to prove
###### Theorem 4
For any non-distillable state $`\rho _{AB}`$ and for any state $`\sigma _{AB}`$ of a bipartite system AB we have
$`S(\sigma _A)S(\sigma _{AB})`$ $``$ $`S(\sigma _{AB}||\rho _{AB})S(\sigma _A||\rho _A),`$ (4)
$`S(\sigma _B)S(\sigma _{AB})`$ $``$ $`S(\sigma _{AB}||\rho _{AB})S(\sigma _B||\rho _B).`$ (5)
It should be noted that the left hand side of the inequality is the negative conditional entropy which is negative for all separable states $`\sigma _{AB}`$, while it can take positive values for entangled states (an example is the singlet state).
Before we discuss the implications of this theorem further let us present its proof.
Proof: Given a non-distillable state $`\rho _{AB}`$, the reduction criterion and the operator monotonicity of the logarithm imply that
$$\mathrm{log}(\rho _A\mathrm{𝟣}\text{ }\text{ }_B)\mathrm{log}(\rho _{AB}).$$
This statement is equivalent to
$`\sigma _{AB}:tr\{\sigma _{AB}\mathrm{log}(\rho _A)\mathrm{𝟣}\text{ }\text{ }_B\}tr\{\sigma _{AB}\mathrm{log}\rho _{AB}\}`$
$``$ $`\sigma _{AB}:tr\{\sigma _{AB}\mathrm{log}(\rho _A)\mathrm{𝟣}\text{ }\text{ }_B\}tr\{\sigma _{AB}\mathrm{log}\rho _{AB}\}`$
$``$ $`\sigma _{AB}:tr\{\sigma _A\mathrm{log}\rho _A\}tr\{\sigma _{AB}\mathrm{log}\rho _{AB}\}.`$
To draw the connection to the relative entropy we use Definition 1 to find the equivalent statement
$`\sigma _{AB}:S(\sigma _{AB})+S(\sigma _A)S(\sigma _A)tr\{\sigma _A\mathrm{log}\rho _A\}`$
$`S(\sigma _{AB})tr\{\sigma _{AB}\mathrm{log}\rho _{AB}\}`$
$``$ $`\sigma _{AB}:S(\sigma _{AB})+S(\sigma _A)+S(\sigma _A||\rho _A)S(\sigma _{AB}||\rho _{AB})`$
$``$ $`\sigma _{AB}:S(\sigma _A)S(\sigma _{AB})S(\sigma _{AB}||\rho _{AB})S(\sigma _A||\rho _A).`$
Interchanging the roles of $`A`$ and $`B`$ we find the second inequality
In the following we present some applications of the new inequality presented in Theorem 4. Firstly, let us demonstrate that from Theorem 4 we can obtain a new lower bound on the relative entropy of entanglement. We find
###### Lemma 5
The relative entropy of entanglement is bounded from below by the negative conditional entropy, i.e. for all $`\sigma _{AB}`$ we have
$$E_{RE}(\sigma _{AB})\mathrm{max}\{S(\sigma _A)S(\sigma _{AB}),S(\sigma _B)S(\sigma _{AB})\}.$$
Proof: The relative entropy of entanglement is defined as
$$E_{RE}(\sigma _{AB})=\underset{\rho _{AB}𝒟}{\mathrm{min}}S(\sigma _{AB}||\rho _{AB})$$
(6)
where $`𝒟`$ either denotes the set of separable states , the set of states with positive partial transpose or the set of non-distillable states . Lemma 5 applies to both definitions and we only prove the strongest one, choosing $`𝒟`$ to be the set of non-distillable states. Let us denote by $`\rho _{AB}^{}`$ the non-distillable state that realizes the relative entropy of entanglement, i.e.
$$E_{RE}(\sigma _{AB})=S(\sigma _{AB}||\rho _{AB}^{}).$$
(7)
From Theorem 4 and the non-negativity of the relative entropy we conclude that
$`S(\sigma _A)S(\sigma _{AB})`$ $``$ $`S(\sigma _{AB}||\rho _{AB}^{})S(\sigma _A||\rho _A^{})`$ (8)
$``$ $`S(\sigma _{AB}||\rho _{AB}^{})`$ (9)
$`=`$ $`E(\sigma _{AB}).`$ (10)
Interchanging systems $`A`$ and $`B`$ and combining the result yields the Lemma 5
A direct consequence of Theorem 1 is a relationship between the relative entropy of entanglement and the entanglement of formation.
###### Corollary 6
For any bipartite state $`\sigma _{AB}`$ we have
$$E_{RE}(\sigma _{AB})E_F(\sigma _{AB})S(\sigma _{AB}).$$
Proof: This follows immediately from Lemma 5 because $`E_F(\sigma _{AB})S(\sigma _A)`$
A remarkable consequence of this Theorem 1 is a very simple proof, that the relative entropy of entanglement for pure states coincides with the entropy of entanglement, i.e. the entropy of the reduced density operator of one subsystem. This statement was first proven in , however, these proofs are very complicated. Using Lemma 5 this proof is simplified considerably.
###### Corollary 7
For pure states $`|\psi _{AB}`$ we find
$$E_{RE}(|\psi _{AB}\psi _{AB}|)=S(\rho _A).$$
(11)
where $`\rho _A=tr_B\{|\psi _{AB}\psi _{AB}|\}`$.
Proof: Up to local unitary operations we can write $`|\psi _{AB}=_{i=1}^n\alpha _i|i_A|i_B`$ for an orthonormal basis $`\{|i\}_{i=1,n}`$. For the mixed state $`\rho _{AB}=_{i=1}^n|\alpha _i|^2|iiii|`$ we find $`E_{RE}(|\psi _{AB}\psi _{AB}|)S(|\psi _{AB}\psi _{AB}|||\rho _{AB})=S(\rho _A)`$. On the other hand Lemma 1 provides $`E_{RE}(|\psi _{AB}\psi _{AB}|)S(\rho _A)`$ and therefore we conclude that Corollary 7 is correct
It is interesting to note that, to our knowledge, all states for which we know the distillable entanglement under local operations and one-way communication, it actually coincides with the negative conditional entropy and one may conjecture that indeed the distillable entanglement under local operations and one way communication is equal to the negative conditional entropy. A similar conjecture has been made by Rains for maximally correlated states .
Another small conclusion we can draw from Theorem 1 is the following
###### Lemma 8
For states $`\sigma _{AB}`$ such that $`E_{RE}(\sigma )=\mathrm{max}\{S(\sigma _A)S(\sigma _{AB}),S(\sigma _B)S(\sigma _{AB})\}`$ the closest state $`\rho _{AB}^{}𝒟`$ must have the same reduced density operator as $`\sigma _{AB}`$.
Proof: Without restricting generality we can assume $`E_{RE}(\sigma _{AB})=S(\sigma _A)S(\sigma _{AB})`$. This implies $`S(\sigma _{AB}||\rho _{AB}^{})=S(\sigma _A)S(\sigma _{AB})`$. But from Theorem 4 we have $`S(\sigma _{AB}||\rho _{AB}^{})S(\sigma _A||\rho _A^{})S(\sigma _A)S(\sigma _{AB})`$ which implies $`S(\sigma _A||\rho _A^{})=0`$ and therefore $`\sigma _A=\rho _A^{}`$
It is important to note, that the lower bound derived in Lemma 5 is actually additive. This allows to draw some conclusions concerning the additivity of the relative entropy of entanglement. In fact, for density operators that achieve the lower bound presented in Lemma 5 the relative entropy is additive.
###### Lemma 9
If two density operators $`\rho _1`$ and $`\rho _2`$ both satisfy $`E_{RE}(\rho _i)=S(\rho _{i,A})S(\rho _{i,AB})`$ then we have
$$E_{RE}(\rho _1\rho _2)=E_{RE}(\rho _1)+E_{RE}(\rho _2),$$
(12)
i.e. the relative entropy of entanglement is additive for $`\rho _1`$ and $`\rho _2`$.
Proof: It is obvious that for any $`\rho _1`$ and $`\rho _2`$ we have $`E_{RE}(\rho _1\rho _2)E_{RE}(\rho _1)+E_{RE}(\rho _2)`$. But on the other hand $`E_{RE}(\rho _1\rho _2)S(\rho _{1A}\rho _{2A})S(\rho _{1AB}\rho _{2AB})=E_{RE}(\rho _1)+E_{RE}(\rho _2)`$ because of additivity of the conditional entropy. Therefore $`E_{RE}(\rho _1\rho _2)=E_{RE}(\rho _1)+E_{RE}(\rho _2)`$
In summary we have proven a new inequality relating the quantum conditional entropy and quantum relative entropy. To demonstrate the usefulness of this inequality, we have used it to derive a new lower bound on the relative entropy of entanglement as well as some remarkably simple proofs of some other properties of the relative entropy of entanglement. Relations between different entanglement measures could be obtained from our inequality and we believe that this will lead to other useful applications in such diverse fields as entanglement, distinguishability or thermodynamics.
Interesting discussions with V. Vedral are acknowledged. This work was supported by EPSRC, The Leverhulme Trust, the EQUIP programme of the European Union and the State Scholarships Foundation of Greece.
|
warning/0002/math0002216.html
|
ar5iv
|
text
|
# About the globular homology of higher dimensional automata
## 1 Introduction
One of the contributions of is the introduction of two homology theories as a starting point for studying branchings and mergings in higher dimensional automata (HDA) from an homological point of view. However these homology theories had an important drawback : roughly speaking, they were not invariant by subdivisions of the observation. Later in , using a model of concurrency by strict globular $`\omega `$-categories borrowed from , two new homology theories are introduced : the negative and positive corner homology theories $`H^{}`$ and $`H^+`$, also called the branching and the merging homologies. It is proved in that they overcome the drawback of Goubault’s homology theories.
Another idea of is the construction of a diagram of abelian groups like in Figure 1, where $`H_{}^{gl}`$ is a new homology theory called the globular homology.
Geometrically, the non-trivial cycles of the globular homology must correspond to the oriented empty globes of $`𝒞`$, and the non-trivial cycles of the branching (resp. the merging) homology theory must correspond to the branching (resp. merging) areas of execution paths. And the morphisms $`h^{}`$ and $`h^+`$ must associate to any globe its corresponding branching area and merging area of execution paths. Many potential applications in computer science of these morphisms are put forward in .
Globular homology was therefore created in order to fulfill two conditions :
* Globular homology must take place in a diagram of abelian groups like in Figure 1. And the geometric meaning of $`h^{}`$ and $`h^+`$ must be exactly as above described.
* Globular homology must be an invariant of HDA with respect to reasonable deformations of HDA, that is of the corresponding $`\omega `$-category.
What is a reasonable deformation of HDA was not yet very clear in . This question is discussed with much more details in .
The old globular homology (i.e. the construction exposed in ) satisfied the first condition, and the second one was supposed to be satisfied by definition (cf. Definition 8.2 of two homotopic $`\omega `$-categories in ), even if some problems were already mentioned, particularly the non-vanishing of the “old” globular homology of $`I^3`$, and more generally of $`I^n`$ for any $`n1`$ in strictly positive dimension.
This latter problem is disturbing because the $`n`$-cube $`I^n`$ (i.e. the corresponding automaton which consists of $`n`$ $`1`$-transitions carried out at the same time) can be deformed by crushing all the $`p`$-faces with $`p>1`$ into an $`\omega `$-category which has only $`0`$-morphisms and $`1`$-morphisms and because the globular homology is supposed to be an invariant by such deformations. The philosophy exposed in tells us similar things : using S-deformations and T-deformations, the $`n`$-cube and the oriented line must be the same up to homotopy, and therefore must have the same globular homology.
The non-vanishing of the second globular homology group of $`I^3`$ (see Figure 2(c)) is due for instance to the $`2`$-dimensional globular cycle
$`\left(R(00)_0R(0++)\right)_1\left(R(0)_0R(0+0)\right)`$
$`\left(R(00)_0R(0++)\right)\left(R(0)_0R(0+0)\right)`$
It is the reason why it was suggested in to add the relation $`A_1B=A+B`$ at least to the $`2`$-dimensional stage of the old globular complex.
But there is then no reason not to add the same relation in the rest of the definition of the old globular complex. For example, if we take the quotient of the old globular complex by the relation $`A_1B=A+B`$ for any pair $`(A,B)`$ of $`2`$-morphisms, then the $`\omega `$-category defined as the free $`\omega `$-category generated by the globular set generated by two $`3`$-morphisms $`A`$ and $`B`$ such that $`t_1A=s_1B`$ gives rise to a $`3`$-dimensional globular cycle $`A_1BAB`$ because $`s_2(A_1BAB)=s_2A_1s_2Bs_2As_2B=0`$ and $`t_2(A_1BAB)=t_2A_1t_2Bt_2At_2B=0`$. So putting the relation $`A_1BAB=0`$ in the old globular complex for any pair of morphisms $`(A,B)`$ of the same dimension sounds necessary. Similar considerations starting from the calculation of the $`(n1)`$-th globular homology group of $`I^n`$ entail the relations $`A_nBAB`$ for any $`n1`$ and for any pair $`(A,B)`$ of $`p`$-morphisms with $`pn+1`$ in the old globular chain complex.
The formal globular homology of Definition 9.3 is exactly equal to the quotient of the old globular complex by these missing relations. It is conjectured (see conjecture 9.5) that this homology theory will coincide for free $`\omega `$-categories generated by semi-cubical sets with the homology theory of Definition 5.2, this latter being the simplicial homology of the globular simplicial nerve $`𝒩^{gl}`$ shifted by one.
We claim that Definition 5.1 (and its simplicial homology shifted by one) cancels the drawback of the old globular homology at least for the following reasons :
* It is noticed in that both corner homologies come from the simplicial homology of two augmented simplicial nerves $`𝒩^{}`$ and $`𝒩^+`$ ; there exists one and only natural transformation $`h^{}`$ (resp. $`h^+`$) from $`𝒩^{gl}`$ to $`𝒩^{}`$ (resp. $`𝒩^+`$) preserving the interior labeling (Theorem 6.1).
* In homology, $`h^{}`$ and $`h^+`$ induce two natural linear maps from $`H_{}^{gl}`$ to resp. $`H_{}^{}`$ and $`H_{}^+`$ which do exactly what we want.
* The globular homology (formal or not) of $`I^n`$ vanishes in strictly positive dimension for any $`n0`$. The globular homology of $`\mathrm{\Delta }^n`$ (the $`n`$-simplex) and of $`2_n`$ (the free $`\omega `$-category generated by one $`n`$-dimensional morphism) as well.
* Using Theorem 9.7 explaining the exact mathematical link between the old construction and the new one, one sees that one does not lose the possible applications in computer science pointed out in .
* The new globular homology, as well as the new globular cut are invariant by S-deformations, that is intuitively by contraction and dilatation of homotopies between execution paths. We will see however that it is not invariant by T-deformations, that is by subdivision of the time, as the old definition and this problem will be a little bit discussed.
This paper is two-fold. The first part introduces the new material. The second part justifies the new definition of the globular homology.
After Section 2 which recalls some conventions and some elementary facts about strict globular $`\omega `$-categories (non-contracting or not) and about simplicial sets, the setting of simplicial cuts of non-contracting $`\omega `$-categories and that of regular cuts are introduced. The first notion allows to enclose the new globular nerve of this paper and both corner nerves in one unique formalism. The notion of regular cuts gives an axiomatic framework for the generalization of the notion of negative and positive folding operators of . Section 4 is an illustration of the previous new notions on the case of corner nerves. In the same section, some non-trivial facts about negative folding operators are recalled. Section 5 provides the definition of the globular nerve of a non-contracting $`\omega `$-category.
The organization of the rest of the paper follows the preceding explanations. First in Section 6, the morphisms $`h^{}`$ and $`h^+`$ are constructed. Section 7 proves that the globular cut is regular. In particular, we get the globular folding operators. Section 8 proves the vanishing of the globular homology of the $`n`$-cube, the $`n`$-simplex and the free $`\omega `$-category generated by one $`n`$-morphism. At last Section 9 makes explicit the exact relation between the new globular homology and the old one. Section 10 speculates about deformations of $`\omega `$-categories considered as a model of HDA and the construction of the bisimplicial set of is detailed.
## 2 Conventions and notations
### 2.1 Globular $`\omega `$-category and cubical set
For us, an $`\omega `$-category will be a strict globular $`\omega `$-category with morphisms of finite dimension. More precisely (see for more details) :
###### Definition 2.1.
A $`1`$-category is a pair $`(A,(,s,t))`$ satisfying the following axioms :
1. $`A`$ is a set
2. $`s`$ and $`t`$ are set maps from $`A`$ to $`A`$ respectively called the source map and the target map
3. for $`x,yA`$, $`xy`$ is defined as soon as $`tx=sy`$
4. $`x(yz)=(xy)z`$ as soon as both members of the equality exist
5. $`sxx=xtx=x`$, $`s(xy)=sx`$ and $`t(xy)=ty`$ (this implies $`ssx=sx`$ and $`ttx=tx`$).
###### Definition 2.2.
A $`2`$-category is a triple $`(A,(_0,s_0,t_0),(_1,s_1,t_1))`$ such that
1. both pairs $`(A,(_0,s_0,t_0))`$ and $`(A,(_1,s_1,t_1))`$ are $`1`$-categories
2. $`s_0s_1=s_0t_1=s_0`$, $`t_0s_1=t_0t_1=t_0`$, and for $`ij`$, $`s_is_j=t_is_j=s_j`$ and $`s_it_j=t_it_j=t_j`$ (Globular axioms)
3. $`(x_0y)_1(z_0t)=(x_1z)_0(y_1t)`$ (Godement axiom or interchange law)
4. if $`ij`$, then $`s_i(x_jy)=s_ix_js_iy`$ and $`t_i(x_jy)=t_ix_jt_iy`$.
###### Definition 2.3.
A globular $`\omega `$-category $`𝒞`$ is a set $`A`$ together with a family $`(_n,s_n,t_n)_{n0}`$ such that
1. for any $`n0`$, $`(A,(_n,s_n,t_n))`$ is a $`1`$-category
2. for any $`m,n0`$ with $`m<n`$, $`(A,(_m,s_m,t_m),(_n,s_n,t_n))`$ is a $`2`$-category
3. for any $`xA`$, there exists $`n0`$ such that $`s_nx=t_nx=x`$ (the smallest of these $`n`$ is called the dimension of $`x`$).
A $`n`$-dimensional element of $`𝒞`$ is called a $`n`$-morphism. A $`0`$-morphism is also called a state of $`𝒞`$, and a $`1`$-morphism an arrow. If $`x`$ is a morphism of an $`\omega `$-category $`𝒞`$, we call $`s_n(x)=d_n^{}(x)`$ the $`n`$-source of $`x`$ and $`t_n(x)=d_n^+(x)`$ the $`n`$-target of $`x`$. The category of all $`\omega `$-categories (with the obvious morphisms) is denoted by $`\omega Cat`$. The corresponding morphisms are called $`\omega `$-functors. The set of morphisms of $`𝒞`$ of dimension at most $`n`$ is denoted by $`tr^n𝒞`$ ; the set of morphisms of $`𝒞`$ of dimension exactly $`n`$ is denoted by $`𝒞_n`$.
Sometime we will use the terminology initial state (resp. final state) for a state $`\alpha `$ which is not the $`0`$-target (resp. the $`0`$-source) of a $`1`$-morphism.
###### Definition 2.4.
A cubical set consists of
* a family of sets $`(K_n)_{n0}`$
* a family of face maps for $`\alpha \{,+\}`$
* a family of degeneracy maps with $`1in`$
which satisfy the following relations
1. $`_i^\alpha _j^\beta =_{j1}^\beta _i^\alpha `$ for all $`i<jn`$ and $`\alpha ,\beta \{,+\}`$
2. $`ϵ_iϵ_j=ϵ_{j+1}ϵ_i`$ for all $`ijn`$
3. $`_i^\alpha ϵ_j=ϵ_{j1}_i^\alpha `$ for $`i<jn`$ and $`\alpha \{,+\}`$
4. $`_i^\alpha ϵ_j=ϵ_j_{i1}^\alpha `$ for $`i>jn`$ and $`\alpha \{,+\}`$
5. $`_i^\alpha ϵ_i=Id`$
A family $`(K_n)_{n0}`$ only equipped with a family of face maps $`_i^\alpha `$ satisfying the same axiom as above is called a semi-cubical set.
###### Definition 2.5.
The corresponding category of cubical sets, with an obvious definition of its morphisms, is isomorphic to the category of presheaves $`Sets^{\mathrm{}^{op}}`$ over a small category $`\mathrm{}`$. The corresponding category of semi-cubical sets , with an obvious definition of its morphisms, is isomorphic to the category of presheaves $`Sets^{\mathrm{}_{}^{semi}{}_{}{}^{op}}`$ over a small category $`\mathrm{}^{semi}`$.
In a simplicial set, the face maps are always denoted by $`_i`$, the degeneracy maps by $`ϵ_i`$. Here are the other conventions about simplicial sets (see for example for further information) :
1. $`Sets`$ : category of sets
2. $`Sets^{\mathrm{\Delta }^{op}}`$ : category of simplicial sets
3. $`Comp\left(Ab\right)`$ : category of chain complexes of abelian groups
4. $`C\left(A\right)`$ : unnormalized chain complex of the simplicial set $`A`$
5. $`H_{}\left(A\right)`$ : simplicial homology of a simplicial set $`A`$
6. $`Ab`$ : category of abelian groups
7. $`Id`$ : identity map
8. $`S`$ : free abelian group generated by the set $`S`$
HDA means higher dimensional automaton. In this paper, this is another term for semi-cubical set, or the corresponding free $`\omega `$-category generated by it.
Various homology theories (see the diagram of Theorem 9.7) will appear in this paper. It is helpful for the reader to keep in mind that the total homology of a semi-cubical set is used nowhere in this work.
### 2.2 Non-contracting $`\omega `$-category
Let $`𝒞`$ be an $`\omega `$-category. We want to define an $`\omega `$-category $`𝒞`$ ($``$ for path) obtained from $`𝒞`$ by removing the $`0`$-morphisms, by considering the $`1`$-morphisms of $`𝒞`$ as the $`0`$-morphisms of $`𝒞`$, the $`2`$-morphisms of $`𝒞`$ as the $`1`$-morphisms of $`𝒞`$ etc. with an obvious definition of the source and target maps and of the composition laws (this new $`\omega `$-category is denoted by $`𝒞[1]`$ in ). The map $`:𝒞𝒞`$ does not induce a functor from $`\omega Cat`$ to itself because $`\omega `$-functors can contract $`1`$-morphisms and because with our conventions, a $`1`$-source or a $`1`$-target can be $`0`$-dimensional. Hence the following definition
###### Proposition and definition 2.6.
For a globular $`\omega `$-category $`𝒞`$, the following assertions are equivalent :
1. $`𝒞`$ is an $`\omega `$-category ; in other terms, $`_i`$, $`s_i`$ and $`t_i`$ for any $`i1`$ are internal to $`𝒞`$ and we can set $`_i^𝒞=_{i+1}^𝒞`$, $`_i^𝒞=_{i+1}^𝒞`$ and $`_i^𝒞=_{i+1}^𝒞`$ for any $`i0`$.
2. The maps $`s_1`$ and $`t_1`$ are non-contracting, that is if $`x`$ is of strictly positive dimension, then $`s_1x`$ and $`t_1x`$ are $`1`$-dimensional (a priori, one can only say that $`s_1x`$ and $`t_1x`$ are of dimension lower or equal than $`1`$)
If Condition (ii) is satisfied, then one says that $`s_1`$ and $`t_1`$ are non-contracting and that $`𝒞`$ is non-contracting.
###### Proof.
Suppose $`s_1`$ and $`t_1`$ non-contracting. Let $`x`$ and $`y`$ be two morphisms of strictly positive dimension and $`p1`$. Then $`s_1s_px=s_1x`$ therefore $`s_px`$ cannot be $`0`$-dimensional. If $`x_py`$ then $`s_1(x_py)=s_1x`$ if $`p=1`$ and if $`p>1`$ for two different reasons. Therefore $`x_py`$ cannot be $`0`$-dimensional as soon as $`p1`$. ∎
###### Definition 2.7.
Let $`f`$ be an $`\omega `$-functor from $`𝒞`$ to $`𝒟`$. The morphism $`f`$ is non-contracting if for any $`1`$-dimensional $`x𝒞`$, the morphism $`f(x)`$ is a $`1`$-dimensional morphism of $`𝒟`$ (a priori, $`f(x)`$ could be either $`0`$-dimensional or $`1`$-dimensional).
###### Definition 2.8.
The category of non-contracting $`\omega `$-categories with the non-contracting $`\omega `$-functors is denoted by $`\omega Cat_1`$.
Notice that in , the word “non-$`1`$-contracting” is used instead of simply “non-contracting”. Since , the philosophy behind the idea of deforming the $`\omega `$-categories viewed as models of HDA is better understood. In particular, the idea of not contracting the morphisms is relevant only for $`1`$-dimensional morphisms. So the “1” in “non-$`1`$-contracting” is not anymore necessary.
###### Definition 2.9.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Then the $`\omega `$-category $`𝒞`$ above defined is called the path $`\omega `$-category of $`𝒞`$. The map $`𝒞𝒞`$ induces a functor from $`\omega Cat_1`$ to $`\omega Cat`$.
Here is a fundamental example of non-contracting $`\omega `$-category. Consider a semi-cubical set $`K`$ and consider the free $`\omega `$-category $`\mathrm{\Pi }(K):=^{\underset{¯}{n}\mathrm{}}K_n.I^n`$ generated by it where
* $`I^n`$ is the free $`\omega `$-category generated by the faces of the $`n`$-cube, whose construction is recalled in Section 4.
* the integral sign denotes the coend construction and $`K_n.I^n`$ means the sum of “cardinal of $`K_n`$” copies of $`I^n`$ (cf. for instance).
Then one has
###### Proposition 2.10.
For any semi-cubical set $`K`$, $`\mathrm{\Pi }(K)`$ is a non-contracting $`\omega `$-category. The functor $`\mathrm{\Pi }:Sets^{\mathrm{}_{}^{semi}{}_{}{}^{op}}\omega Cat`$ from the category of semi-cubical sets to that of $`\omega `$-categories yields a functor from $`Sets^{\mathrm{}_{}^{semi}{}_{}{}^{op}}`$ to the category of non-contracting $`\omega `$-categories $`\omega Cat_1`$.
###### Proof.
The characterization of Proposition 2.6 gives the solution. ∎
## 3 Cut of globular higher dimensional categories
Before introducing the globular nerve of an $`\omega `$-category, let us introduce the formalism of regular simplicial cuts of $`\omega `$-categories. The notion of simplicial cuts enables us to put together in the same framework both corner nerves constructed in and the new globular nerve of Section 5. The notion of regular cuts enables to generalize the notion of negative (resp. positive) folding operators associated to the branching (resp. merging) nerve (cf. ). It is also an attempt to finding a way of characterizing these three nerves. There are no much more things known about this problem.
###### Definition 3.1.
An augmented simplicial set is a simplicial set
$$((X_n)_{n0},(_i:X_{n+1}X_n)_{0in+1},(ϵ_i:X_nX_{n+1})_{0in})$$
together with an additional set $`X_1`$ and an additional map $`_1`$ from $`X_0`$ to $`X_1`$ such that $`_1_0=_1_1`$. A morphism of augmented simplicial set is a map of $``$-graded sets which commutes with all face and degeneracy maps. We denote by $`Sets_+^{\mathrm{\Delta }^{op}}`$ the category of augmented simplicial sets.
The “chain complex” functor of an augmented simplicial set $`X`$ is defined by $`C_n(X)=X_n`$ for $`n1`$ endowed with the simplicial differential map (denoted by $``$) in positive dimension and the map $`_1`$ from $`C_0(X)`$ to $`C_1(X)`$. The “simplicial homology” functor $`H_{}`$ from the category of augmented simplicial sets $`Sets_+^{\mathrm{\Delta }^{op}}`$ to the category of abelian groups $`Ab`$ is defined as the usual one for $`1`$ and by setting $`H_0(X)=Ker(_1)/Im(_0_1)`$ and $`H_1(X)=X_1/Im(_1)`$ whenever $`X`$ is an augmented simplicial set.
###### Definition 3.2.
A (simplicial) cut is a functor $`:\omega Cat_1Sets_+^{\mathrm{\Delta }^{op}}`$ together with a family $`ev=(ev_n)_{n0}`$ of natural transformations $`ev_n:F_ntr^n`$ where $`F_n`$ is the set of $`n`$-simplexes of $``$. A morphism of cuts from $`(,ev)`$ to $`(𝒢,ev)`$ is a natural transformation of functors $`\varphi `$ from $``$ to $`𝒢`$ which makes the following diagram commutative for any $`n0`$ :
The terminology of “cuts” is borrowed from . It will be explained later : cf. the explanations around Figure 3 and also Section 10.
There is no ambiguity to denote all $`ev_n`$ by the same notation $`ev`$ in the sequel. The map $`ev`$ of $``$-graded sets is called the evaluation map and a cut $`(,ev)`$ will be always denoted by $``$.
If $``$ is a functor from $`\omega Cat_1`$ to $`Sets_+^{\mathrm{\Delta }^{op}}`$, let $`C_{n+1}^{}(𝒞):=C_n((𝒞))`$ and let $`H_{n+1}^{}`$ be the corresponding homology theory for $`n1`$.
Let $`M_n^{}:\omega Cat_1Ab`$ be the functor defined as follows : the group $`M_n^{}(𝒞)`$ is the subgroup generated by the elements $`x_{n1}(𝒞)`$ such that $`ev(x)tr^{n2}𝒞`$ for $`n2`$ and with the convention $`M_0^{}(𝒞)=M_1^{}(𝒞)=0`$ and the definition of $`M_n^{}`$ is obvious on non-contracting $`\omega `$-functors. The elements of $`M_{}^{}(𝒞)`$ are called thin.
Let $`CR_n^{}:\omega Cat_1Comp(Ab)`$ be the functor defined by $`CR_n^{}:=C_n^{}/(M_n^{}+M_{n+1}^{})`$ and endowed with the differential map $``$. This chain complex is called the reduced complex associated to the cut $``$ and the corresponding homology is denoted by $`HR_{}^{}`$ and is called the reduced homology associated to $``$. A morphism of cuts from $``$ to $`𝒢`$ yields natural morphisms from $`H_{}^{}`$ to $`H_{}^𝒢`$ and from $`HR_{}^{}`$ to $`HR_{}^𝒢`$. There is also a canonical natural transformation $`R^{}`$ from $`H_{}^{}`$ to $`HR_{}^{}`$, functorial with respect to $``$, that is making the following diagram commutative :
###### Definition 3.3.
A cut $``$ is regular if and only if it satisfies the following properties :
1. For any $`\omega `$-category $`𝒞`$, the set $`_1(𝒞)`$ only depends on $`tr^0𝒞=𝒞_0`$ : i.e. for any $`\omega `$-categories $`𝒞`$ and $`𝒟`$, $`𝒞_0=𝒟_0`$ implies $`_1(𝒞)=_1(𝒟)`$.
2. $`_0:=tr^0`$.
3. $`evϵ_i=ev`$.
4. for any natural transformation of functors $`\mu `$ from $`_{n1}`$ to $`_n`$ with $`n1`$, and for any natural map $`\mathrm{}`$ from $`tr^{n1}`$ to $`_{n1}`$ such that $`ev\mathrm{}=Id_{tr^{n1}}`$, there exists one and only one natural transformation $`\mu .\mathrm{}`$ from $`tr^n`$ to $`_n`$ such that the following diagram commutes
where $`i_n`$ is the canonical inclusion functor from $`tr^{n1}`$ to $`tr^n`$.
5. let $`\mathrm{}_1^{}:=Id__0`$ and $`\mathrm{}_n^{}:=ϵ_{n2}.\mathrm{}ϵ_0.\mathrm{}_1^{}`$ a natural transformation from $`tr^{n1}`$ to $`_{n1}`$ for $`n2`$ ; then the natural transformations $`_i\mathrm{}_n^{}`$ for $`0in1`$ from $`tr^{n1}`$ to $`_{n2}`$ satisfy the following properties
1. $`\{ev_{n2}\mathrm{}_n^{},ev_{n1}\mathrm{}_n^{}\}=\{s_{n1},t_{n1}\}`$.
2. if for some $`\omega `$-category $`𝒞`$ and some $`u𝒞_n`$, $`ev_i\mathrm{}_n^{}(u)=d_p^\alpha u`$ for some $`pn`$ and for some $`\alpha \{,+\}`$, then $`_i\mathrm{}_n^{}=_i\mathrm{}_n^{}d_p^\alpha `$.
6. Let $`\mathrm{\Phi }_n^{}:=\mathrm{}_n^{}ev`$ be a natural transformation from $`_{n1}`$ to itself ; then $`\mathrm{\Phi }_n^{}`$ induces the identity natural transformation on $`CR_n^{}`$.
7. if $`x`$, $`y`$ and $`z`$ are three elements of $`_n(𝒞)`$, and if $`ev(x)_pev(y)=ev(z)`$ for some $`1pn`$, then $`x+y=z`$ in $`CR_{n+1}^{}(𝒞)`$ and in a functorial way.
If $``$ is a regular cut, then the natural transformation $`\mathrm{\Phi }_n^{}`$ is called the $`n`$-dimensional folding operator of the cut $``$. By convention, one sets $`\mathrm{}_0^{}=Id__1`$ and $`\mathrm{\Phi }_0^{}=Id__1`$. There is no ambiguity to set $`\mathrm{\Phi }^{}(x):=\mathrm{\Phi }_{n+1}^{}(x)`$ for $`x_n(𝒞)`$ for some $`\omega `$-category $`𝒞`$. So $`\mathrm{\Phi }^{}`$ defines a natural transformation, and even a morphism of cuts, from $``$ to itself. However beware of the fact that there is really an ambiguity in the notation $`\mathrm{}^{}`$ : so this latter will not be used.
Condition 3 tells us that the $`ϵ_i`$ operations are really degeneracy maps. Condition 4 ensures the existence and the uniqueness of the folding operator associated to the cut.
Condition 5 tells us several things. A priori, a natural transformation like $`ev_i\mathrm{}_n^{}`$ from $`tr^{n1}`$ to $`tr^{n2}`$ is necessarily of the form $`d_p^\alpha `$ for some $`pn1`$ and for some $`\alpha \{,+\}`$. Indeed consider the free $`\omega `$-category $`2_n(A)`$ generated by some $`n`$-morphism $`A`$. Then $`ev_i\mathrm{}_n^{}(A)2_n(A)`$ and therefore $`ev_i\mathrm{}_n^{}(A)=d_p^\alpha (A)`$ for some $`p`$ and some $`\alpha `$. By naturality, this implies that $`ev_i\mathrm{}_n^{}=d_p^\alpha `$. If $`0i<n2`$, then
$`ev_i\mathrm{}_n^{}`$ $`=ev_i\mathrm{}_n^{}d_{n1}^\beta `$ for some $`\beta \{,+\}`$
$`=ev_i\mathrm{}_n^{}i_{n1}d_{n1}^\beta `$
$`=ev_iϵ_{n2}\mathrm{}_{n1}^{}d_{n1}^\beta `$ by construction of $`\mathrm{}_n^{}`$
$`=evϵ_{n3}_i\mathrm{}_{n1}^{}d_{n1}^\beta `$
$`=ev_i\mathrm{}_{n1}^{}d_{n1}^\beta `$ by rule 3
$`=d_p^\alpha d_{n1}^\beta `$ for some $`pn2`$
$`=d_p^\alpha `$
Therefore $`_i\mathrm{}_n^{}`$ is thin. Now if $`n2in1`$, then
$`ev_i\mathrm{}_n^{}`$ $`=ev_i\mathrm{}_n^{}d_{n1}^\beta `$ for some $`\beta \{,+\}`$
$`=ev_i\mathrm{}_n^{}i_{n1}d_{n1}^\beta `$
$`=ev_iϵ_{n2}\mathrm{}_{n1}^{}d_{n1}^\beta `$ by construction of $`\mathrm{}_n^{}`$
$`=ev\mathrm{}_{n1}^{}d_{n1}^\beta `$
$`=d_{n1}^\beta `$ by construction of $`\mathrm{}_n^{}`$
Therefore $`\{ev_{n2}\mathrm{}_n^{},ev_{n1}\mathrm{}_n^{}\}\{s_{n1},t_{n1}\}`$ always holds. Condition 5 states more precisely that these latter sets are actually equal. In other terms, the operator $`\mathrm{}_n^{}`$ concentrates the “weight” on the faces $`_{n2}\mathrm{}_n^{}`$ and $`_{n1}\mathrm{}_n^{}`$.
Condition 6 explains the link between the thin elements of the cut and the folding operators. Intuitively, the folding operators move the labeling of the elements of the cuts in a canonical position without changing the total sum on the source and target sides. What is exactly this canonical position is precisely described by Proposition 3.5. Conditions 5 and 7 ensure that by moving the labeling of an element, we stay in the same equivalence class modulo thin elements.
Now here are some trivial remarks about regular cuts :
* Let $`f`$ be a natural set map from $`tr^0𝒞=𝒞_1`$ to itself. Let $`2_1`$ be the $`\omega `$-category generated by one $`1`$-morphism $`A`$. Then necessarily $`f(A)=A`$ and therefore $`f=Id`$. So the above axioms imply that $`ev_0=Id`$.
* The map $`\mathrm{\Phi }_n^{}`$ induces the identity natural transformation on $`HR_n^{}`$.
* For any $`n1`$, there exists non-thin elements $`x`$ in $`_{n1}(𝒞)`$ as soon as $`𝒞_n\mathrm{}`$. Indeed, if $`u𝒞_n`$, $`ev\mathrm{}_n^{}(u)=u`$, therefore $`\mathrm{}_n^{}(u)`$ is a non-thin element of $`_{n1}(𝒞)`$.
We end this section by some general facts about regular cuts.
###### Proposition 3.4.
Let $`f`$ be a morphism of cuts from $``$ to $`𝒢`$. Suppose that $``$ and $`𝒢`$ are regular. Then $`\mathrm{\Phi }^𝒢f=f\mathrm{\Phi }^{}`$ as natural transformation from $``$ to $`𝒢`$. In other terms, the following diagram is commutative :
###### Proof.
Let $`n0`$ and let $`P(n)`$ be the property : “for any $`\omega `$-category $`𝒞`$ and any $`xtr^n𝒞`$, then $`f\mathrm{}_{n+1}^{}(x)=\mathrm{}_{n+1}^𝒢x`$.”
One has $`\mathrm{\Phi }_1^{}:=Id__0`$, $`\mathrm{\Phi }_1^𝒢:=Id_{G_0}`$ and necessarily $`f_0=Id`$ by definition of a morphism of cuts. Therefore $`P(0)`$ holds. Now suppose $`P(n)`$ proved for some $`n0`$. One has $`evf\mathrm{}_{n+2}^{}=ev\mathrm{}_{n+2}^{}=Id_{tr^{n+1}}`$ since $`f`$ is a morphism of cuts and
$`f\mathrm{}_{n+2}^{}i_{n+1}`$ $`=f(ϵ_n.\mathrm{}_{n+1}^{})i_{n+1}`$
$`=fϵ_n\mathrm{}_{n+1}^{}`$ by definition of $`ϵ_n.\mathrm{}_{n+1}^{}`$
$`=ϵ_nf\mathrm{}_{n+1}^{}`$ since $`f`$ morphism of simplicial sets
$`=ϵ_n\mathrm{}_{n+1}^𝒢`$ by induction hypothesis
Therefore the natural transformation $`f\mathrm{}_{n+2}^{}`$ from $`tr^{n+1}`$ to $`𝒢_{n+1}`$ can be identified with $`ϵ_n.\mathrm{}_{n+1}^𝒢`$ which is precisely $`\mathrm{}_{n+2}^𝒢`$. Therefore $`P(n+1)`$ is proved.
At last, if $`x_n(𝒞)`$, then
$`\mathrm{\Phi }^𝒢f(x)`$ $`=\mathrm{}_{n+1}^𝒢evf(x)`$ by definition of folding operators
$`=\mathrm{}_{n+1}^𝒢ev(x)`$ since $`f`$ preserves the evaluation map
$`=f\mathrm{}_{n+1}^{}ev(x)`$ since $`P(n)`$ holds
$`=f\mathrm{\Phi }^{}(x)`$ by definition of folding operators
###### Proposition 3.5.
If $`u`$ is a $`(n+1)`$-morphism of $`𝒞`$ with $`n1`$, then $`\mathrm{}_{n+1}^{}u`$ is an homotopy within the simplicial set $`(𝒞)`$ between $`\mathrm{}_n^{}s_nu`$ and $`\mathrm{}_n^{}t_nu`$.
###### Proof.
The natural map $`ev_i\mathrm{}_{n+1}^{}`$ for $`0in`$ from $`tr^n`$ to $`tr^{n1}`$ is of the form $`d_{m_i}^{\alpha _i}`$ for $`m_in`$ with $`m_in1`$ for $`0in2`$ and $`\{ev_{n1}\mathrm{}_{n+1}^{},ev_n\mathrm{}_{n+1}^{}\}=\{s_n,t_n\}`$. Therefore for $`0in2`$, $`_i\mathrm{}_{n+1}^{}=_i\mathrm{}_{n+1}^{}s_n=_i\mathrm{}_{n+1}^{}t_n`$ by rule 5b of Definition 3.3. And by construction of $`\mathrm{}_{n+1}^{}`$, one obtains $`_i\mathrm{}_{n+1}^{}=ϵ_{n2}_i\mathrm{}_n^{}s_n=ϵ_{n2}_i\mathrm{}_n^{}t_n`$. ∎
###### Corollary 3.6.
If $`xCR_{n+1}^{}(𝒞)`$, then $`x=\mathrm{}_{n+1}^{}x=\mathrm{}_n^{}s_nx\mathrm{}_n^{}t_nx`$ in $`CR_n^{}(𝒞)`$. In other terms, the differential map from $`CR_{n+1}^{}(𝒞)`$ to $`CR_n^{}(𝒞)`$ with $`n1`$ is induced by the map $`s_nt_n`$.
## 4 The cuts of branching and merging nerves
We see now that the corner nerves $`𝒩^\eta `$ defined in are two examples of regular cuts with the correspondence $`\mathrm{}_n^\eta :=\mathrm{}_n^{𝒩^\eta }`$, $`\mathrm{\Phi }_n^\eta :=\mathrm{\Phi }_n^{𝒩^\eta }`$, $`H_n^\eta :=H_n^{𝒩^\eta }`$, $`HR_n^\eta :=HR_n^{𝒩^\eta }`$ and $`ev(x)=x(0_{dim(x)})`$.
Let us first recall the construction of the free $`\omega `$-category $`I^n`$ generated by the faces of the $`n`$-cube. The faces of the $`n`$-cube are labeled by the words of length $`n`$ in the alphabet $`\{,0,+\}`$, one word corresponding to the barycenter of one face. We take the convention that $`00\mathrm{}0\text{ (}n\text{ times)}=:0_n`$ corresponds to its interior and that $`_n`$ (resp. $`+_n`$) corresponds to its initial state $`\mathrm{}\text{ (}n\text{ times)}`$ (resp. to its final state $`++\mathrm{}+\text{ (}n\text{ times)}`$). If $`x`$ is a face of the $`n`$-cube, let $`R(x)`$ be the set of faces of $`x`$. If $`X`$ is a set of faces, then let $`R(X)=_{xX}R(x)`$. Notice that $`R(XY)=R(X)R(Y)`$ and that $`R(\{x\})=R(x)`$. Then $`I^n`$ is the free $`\omega `$-category generated by the $`R(x)`$ with the rules
1. For $`x`$ $`p`$-dimensional with $`p1`$,
$$s_{p1}(R(x))=R(s_x)$$
and
$$t_{p1}(R(x))=R(t_x)$$
where $`s_x`$ and $`t_x`$ are the sets of faces defined below.
2. If $`X`$ and $`Y`$ are two elements of $`I^n`$ such that $`t_p(X)=s_p(Y)`$ for some $`p`$, then $`XY`$ belongs to $`I^n`$ and $`XY=X_pY`$.
The set $`s_x`$ is the set of subfaces of the faces obtained by replacing the $`i`$-th zero of $`x`$ by $`()^i`$, and the set $`t_x`$ is the set of subfaces of the faces obtained by replacing the $`i`$-th zero of $`x`$ by $`()^{i+1}`$. For example, $`s_{0+00}=\{\text{-+00},\text{0++0},\text{0+0-}\}`$ and $`t_{0+00}=\{\text{++00},\text{0+-0},\text{0+0+}\}`$. Figure 2(c) represents the free $`\omega `$-category generated by the $`3`$-cube.
The branching and merging nerves are dual from each other. We set
$`\omega Cat(I^{n+1},𝒞)^\eta :=\{x\omega Cat(I^{n+1},𝒞),d_0^\eta (u)=\eta _{n+1}`$
$`\text{and }dim(u)=1dim(x(u))=1\}`$
where $`\eta \{,+\}`$ and where $`\eta _{n+1}`$ is the initial state (resp. final state) of $`I^{n+1}`$ if $`\eta =`$ (resp. $`\eta =+`$). For all $`(i,n)`$ such that $`0in`$, the face maps $`_i`$ from $`\omega Cat(I^{n+1},𝒞)^\eta `$ to $`\omega Cat(I^n,𝒞)^\eta `$ are the arrows $`_{i+1}^\eta `$ defined by
$$_{i+1}^\eta (x)(k_1\mathrm{}k_{n+1})=x(k_1\mathrm{}[\eta ]_{i+1}\mathrm{}k_{n+1})$$
and the degeneracy maps $`ϵ_i`$ from $`\omega Cat(I^n,𝒞)^\eta `$ to $`\omega Cat(I^{n+1},𝒞)^\eta `$ are the arrows $`\mathrm{\Gamma }_{i+1}^\eta `$ defined by setting
$`\mathrm{\Gamma }_i^{}(x)(k_1\mathrm{}k_n):=x(k_1\mathrm{}\mathrm{max}(k_i,k_{i+1})\mathrm{}k_n)`$
$`\mathrm{\Gamma }_i^+(x)(k_1\mathrm{}k_n):=x(k_1\mathrm{}\mathrm{min}(k_i,k_{i+1})\mathrm{}k_n)`$
with the order $`<0<+`$.
###### Proposition and definition 4.1.
Let $`𝒞`$ be an $`\omega `$-category. The $``$-graded set $`𝒩^\eta (𝒞)`$ together with the convention $`𝒩_1^\eta (𝒞)=𝒞_0`$, endowed with the maps $`_i`$ and $`ϵ_i`$ above defined with moreover $`_1=s_0`$ (resp. $`_1=t_0`$) if $`\eta =`$ (resp. $`\eta =+`$) and with $`ev(x)=x(0_n)`$ for $`x\omega Cat(I^n,𝒞)`$ is a simplicial cut. It is called the $`\eta `$-corner simplicial nerve $`𝒩^\eta `$ of $`𝒞`$.
Set $`H_{n+1}^\eta (𝒞):=H_n(𝒩^\eta (𝒞))`$ for $`n1`$. These homology theories are called branching and merging homology respectively and are exactly the same homology theories as that defined in and studied in .
And we have
###### Theorem 4.2.
The simplicial cut $`𝒩^\eta `$ is regular. The associated folding operator $`\mathrm{}_n^{𝒩^\eta }`$ coincides with the operator $`\mathrm{}_n^\eta `$ defined in . And therefore the associated homology theory $`HR_n^{𝒩^\eta }`$ coincide with the reduced corner homology $`HR_n^\eta `$ defined in .
It is useful for the sequel to remind some important properties of the folding operators associated to corner nerves.
###### Theorem 4.3.
Let $`𝒞`$ be an $`\omega `$-category. Let $`x`$ be an element of $`𝒩_n^{}(𝒞)`$. Then the following two conditions are equivalent :
1. the equality $`x=\mathrm{\Phi }_n^{}(x)`$ holds
2. for $`1in`$, one has $`ev_i^+x=_i^+x(0_n)`$ is $`0`$-dimensional and for $`1in2`$, one has $`_i^{}xIm(\mathrm{\Gamma }_{n2}^{}\mathrm{}\mathrm{\Gamma }_i^{})`$.
Another operator coming from which matters for this paper is the operator $`\theta _i^{}`$.
###### Definition 4.4.
Let $`x𝒩_n^{}(𝒞)`$ for some $`𝒞`$ such that for any $`1jn+1`$, $`_j^+x`$ is $`0`$-dimensional. Then $`x`$ is called a negative element of the branching nerve.
###### Theorem 4.5.
Let $`n2`$. There exists natural transformations
$$\theta _1^{},\mathrm{},\theta _{n1}^{}$$
from $`𝒩_n^{}`$ to itself satisfying the following properties :
1. If $`x`$ is a negative element of $`𝒩_n^{}(𝒞)`$, then for any $`1in1`$, $`\theta _i^{}x`$ is a negative element as well.
2. If $`x`$ is a negative element of $`𝒩_n^{}(𝒞)`$, then for any $`1in1`$, there exists a thin negative element $`y_i`$ of $`𝒩_{n+1}^{}(𝒞)`$ such that $`^{}y_ix`$ is a linear combination of thin negative elements.
3. There exists a composite of $`\theta _1^{},\mathrm{},\theta _{n1}^{}`$ which coincides with the negative folding operators on negative elements of $`𝒩_n^{}`$.
###### Sketch of proof.
Consider the $`\theta _1^{},\mathrm{},\theta _{n1}^{}`$ of . One has
$`_j^+\theta _i^{}=\{\begin{array}{c}\theta _{i1}^{}_j^+\text{ if }j<i\\ \theta _i^{}_j^+\text{ if }j>i+2\end{array}`$
$`_i^+\theta _i^{}={}_{}{}^{v}\psi _{i}^{}_i^+`$
$`_{i+1}^+\theta _i^{}=ϵ_{i+1}_{i+1}^+_i^{}+_iϵ_{i+1}_{i+1}^+_{i+1}^+`$
$`_{i+2}^+\theta _i^{}={}_{}{}^{v}\psi _{i}^{+}_{i+2}^+`$
where, for the last formula, $`{}_{}{}^{v}\psi _{i}^{\pm }`$ are other operators which is not important to explicitly define here : the only important thing is that $`_i^+\theta _i^{}`$ remains $`0`$-dimensional if the argument is $`0`$-dimensional. Hence property 1. As for property 2, it is enough to check it for $`i=1`$. And in this case, $`y`$ is a thin $`4`$-cube satisfying
$`_1^+y={}_{}{}^{v}\psi _{2}^{}\mathrm{\Gamma }_1^{}_1^+x`$
$`_2^+y=\mathrm{\Gamma }_2^{}_2^+x`$
$`_3^+y=ϵ_3(\mathrm{\Gamma }_1^{}_2^+_1^{}x+_1ϵ_2_2^+_2^+x)`$
$`_4^+y={}_{}{}^{v}\psi _{2}^{+}\mathrm{\Gamma }_2^{}_3^+x`$
Once again, we refer to for the precise definition of the operators involved in the above formulas. The only thing that matters here is the dimension of $`_i^+y`$.
By , we know that $`\mathrm{\Phi }^{}=\mathrm{\Theta }\mathrm{\Psi }`$ when $`\mathrm{\Theta }`$ is a composite of $`\theta _i^{}`$ and such that for $`x`$ negative, $`\mathrm{\Psi }x=x`$. Hence property 3. ∎
The graded set $`(\omega Cat(I^n,𝒞))_{n0}`$ endowed with the operations $`_i^\pm `$ above defined and by the maps $`ϵ_i(x)(k_1\mathrm{}k_{n+1})=x(k_1\mathrm{}\widehat{k_i}\mathrm{}k_{n+1})`$ for $`x\omega Cat(I^n,𝒞)`$ and $`1in+1`$ is a cubical set and is usually known as the cubical singular nerve of $`𝒞`$ . The use of the same notation $`ϵ_i`$ for the degeneracy maps of the cubical singular nerve and the degeneracy maps of the three simplicial nerves appearing in this paper is very confusing. Fortunately, we will not need the degeneracy maps of the cubical singular nerve in this work except for Theorem 4.5 right above.
## 5 The globular cut
The most direct way of constructing a cut of $`\omega `$-categories consists of using the composite of both functors $`:𝒞𝒞`$ and $`𝒩`$ where $`𝒩`$ is the simplicial nerve functor defined by Street <sup>1</sup><sup>1</sup>1Of course, the functor $`𝒩`$ can be viewed as a functor from $`\omega Cat_1`$ to $`Sets_+^{\mathrm{\Delta }^{op}}`$, but a “good” cut should not be extendable to a functor from $`\omega Cat`$ to $`Sets_+^{\mathrm{\Delta }^{op}}`$..
Let us start this section by recalling the construction of the free $`\omega `$-category $`\mathrm{\Delta }^n`$ generated by the faces of the $`n`$-simplex. The faces of the $`n`$-simplex are labeled by the strictly increasing sequences of elements of $`\{0,1,\mathrm{},n\}`$. The length of a sequence is equal to the dimension of the corresponding face plus one. If $`x`$ is a face of the $`n`$-simplex, its subfaces are all increasing sequences of $`\{0,1,\mathrm{},n\}`$ included in $`x`$. If $`x`$ is a face of the $`n`$-simplex, let $`R(x)`$ be the set of faces of $`x`$. If $`X`$ is a set of faces, then let $`R(X)=_{xX}R(x)`$. Notice that $`R(XY)=R(X)R(Y)`$ and that $`R(\{x\})=R(x)`$. Then $`\mathrm{\Delta }^n`$ is the free $`\omega `$-category generated by the $`R(x)`$ with the rules
1. For $`x`$ $`p`$-dimensional with $`p1`$,
$$s_{p1}(R(x))=R(s_x)$$
and
$$t_{p1}(R(x))=R(t_x)$$
where $`s_x`$ and $`t_x`$ are the sets of faces defined below.
2. If $`X`$ and $`Y`$ are two elements of $`\mathrm{\Delta }^n`$ such that $`t_p(X)=s_p(Y)`$ for some $`p`$, then $`XY`$ belongs to $`\mathrm{\Delta }^n`$ and $`XY=X_pY`$.
where $`s_x`$ (resp. $`t_x`$) is the set of subfaces of $`x`$ obtained by removing one element in odd position (resp. in even position). For instance, $`s_{(04589)}=\{(4589),(0489),(0458)\}`$ and $`t_{(04589)}=\{(0589),(0459)\}`$.
Sometimes we will write (for instance) $`(0<4<5<8<9)`$ instead of simply $`(04589)`$. Figure 2(b) gives the example of the $`2`$-simplex.
Let $`x\omega Cat(\mathrm{\Delta }^n,𝒞)`$. Then consider the labeling of the faces of respectively $`\mathrm{\Delta }^{n+1}`$ and $`\mathrm{\Delta }^{n1}`$ defined by :
* $`ϵ_i(x)(\sigma _0<\mathrm{}<\sigma _r)=x(\sigma _0<\mathrm{}<\sigma _{k1}<\sigma _k1<\mathrm{}<\sigma _r1)`$ if $`\sigma _{k1}<i`$ and $`\sigma _k>i`$.
* $`x(\sigma _0<\mathrm{}<\sigma _{k1}<i<\sigma _{k+1}1<\mathrm{}<\sigma _r1)`$ if $`\sigma _{k1}<i`$, $`\sigma _k=i`$ and $`\sigma _{k+1}>i+1`$.
* $`x(\sigma _0<\mathrm{}<\sigma _{k1}<i<\sigma _{k+2}1<\mathrm{}<\sigma _r1)`$ if $`\sigma _{k1}<i`$, $`\sigma _k=i`$ and $`\sigma _{k+1}=i+1`$.
and
$$_i(x)(\sigma _0<\mathrm{}<\sigma _s)=x(\sigma _0<\mathrm{}<\sigma _{k1}<\sigma _k+1<\mathrm{}<\sigma _s+1)$$
where $`\sigma _k,\mathrm{},\sigma _si`$ and $`\sigma _{k1}<i`$.
It can be checked that $`ϵ_i(x)`$ (resp. $`_i(x)`$) are $`\omega `$-functors from $`\mathrm{\Delta }^{n+1}`$ (resp. $`\mathrm{\Delta }^{n1}`$) to $`𝒞`$ . By construction, the map $`[n]\mathrm{\Delta }^n`$ induces then a functor from the well-known category $`\mathrm{\Delta }`$ whose associated presheaves are the simplicial sets to $`\omega Cat`$. Therefore $`𝒩(𝒞)=(\omega Cat(\mathrm{\Delta }^{},𝒞),_i,ϵ_i)`$ is a simplicial set which is called the simplicial nerve of $`𝒞`$.
###### Definition 5.1.
The globular cut $`𝒩^{gl}`$ (or the globular nerve) is the functor from $`\omega Cat_1`$ to $`Sets_+^{\mathrm{\Delta }^{op}}`$ defined by $`𝒩_n^{gl}(𝒞)=\omega Cat(\mathrm{\Delta }^n,𝒞)`$ for $`n0`$ and with $`𝒩_1^{gl}(𝒞)=𝒞_0\times 𝒞_0`$, and endowed with the augmentation map $`_1`$ from $`𝒩_0^{gl}(𝒞)=𝒞_1`$ to $`𝒩_1^{gl}(𝒞)=𝒞_0\times 𝒞_0`$ defined by $`_1x=(s_0x,t_0x)`$. The evaluation map $`ev`$ is defined by $`ev(x)=x((0\mathrm{}n))`$ for $`x\omega Cat(\mathrm{\Delta }^n,𝒞)`$. The homology theory $`H_n^{gl}:=H_n^{𝒩^{gl}}`$ is called the globular homology and $`HR_n^{gl}:=HR_n^{𝒩^{gl}}`$ the reduced globular homology.
Geometrically, the elements of $`𝒩_n^{gl}(𝒞)`$ are full $`(n+1)`$-globes. Figure 3 depicts a $`2`$-simplex in the globular nerve. The simplexes seen by the globular cut are intuitively transverse to the execution paths, as well as those of corner nerves. Hence the terminology of cuts.
Here is now the new definition of the globular homology of a globular $`\omega `$-category $`𝒞`$ :
###### Definition 5.2.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. We set
$$H_{n+1}^{gl}(𝒞):=H_n(𝒩^{gl}(𝒞))$$
for $`n1`$ and this homology theory is called the globular homology of $`𝒞`$.
## 6 Associating to any globe its corners
The purpose of the rest of the paper is to justify that Definition 5.2 is the right definition. This is not a mathematical statement of course ! We follow the order of the remarks at the very end of Section 1 which explain what kind of conditions the globular homology must fulfill. So we have first to construct $`h^{}`$ and $`h^+`$ and we must verify that geometrically, in homology, $`h^{}`$ and $`h^+`$ do what we expect to find. In fact, we refer to for intuitive explanations of $`h^{}`$ and $`h^+`$. We only recall here Figure 4 as an illustration and care only about the construction of $`h^{}`$.
###### Theorem 6.1.
Let $`\alpha \{,+\}`$. There exists one and only one morphism of cuts $`h^\alpha `$ from $`𝒩^{gl}`$ to $`𝒩^\alpha `$. Moreover, for any non-contracting $`\omega `$-category $`𝒞`$, both morphisms $`h^\alpha `$ from $`𝒩^{gl}(𝒞)`$ to $`𝒩^\alpha (𝒞)`$ are injective.
The rest of the section is devoted to the proof of Theorem 6.1. The following sequence of propositions establishes the existence of $`h^{}`$. The term $`\underset{¯}{cub}^n`$ denotes the set of faces of the $`n`$-cube, as described in Section 4.
We briefly recall how filling shells in the cubical singular nerve. This technical tool already appears in for $`\omega `$-groupoids and in for $`\omega `$-categories. A particular case can be found in .
###### Definition 6.2.
A $`n`$-shell in the cubical singular nerve is a family of $`2(n+1)`$ elements $`x_i^\pm `$ of $`\omega Cat(I^n,𝒞)^{}`$ such that $`_i^\alpha x_j^\beta =_{j1}^\beta x_i^\alpha `$ for $`1i<jn+1`$ and $`\alpha ,\beta \{,+\}`$.
If $`x_i^\pm `$ is a $`n`$-shell, then it induces a labeling $`x`$ on the set of faces of dimension at most $`n`$ of the $`(n+1)`$-cube in the following manner : let $`k_1\mathrm{}k_{n+1}`$ be a face of dimension at most $`n`$ ; then there exists $`i`$ such that $`k_i0`$ ; then let $`x(k_1\mathrm{}k_{n+1}):=x_i(k_1\mathrm{}\widehat{k_i}\mathrm{}k_{n+1})`$. The axiom satisfied by an $`n`$-shell ensures the coherence of the definition.
###### Proposition and definition 6.3.
Let $`x_i^\pm `$ be an $`(n1)`$-shell with $`n1`$.
* The labeling of the faces of dimension at most $`(n1)`$ of $`I^n`$ defined by $`x_i^\pm `$ always induces an $`\omega `$-functor and only one from $`I^n\backslash \{R(0_n)\}`$ to $`𝒞`$. Denote it by $`x`$.
* The $`n`$-shell $`(x_i^\pm )`$ is said fillable if there exists a morphism $`u`$ of $`𝒞`$ such that $`s_{n1}u=x\left(s_{n1}R(0_n)\right)`$ and $`t_{n1}u=x\left(t_{n1}R(0_n)\right)`$. In this case, there exists a unique $`\omega `$-functor $`x`$ from $`I^n`$ to $`𝒞`$ such that $`_i^\pm x=x_i^\pm `$ for $`1in`$ and $`x(0_n)=u`$.
###### Proof.
Using the freeness of $`I^n`$, the construction in the proof of Proposition 5.1 yields the $`\omega `$-functor $`x`$ from $`I^n\backslash \{R(0_n)\}`$ to $`𝒞`$. The hypotheses stated in were too strong indeed. If moreover the shell is fillable in the above sense, one concludes still as in the proof of Proposition 5.1. ∎
Now we can construct $`h^{}`$.
###### Theorem 6.4.
Let $`x`$ be an $`n`$-simplex of the globular simplicial nerve of $`𝒞`$. Then the map $`h_n^{}(x)`$ from $`\underset{¯}{cub}^{n+1}`$ to $`𝒞`$ defined by
1. $`+\{k_1\mathrm{}k_{n+1}\}`$ implies $`h_n^{}(x)(k_1\mathrm{}k_{n+1})=t_0x((0))`$ (notice that $`(0)`$ is the final state of $`\mathrm{\Delta }^n`$)
2. $`\{k_1,\mathrm{},k_{n+1}\}\{,0\}`$ and
$$\{k_1,\mathrm{},k_{n+1}\}\{0\}=\{k_{\sigma _0+1},\mathrm{},k_{\sigma _r+1}\}$$
with $`\sigma _0<\mathrm{}<\sigma _r`$ implies $`h_n^{}(x)(k_1\mathrm{}k_{n+1})=x((\sigma _0\mathrm{}\sigma _r))`$
3. $`h_n^{}(x)(_{n+1})=s_0x((n))`$ (notice that $`(n)`$ is the initial state of $`\mathrm{\Delta }^n`$)
yields an $`\omega `$-functor from $`I^{n+1}`$ to $`𝒞`$. Moreover, $`h^{}`$ induces a morphism of simplicial sets from the globular nerve of $`𝒞`$ to its negative corner nerve. And the map from $`𝒩_1^{gl}(𝒞)`$ to $`𝒩_1^{}(𝒞)`$ defined by $`(x,y)x`$ extends the previous morphism to the corresponding augmented simplicial nerves. Moreover for $`n0`$, $`h_n^{}`$ is a one-to-one map and the image of $`h_n^{}`$ contains exactly all cubes $`x`$ of the negative corner nerve such that as soon as $`_i^+x`$ exists, then it is $`0`$-dimensional.
There is no ambiguity to set $`h^{}(x)=h_n^{}(x)`$ if $`x`$ is an $`n`$-simplex of the globular cut.
In the sequel, in order to make easier the reading of the calculations, we suppose that an expression like $`(\sigma _0<\sigma _j\widehat{k}<\sigma _{j+1}<\mathrm{}<\sigma _r)`$ is the same thing as $`(\sigma _0<\sigma _j<\sigma _{j+1}<\mathrm{}<\sigma _r)`$ in $`\mathrm{\Delta }^{}`$ but with an additional information given within the calculation itself : here that $`\sigma _j\widehat{k}<\sigma _{j+1}`$ holds.
###### Proof.
One proves by induction on $`n`$ the following property $`P(n)`$ : “ For any $`n`$-simplex $`x`$ of the globular simplicial nerve of any $`\omega `$-category $`𝒞`$, the map $`h^{}(x)`$ from $`\underset{¯}{cub}^{n+1}`$ to $`𝒞`$ induces an $`\omega `$-functor and moreover an element of $`\omega Cat(I^{n+1},𝒞)^{}`$.”
Let $`x`$ be a $`0`$-simplex of the globular nerve of $`𝒞`$. Then $`x`$ is an $`\omega `$-functor from $`\mathrm{\Delta }^0`$ to $`𝒞`$, and therefore it can be identified with the $`1`$-morphism $`x((0))`$ of $`𝒞`$. Therefore
$`h^{}(x)(0)=x((0))`$ by rule 2
$`h^{}(x)(+)=t_0x((0))`$ by rule 1
$`h^{}(x)()=s_0x((0))`$ by rule 3
Therefore $`P(0)`$ is proved.
Now suppose that $`P(n)`$ is proved for $`n0`$. Let $`x`$ be a $`(n+1)`$-simplex of the globular simplicial nerve of some $`\omega `$-category $`𝒞`$. If $`+\{k_1,\mathrm{},k_{n+1}\}`$, then
$`_i^{}(h^{}(x))(k_1\mathrm{}k_{n+1})`$
$`=h^{}(x)(k_1\mathrm{}k_{i1}k_i\mathrm{}k_{n+1})`$ by definition of $`_i^{}`$ for $`1in+2`$
$`=t_0x((0))`$ by rule 1
$`=h^{}(_{i1}x)(k_1\mathrm{}k_{n+1})`$ again by rule 1
If $`+\{k_1,\mathrm{},k_{n+1}\}`$, i.e. if $`\{k_1,\mathrm{},k_{n+1}\}\{,0\}`$, set
$$\{k_1,\mathrm{},k_{n+1}\}\{0\}=\{k_{\sigma _0+1},\mathrm{},k_{\sigma _r+1}\}$$
with $`\sigma _0<\mathrm{}<\sigma _r`$. For a given $`i`$ such that $`1in+2`$, set
$$w_1\mathrm{}w_{n+2}=k_1\mathrm{}k_{i1}k_i\mathrm{}k_{n+1}$$
as word. Then let
$$\{w_1,\mathrm{},w_{n+2}\}\{0\}=\{w_{\tau _0+1},\mathrm{},w_{\tau _r+1}\}$$
with $`\tau _0<\mathrm{}<\tau _r`$. The relation between the sequence of $`\sigma _j`$ and the sequence of $`\tau _j`$ is as follows :
$`\sigma _j+1i1\sigma _j=\tau _j`$
$`\sigma _j+1i\sigma _j+1=\tau _j`$
And we have
$`_i^{}(h^{}(x))(k_1\mathrm{}k_{n+1})`$
$`=h^{}(x)(k_1\mathrm{}k_{i1}k_i\mathrm{}k_{n+1})\text{ by definition of }_i^{}`$
$`=x((\tau _0\mathrm{}\tau _r))\text{ by rule }\text{2}`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_0}\widehat{i2}<\widehat{i1}<\sigma _{j_0+1}+1<\mathrm{}<\sigma _r+1))`$
$`=(_{i1}x)((\sigma _0\mathrm{}\sigma _r))\text{ by definition of }_{i1}`$
$`=h^{}(_{i1}x)(k_1\mathrm{}k_{n+1})\text{ by rule }\text{2}`$
Therefore $`_i^{}(h^{}(x))=h^{}(_{i1}x)`$. And by rule 1, $`_i^+(h^{}(x))`$ is the constant $`\omega `$-functor from $`\underset{¯}{cub}^{n+1}`$ to $`𝒞`$ which sends any face of $`I^{n+1}`$ on $`t_0x((0))`$. Therefore $`(_i^\pm (h^{}(x)))_{1in+1}`$ is a $`(n+1)`$-shell in the cubical nerve of $`𝒞`$ which is fillable. By Proposition 6.3, the labeling $`h^{}(x)`$ of $`\underset{¯}{cub}^{n+2}`$ induces an $`\omega `$-functor from $`I^{n+2}`$ to $`𝒞`$ and $`P(n+1)`$ is proved.
By construction, the equality $`_i^{}(h^{}(x))=h^{}(_{i1}x)`$ holds for any $`n`$-simplex $`x`$ of the globular nerve and for $`1in+1`$. It remains to check that for such a simplex $`x`$, $`\mathrm{\Gamma }_i^{}(h^{}(x))=h^{}(ϵ_{i1}x)`$ for $`i1n+1`$. Consider a face $`k_1\mathrm{}k_{n+2}`$ of the $`(n+2)`$-cube. If $`+\{k_1,\mathrm{},k_{n+2}\}`$, then
$`\mathrm{\Gamma }_i^{}(h^{}(x))(k_1\mathrm{}k_{n+2})`$
$`=h^{}(x)(k_1\mathrm{}max(k_i,k_{i+1})\mathrm{}k_{n+2})`$ by definition of $`\mathrm{\Gamma }_i^{}`$
$`=t_0x((0))`$ by rule 1
$`=h^{}(ϵ_{i1}x)(k_1\mathrm{}k_{n+2})`$ again by rule 1
If $`+\{k_1,\mathrm{},k_{n+2}\}`$, i.e. if $`\{k_1,\mathrm{},k_{n+2}\}\{,0\}`$, set
$$\{k_1,\mathrm{},k_{n+2}\}\{0\}=\{k_{\sigma _0+1},\mathrm{},k_{\sigma _r+1}\}$$
with $`\sigma _0<\mathrm{}<\sigma _r`$. For a given $`i`$ such that $`1in+1`$,
$$\{k_1,\mathrm{},max(k_i,k_{i+1}),\mathrm{},k_{n+2}\}\{,0\}$$
and set $`w_1\mathrm{}w_{n+1}=k_1\mathrm{}max(k_i,k_{i+1})\mathrm{}k_{n+2}`$ as word. Then let
$$\{w_1,\mathrm{},w_{n+1}\}\{0\}=\{w_{\tau _0+1},\mathrm{},w_{\tau _s+1}\}$$
with $`\tau _0<\mathrm{}<\tau _s`$. One has to calculate
$`\mathrm{\Gamma }_i^{}(h^{}(x))(k_1\mathrm{}k_{n+2})`$
$`=h^{}(x)(k_1\mathrm{}max(k_i,k_{i+1})\mathrm{}k_{n+2})`$ by definition of $`\mathrm{\Gamma }_i^{}`$
$`=x((\tau _0\mathrm{}\tau _s))`$ by definition of $`h^{}`$
for some $`1in+2`$.
The situation can be decomposed in three mutually exclusive cases :
1. $`k_i=k_{i+1}=0`$. In this case, there exists a unique $`j_0`$ such that $`\sigma _{j_0}+1=i`$, $`s=r1`$ and
$`\sigma _j+1i1\sigma _j=\tau _j\text{ (in this case, }j<j_0\text{)}`$
$`\tau _{j_0}+1=i=\sigma _{j_0}+1`$
$`\sigma _j+1i+2\sigma _j1=\tau _{j1}\text{ (in this case, }j>j_0+1\text{)}`$
Then $`\sigma _{j_0+2}i+1`$ and
$`x((\tau _0\mathrm{}\tau _s))`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_0}=\widehat{i1}<\sigma _{j_0+2}1<\mathrm{}<\sigma _{s+1}1))`$
$`=(ϵ_{i1}x)(\sigma _0\mathrm{}\sigma _{j_0}\sigma _{j_0+1}\sigma _{j_0+2}\mathrm{}\sigma _{s+1})\text{ by definition of }ϵ_i`$
and since $`\sigma _{j_0+1}=i`$
$`=(h^{}(ϵ_{i1}x))(k_1\mathrm{}k_{n+2})\text{ by definition of }h^{}`$
2. $`k_i=k_{i+1}=`$. In this case, $`s=r`$ and
$`\sigma _j+1i1\sigma _j=\tau _j`$
$`\sigma _j+1i+2\sigma _j1=\tau _j`$
Then for some $`k`$,
$`x((\tau _0\mathrm{}\tau _s))`$
$`=x((\sigma _0<\mathrm{}<\sigma _k<\widehat{i1}<\sigma _{k+1}1<\mathrm{}<\sigma _r1))`$
$`=(ϵ_{i1}x)((\sigma _0\mathrm{}\sigma _k\sigma _{k+1}\mathrm{}\sigma _r))\text{ by definition of }ϵ_i`$
$`=(h^{}(ϵ_{i1}x))(k_1\mathrm{}k_{n+2})\text{ by definition of }h^{}`$
3. $`k_ik_{i+1}`$. Now $`s=r`$ and since $`\{k_i,k_{i+1}\}\{,0\}`$, then there exists a unique $`j_0`$ such that $`\sigma _{j_0}+1\{i,i+1\}`$ and we have
$`\sigma _j+1i1\sigma _j=\tau _j\text{ (in this case, }j<j_0\text{)}`$
$`\tau _{j_0}+1=i`$
$`\sigma _j+1i+2\sigma _j1=\tau _j\text{ (in this case, }j>j_0\text{)}`$
There are two subcases : $`\sigma _{j_0}+1=i`$ and $`\sigma _{j_0}+1=i+1`$. In the first situation,
$`x((\tau _0\mathrm{}\tau _s))`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_01}<\sigma _{j_0}=i1<\sigma _{j_0+1}1<\mathrm{}<\sigma _r1))`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_01}<\sigma _{j_0}<\sigma _{j_0+1}1<\mathrm{}<\sigma _r1))`$
$`=(ϵ_{i1}x)((\sigma _0<\mathrm{}<\sigma _{j_0}<\sigma _{j_0+1}<\mathrm{}<\sigma _r))\text{ by definition of }ϵ_i`$
$`=(h^{}(ϵ_{i1}x))(k_1\mathrm{}k_{n+2})\text{ by definition of }h^{}`$
In the second situation,
$`x((\tau _0\mathrm{}\tau _s))`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_01}<\sigma _{j_0}1=i1<\sigma _{j_0+1}1<\mathrm{}<\sigma _r1))`$
$`=x((\sigma _0<\mathrm{}<\sigma _{j_01}<\sigma _{j_0}1<\sigma _{j_0+1}1<\mathrm{}<\sigma _r1))`$
$`=(ϵ_{i1}x)((\sigma _0<\mathrm{}<\sigma _{j_0}<\sigma _{j_0+1}<\mathrm{}<\sigma _r))\text{ by definition of }ϵ_i`$
$`=(h^{}(ϵ_{i1}x))(k_1\mathrm{}k_{n+2})\text{ by definition of }h^{}`$
Notice that $`h^{}`$ induces a natural transformation from $`CR_{}^{gl}`$ to $`CR_{}^{}`$ which is not injective. Consider for example the $`\omega `$-category consisting of two composable $`1`$-morphisms $`u`$ and $`v`$ with $`t_0u=s_0v`$. The $`0`$-simplexes $`u`$ and $`u_0v`$ of $`𝒩_0^{gl}`$ have indeed the same image by $`h^{}`$ in $`CR_1^{}`$. To see that, consider the thin square $`c`$ from $`I^2`$ to $`𝒞`$ defined by $`c(0)=u_0v`$, $`c(0+)=t_0v`$, $`c(0)=u`$, $`c(+0)=v`$ and $`c(00)=u_0v`$.
Now we arrive at :
###### Theorem 6.5.
There exists one and only one morphism of cuts from $`𝒩^{gl}`$ to $`𝒩^{}`$.
The proof of this theorem uses Theorem 8.3 assertion 1 as shortcut. There is no vicious circle because the uniqueness of $`h^{}`$ and $`h^+`$ is used nowhere in this paper. The only fact which is used is that Theorem 6.4 provides a natural transformation from $`𝒩^{gl}`$ to $`𝒩^{}`$ which is injective on the underlying sets.
###### Proof.
Let $`h`$ and $`h^{}`$ be two morphisms of cuts from $`𝒩^{gl}`$ to $`𝒩^{}`$. One proves by induction on $`n`$ that $`h_n`$ and $`h_n^{}`$ from $`𝒩_n^{gl}`$ to $`𝒩_n^{}`$ coincide. For $`n=0`$, $`𝒩_0^{gl}=𝒩_n^{}=tr^0`$. The only natural transformation from $`tr^0`$ to itself is $`Id_{tr^0}`$, therefore $`h_0=h_0^{}`$.
Suppose $`P(n)`$ proved for some $`n0`$. Then for any $`x𝒩_{n+1}^{gl}(𝒞)`$, and for any $`0in+1`$,
$`_{i+1}^{}h_{n+1}(x)`$ $`=h_n(_ix)`$ since $`h`$ morphism of simplicial sets
$`=h_n^{}(_ix)`$ by induction hypothesis
$`=_{i+1}^{}h_{n+1}^{}(x)`$ since $`h^{}`$ morphism of simplicial sets
Now with $`1jn+2`$,
$`(_j^+h_{n+1}(x))(_{n+1})`$
$`=h_{n+1}(x)(\mathrm{}[+]_j\mathrm{})`$
$`=h_{n+1}(x)\left(t_0R(\mathrm{}[0]_j\mathrm{})\right)`$
$`=t_0\left(h_{n+1}(x)(R(\mathrm{}[0]_j\mathrm{}))\right)\text{ since }h_{n+1}(x)\text{ }\omega \text{-functor}`$
$`=t_0\left((_1^{}\mathrm{}\widehat{_j^{}}\mathrm{}_{n+2}^{}h_{n+1}(x))(0)\right)`$
$`=t_0\left(h_0(_0\mathrm{}\widehat{_{j1}}\mathrm{}_{n+1}x)(0)\right)\text{ since }h\text{ morphism of simplicial sets}`$
$`=t_0\left((_0\mathrm{}\widehat{_{j1}}\mathrm{}_{n+1}x)((0))\right)`$
So the $`0`$-morphism $`_j^+h_{n+1}(x))(_{n+1})`$ is the value of the constant map $`t_0x`$ of Theorem 8.3 (denoted by $`T(x)`$ in Section 10).
Let $`𝒟`$ be the unique $`\omega `$-category such that $`𝒟=\mathrm{\Delta }^{n+1}`$ and with $`𝒟_0=\{\alpha ,\beta \}`$, $`s_0(𝒟)=\{\alpha \}`$, $`t_0(𝒟)=\{\beta \}`$ and $`\alpha \beta `$. And consider $`Id_{\mathrm{\Delta }^{n+1}}𝒩_{n+1}^{gl}(𝒟)`$.
Suppose that $`+\{k_1,\mathrm{},k_{n+2}\}\{,+\}`$ and suppose that at least two $`k_i`$ are equal to $`+`$. Then there exists a $`1`$-morphism $`u`$ of $`I^{n+2}`$ such that $`s_0u=\mathrm{}_1\mathrm{}\mathrm{}_{n+2}`$ with exactly one $`\mathrm{}_i`$ equal to $`+`$ and such that $`t_0u=k_1\mathrm{}k_{n+2}`$. Then
$$s_0\left(h_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(u)\right)=h_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(\mathrm{}_1\mathrm{}\mathrm{}_{n+2})=\beta $$
by the previous calculation. Since $`\beta `$ is the unique morphism of $`𝒟`$ with $`0`$-source $`\beta `$, then $`h_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(u)=\beta `$ and therefore
$$h_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(k_1\mathrm{}k_{n+2})=\beta .$$
Suppose now that $`+\{k_1,\mathrm{},k_{n+2}\}`$ with perhaps some $`0`$ in the set. Then
$$s_0\left(h_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(k_1\mathrm{}k_{n+2})\right)=\beta $$
and therefore
$$evh_{n+1}(Id_{\mathrm{\Delta }^{n+1}})(k_1\mathrm{}k_{n+2})=\beta =T\left(Id_{\mathrm{\Delta }^{n+1}}\right).$$
The $`\omega `$-functor $`x`$ from $`\mathrm{\Delta }^{n+1}`$ to $`𝒞`$ induces a non-contracting $`\omega `$-functor $`\overline{x}`$ from $`𝒟`$ to $`𝒞`$ with $`\overline{x}(\alpha )=S(x)`$ ($`S(x)`$ being the value of the constant map $`s_0x`$ by Theorem 8.3) and $`\overline{x}(\beta )=T(x)`$ which sends $`Id_{\mathrm{\Delta }^{n+1}}𝒩_{n+1}^{gl}(𝒟)`$ on $`x𝒩_{n+1}^{gl}(𝒞)`$. So by naturality,
$$evh_{n+1}(x)(k_1\mathrm{}k_{n+2})=T(x).$$
Therefore for any $`1jn+2`$, $`_j^+h_{n+1}(x)=_j^+h_{n+1}^{}(x)`$. By hypothesis, $`ev(h_{n+1}(x))=ev(x)=ev(h_{n+1}^{}(x))`$. So $`h_{n+1}(x)`$ and $`h_{n+1}^{}(x)`$ induce the same labeling of the faces of $`I^{n+2}`$ and $`P(n+1)`$ is proved. ∎
Without explanation, here is the construction of $`h^+`$ :
###### Proposition 6.6.
Let $`x`$ be an $`n`$-simplex of the globular simplicial nerve of $`𝒞`$. Then the map $`h_n^+(x)`$ from $`\underset{¯}{cub}^{n+1}`$ to $`𝒞`$ defined by
1. $`\{k_1\mathrm{}k_{n+1}\}`$ implies $`h_n^+(x)(k_1\mathrm{}k_{n+1})=s_0x((n))`$ (notice that $`(n)`$ is the initial state of $`\mathrm{\Delta }^n`$)
2. $`\{k_1,\mathrm{},k_{n+1}\}\{+,0\}`$ and
$$\{k_1,\mathrm{},k_{n+1}\}\{0\}=\{k_{\sigma _0+1},\mathrm{},k_{\sigma _r+1}\}$$
with $`\sigma _0<\mathrm{}<\sigma _r`$ implies $`h_n^+(x)(k_1\mathrm{}k_{n+1})=x((\sigma _0\mathrm{}\sigma _r))`$
3. $`h_n^+(x)(+_{n+1})=t_0x((0))`$ (notice that $`(0)`$ is the final state of $`\mathrm{\Delta }^n`$)
yields an $`\omega `$-functor from $`I^{n+1}`$ to $`𝒞`$. Moreover, $`h^+`$ induces a morphism of simplicial sets from the globular nerve of $`𝒞`$ to its positive corner nerve. And the map from $`𝒩_1^{gl}(𝒞)`$ to $`𝒩_1^+(𝒞)`$ defined by $`(x,y)y`$ extends the previous morphism to the corresponding augmented simplicial nerves. Moreover for $`n0`$, $`h_n^+`$ is a one-to-one map and the image of $`h_n^+`$ contains exactly all cubes $`x`$ of the positive corner nerve such that as soon as $`_i^{}x`$ exists, then it is $`0`$-dimensional.
###### Question 6.7.
Is it possible to find an appropriate setting where the globular cut would be an initial object ? Is it possible to characterize the diagram of cuts of Figure 1 ?
As immediate corollary of the construction of $`h^{}`$ and its injectivity, let us introduce the analogue of Proposition 6.3 in the globular nerve.
###### Definition 6.8.
In a simplicial set $`A`$, a $`n`$-shell is a family $`(x_i)_{i=0,\mathrm{},n+1}`$ of $`(n+2)`$ $`n`$-simplexes of $`A`$ such that for any $`0i<jn+1`$, $`_ix_j=_{j1}x_i`$.
###### Proposition 6.9.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Consider a $`n`$-shell $`(x_i)_{i=0,\mathrm{},n+1}`$ of the globular simplicial nerve of $`𝒞`$. Then
1. The labeling defined by $`(x_i)_{i=0,\mathrm{},n+1}`$ yields an $`\omega `$-functor $`x`$ (and necessarily exactly one) from $`\mathrm{\Delta }^{n+1}\backslash \{(01\mathrm{}n+1)\}`$ to $`𝒞`$.
2. Let $`u`$ be a morphism of $`𝒞`$ such that
$$s_nu=x\left(s_nR((01\mathrm{}n+1))\right)$$
and
$$t_nu=x\left(t_nR((01\mathrm{}n+1))\right)$$
Then there exists one and only one $`\omega `$-functor still denoted by $`x`$ from $`\mathrm{\Delta }^{n+1}`$ to $`𝒞`$ such that for any $`0in+1`$, $`_ix=x_i`$ and
$$x((01\mathrm{}n+1))=u.$$
## 7 Regularity of the globular cut
This section is devoted to the proof of the following theorem.
###### Theorem 7.1.
The globular cut is regular.
The principle of this proof is to use the injectivity of the natural transformation $`h^{}`$ from $`𝒩^{gl}`$ to $`𝒩^{}`$ and to use the regularity of $`𝒩^{}`$.
The folding operator $`\mathrm{\Phi }_n^{gl}:=\mathrm{\Phi }_n^{𝒩^{gl}}`$ is called the $`n`$-dimensional globular folding operator and we set $`\mathrm{}_n^{gl}:=\mathrm{}_n^{𝒩^{gl}}`$. It is clear that rule 1 and rule 2 of Definition 3.3 are satisfied. We have to check the rest of it.
###### Theorem 7.2.
For any natural transformation of functors $`\mu `$ from $`𝒩_{n1}^{gl}`$ to $`𝒩_n^{gl}`$ with $`n1`$, and for any natural map $`\mathrm{}`$ from $`tr^{n1}`$ to $`𝒩_{n1}^{gl}`$ such that $`ev\mathrm{}=Id_{tr^{n1}}`$, there exists one and only one natural transformation denoted by $`\mu .\mathrm{}`$ from $`tr^n`$ to $`𝒩_n^{gl}`$ such that the following diagram commutes
where $`i_n`$ is the canonical inclusion functor from $`tr^{n1}`$ to $`tr^n`$.
###### Proof.
The natural transformation $`h^{}\mathrm{}`$ from $`tr^{n1}`$ to $`𝒩_{n1}^{}`$ can be lifted to a natural transformation $`(h^{}(\mu )).(h^{}\mathrm{})`$ from $`tr^n`$ to $`𝒩_n^{}`$ since the cut $`𝒩^{}`$ is regular. Since $`h^{}(\mu .\mathrm{})=(h^{}(\mu )).(h^{}\mathrm{})`$ and since $`h^{}`$ is one-to-one in positive degree, there is at most one solution for this lifting problem.
Let $`x𝒞_{n+1}`$. For $`0in`$, the natural transformation
$$ev_i(h^{}(\mu ).(h^{}\mathrm{})):tr^ntr^{n1}$$
is of the form $`d_{m_i}^{\alpha _i}`$ for some $`\alpha _i\{,+\}`$ and some $`m_in`$. Therefore
$`_i(h^{}(\mu ).(h^{}\mathrm{}))`$
$`=_i(h^{}(\mu ).(h^{}\mathrm{}))i_nd_{m_i}^{\alpha _i}`$ by Definition 3.3 rule 5b
$`=_ih^{}(\mu )h^{}\mathrm{}d_{m_i}^{\alpha _i}`$ by hypothesis
$`=_ih^{}\mu \mathrm{}d_{m_i}^{\alpha _i}`$
$`=h^{}_i\mu \mathrm{}d_{m_i}^{\alpha _i}`$ since $`h^{}`$ morphism of simplicial sets
So $`_i(h^{}(\mu ).(h^{}\mathrm{}))(x)h^{}(𝒩_{n1}^{gl}(𝒞))`$ for any $`0in`$ and by Proposition 6.9, $`(h^{}(\mu ).(h^{}\mathrm{}))(x)h^{}(𝒩_n^{gl}(𝒞))`$. Let $`\mathrm{}^{}(x)`$ be the unique element of $`𝒩_n^{gl}(𝒞)`$ such that
$$h^{}\mathrm{}^{}(x):=(h^{}(\mu ).(h^{}\mathrm{}))(x)$$
Then $`\mathrm{}^{}`$ is a solution. ∎
###### Corollary 7.3.
The equalities $`h^{}\mathrm{\Phi }^{gl}=\mathrm{\Phi }^{}h^{}`$ and $`h^+\mathrm{\Phi }^{gl}=\mathrm{\Phi }^+h^+`$ hold.
###### Proof.
It is a consequence of the naturality of $`h^{}`$ and $`h^+`$ and of Proposition 3.4. ∎
Now here is a characterization of globular folding operators :
###### Proposition 7.4.
Let $`x`$ be a $`n`$-simplex of the globular nerve of $`𝒞`$. Then $`x=\mathrm{\Phi }^{gl}(x)`$ if and only if for $`0in2`$, $`_ixIm(ϵ_{n2}\mathrm{}ϵ_i)`$.
###### Proof.
The equality $`x=\mathrm{\Phi }^{gl}(x)`$ implies $`h^{}(x)=\mathrm{\Phi }^{}(h^{}(x))`$, implies by Theorem 4.3 that for $`1in1`$,
$`h^{}(_{i1}x)`$ $`=`$ $`_i^{}(h^{}(x))=\mathrm{\Gamma }_{n1}^{}\mathrm{}\mathrm{\Gamma }_i^{}\mathrm{}_i^{}d_i^{()}h^{}(x)(0_{n+1})`$
$`=`$ $`h^{}\left(ϵ_{n2}\mathrm{}ϵ_{i1}\mathrm{}_i^{gl}s_ix((0\mathrm{}n))\right)`$
therefore $`_{i1}xIm(ϵ_{n2}\mathrm{}ϵ_{i1})`$. Conversely, if for $`0in2`$, $`_ixIm(ϵ_{n2}\mathrm{}ϵ_i)`$, then $`h^{}(x)=\mathrm{\Phi }^{}h^{}(x)=h^{}\mathrm{\Phi }^{gl}(x)`$ and therefore $`x=\mathrm{\Phi }^{gl}(x)`$. ∎
###### Theorem 7.5.
The globular folding operator $`\mathrm{\Phi }^{gl}`$ induces the identity map on the globular reduced chain complex $`CR_{}^{gl}`$.
###### Proof.
Consider the $`\theta _i^{}`$ operators of Theorem 4.5. If $`x𝒩_n^{gl}`$, then $`h^{}x`$ is negative. So $`\theta _i^{}h^{}x`$ is also negative by Theorem 4.5(1) and determines a unique element $`\theta _i^{gl}x𝒩_n^{gl}`$ such that $`h^{}\theta _i^{gl}x=\theta _i^{}h^{}x`$. It is clear that these operators $`\theta _i^{gl}`$ induces the identity map on the reduced globular complex by Theorem 4.5(2). Since $`\mathrm{\Phi }^{}h^{}x`$ is also negative, then by Theorem 4.5(3),
$$\mathrm{\Phi }^{}h^{}x=\theta _{i_1}^{}\mathrm{}\theta _{i_s}^{}h^{}x$$
for some sequence $`i_1,\mathrm{},i_s`$. Therefore by the injectivity of $`h^{}`$,
$$\mathrm{\Phi }^{gl}x=\theta _{i_1}^{gl}\mathrm{}\theta _{i_s}^{gl}x$$
###### Theorem 7.6.
In the reduced globular complex, one has
$$\mathrm{}_n^{gl}(x_py)=\mathrm{}_n^{gl}(x)+\mathrm{}_n^{gl}(y)$$
for any morphisms $`x`$ and $`y`$ of $`𝒞`$ of dimension $`n`$ and for $`1pn1`$.
###### Sketch of proof.
One has
$`h^{}(\mathrm{}_n^{gl}(x_py))`$ $`=`$ $`\mathrm{}_n^{}(x_py)`$
$`=`$ $`\mathrm{}_n^{}(x)+\mathrm{}_n^{}(y)+t_1+^{}t_2`$
$`=`$ $`h^{}(\mathrm{}_n^{gl}(x))+h^{}(\mathrm{}_n^{gl}(y))+t_1+^{}t_2`$
with $`t_1`$ a thin $`(n+1)`$-cube and $`t_2`$ a thin $`(n+2)`$-cube. The proof made in shows that $`t_1`$ and $`t_2`$ are in the image of $`h^{}`$. Indeed, the existence of $`t_1`$ and $`t_2`$ comes from the vanishing of some globular nerve. Therefore $`t_1=h^{}(T_1)`$ and $`t_2=h^{}(T_2)`$ where $`T_1`$ is a thin $`n`$-simplex and $`T_2`$ a thin $`(n+1)`$-simplex. This completes the proof. ∎
In fact one can explicitly verify that if $`x`$ and $`y`$ are two $`n`$-morphisms of $`𝒞`$, then $`\mathrm{}_n^{gl}(x_{n1}y)\mathrm{}_n^{gl}(x)\mathrm{}_n^{gl}(y)`$ is a boundary in the normalized globular complex. It suffices to consider the thin $`(n+1)`$-cube $`B_{n1}^n(x,y)`$ of which turns to be in the image of $`h^{}`$ because it is negative. Therefore with $`b(x,y)\omega Cat(\mathrm{\Delta }^n,𝒞)`$ defined by $`_ib(x,y)=ϵ_{n2}\mathrm{}ϵ_i\mathrm{}_{i+1}^{gl}d_{i+1}^{()^{i+1}}x`$ for $`0in3`$ (observe that $`d_{i+1}^{()^{i+1}}x=d_{i+1}^{()^{i+1}}y`$), $`_{n2}b(x,y)=\mathrm{}_n^{gl}y`$, $`_{n1}b(x,y)=\mathrm{}_n^{gl}(x_{n1}y)`$, $`_nb(x,y)=\mathrm{}_n^{gl}x`$, one has
$$b(x,y)=\pm \left(\mathrm{}_n^{gl}(x_{n1}y)\mathrm{}_n^{gl}(x)\mathrm{}_n^{gl}(y)\right)+\text{ degenerate elements}.$$
## 8 Example of calculations of globular homology
The main goal of this section is to prove the vanishing of the globular homology of the $`n`$-cube in positive dimension for all $`n0`$. However we also study the case of the $`\omega `$-category $`2_n`$ generated by one $`n`$-morphism and pose some questions about the globular homology of the $`\omega `$-category generated by a composable pasting scheme in the sense of .
###### Theorem 8.1.
For any $`p>0`$ and any $`n0`$, $`H_p^{gl}(2_n)=0`$.
###### Proof.
For $`p=1`$, it is obvious. For $`p>1`$, one has
$$H_p^{gl}(2_n)H_{p1}(2_n)H_{p1}(2_{n1})=0$$
where $`H_{}(𝒟)`$ means the simplicial homology of the simplicial nerve of the $`\omega `$-category $`𝒟`$. ∎
###### Definition 8.2.
Let $`𝒞`$ be an $`\omega `$-category and let $`\alpha `$ and $`\beta `$ be two $`0`$-morphisms of $`𝒞`$. Then the bilocalization of $`𝒞`$ with respect to $`\alpha `$ and $`\beta `$ is the $`\omega `$-subcategory of $`𝒞`$ obtained by keeping in dimension $`0`$ only $`\alpha `$ and $`\beta `$ and by keeping in positive dimension all morphisms $`x`$ such that $`s_0x=\alpha `$ and $`t_0x=\beta `$. It is denoted by $`𝒞[\alpha ,\beta ]`$.
###### Theorem 8.3.
Let $`𝒞`$ be a non-contracting $`\omega `$-category.
1. Let $`x`$ be an $`\omega `$-functor from $`\mathrm{\Delta }^n`$ to $`𝒞`$ for some $`n0`$. Then the set maps
$$(\sigma _0\mathrm{}\sigma _r)s_0x((\sigma _0\mathrm{}\sigma _r))$$
and
$$(\sigma _0\mathrm{}\sigma _r)t_0x((\sigma _0\mathrm{}\sigma _r))$$
from the underlying set of faces of $`\mathrm{\Delta }^n`$ to $`𝒞_0`$ are constant. The unique value of $`s_0x`$ is denoted by $`S(x)`$ and the unique value of $`t_0x`$ is denoted by $`T(x)`$.
2. For any pair $`(\alpha ,\beta )`$ of $`0`$-morphisms of $`𝒞`$, for any $`n1`$, and for any $`0in`$, then $`_i\left(𝒩_n^{gl}(𝒞[\alpha ,\beta ])\right)𝒩_{n1}^{gl}(𝒞[\alpha ,\beta ])`$.
3. For any pair $`(\alpha ,\beta )`$ of $`0`$-morphisms of $`𝒞`$, for any $`n0`$, and for any $`0in`$, then $`ϵ_i\left(𝒩_n^{gl}(𝒞[\alpha ,\beta ])\right)𝒩_{n+1}^{gl}(𝒞[\alpha ,\beta ])`$.
4. By setting, $`G^{\alpha ,\beta }𝒩_n^{gl}(𝒞):=𝒩_n^{gl}(𝒞[\alpha ,\beta ])`$ for $`n0`$ and $`G^{\alpha ,\beta }𝒩_1^{gl}(𝒞):=\{(\alpha ,\beta ),(\beta ,\alpha )\}`$, one obtains a $`(𝒞_0\times 𝒞_0)`$-graduation on the globular nerve ; in particular, one has the direct sum of augmented simplicial sets
$$𝒩_{}^{gl}(𝒞)=\underset{(\alpha ,\beta )𝒞_0\times 𝒞_0}{}G^{\alpha ,\beta }𝒩_{}^{gl}(𝒞)$$
and $`G^{\alpha ,\beta }𝒩_{}^{gl}(𝒞)=𝒩_{}^{gl}(𝒞[\alpha ,\beta ])`$.
###### Proof.
The only non-trivial part is the first assertion. Let $`P(n)`$ be the property : “for any non-contracting $`\omega `$-category $`𝒞`$ and any $`\omega `$-functor $`x`$ from $`\mathrm{\Delta }^n`$ to $`𝒞`$, the set map $`(\sigma _0\mathrm{}\sigma _r)s_0x((\sigma _0\mathrm{}\sigma _r))`$ from the set of faces of $`\mathrm{\Delta }^n`$ to $`𝒞_0`$ is constant.”
There is nothing to check for $`P(0)`$. For $`P(1)`$, if $`x`$ is an $`\omega `$-functor from $`\mathrm{\Delta }^1`$ to $`𝒞`$, then $`s_1x((01))=x((1))`$ and $`t_1x((01))=x((0))`$ in $`𝒞`$. Therefore
$$s_0x((01))=s_0s_1x((01))=s_0x((1))$$
and
$$s_0x((0))=s_0t_1x((01))=s_0x((01)).$$
Therefore $`P(1)`$ is true.
Suppose $`P(n)`$ proved for some $`n1`$ and let us prove $`P(n+1)`$. For any $`1in`$, the $`\omega `$-functor $`x:\mathrm{\Delta }^{n+1}𝒞`$ induces an $`\omega `$-functor on the $`\omega `$-category $`\mathrm{\Delta }_i^{n+1}`$ generated by the face $`(0\mathrm{}\widehat{i}\mathrm{}n+1)`$ and its subfaces. One has an isomorphism of $`\omega `$-categories $`\mathrm{\Delta }^n\mathrm{\Delta }_i^{n+1}`$. Therefore the restriction of $`s_0x`$ to the faces of $`\mathrm{\Delta }_i^{n+1}`$ is constant by induction hypothesis. Now it is clear that $`\mathrm{\Delta }_i^{n+1}\mathrm{\Delta }_{i+1}^{n+1}\mathrm{\Delta }^{n1}\mathrm{}`$ since $`n1`$. Therefore the set map $`s_0x`$ restricted to $`\mathrm{\Delta }_i^{n+1}\mathrm{\Delta }_{i+1}^{n+1}`$ is constant. Therefore the restriction of the set map $`s_0x`$ to the faces of dimension at most $`n`$ of $`\mathrm{\Delta }^{n+1}`$ is constant. We know that
$$s_nR((01\mathrm{}n+1))=\mathrm{\Psi }(X_0,X_1,\mathrm{},X_s)$$
where $`X_0,X_1,\mathrm{},X_s`$ are faces of $`\mathrm{\Delta }^{n+1}`$ of dimension at most $`n`$. So
$`s_0x((01\mathrm{}n+1))`$ $`=s_0s_{n+1}x((01\mathrm{}n+1))`$
$`=s_0x\left(s_nR((01\mathrm{}n+1))\right)`$ since $`x`$ $`\omega `$-functor
$`=s_0x\mathrm{\Psi }(X_0,X_1,\mathrm{},X_s)`$
where $`\mathrm{\Psi }`$ is a function using only the compositions of $`\mathrm{\Delta }^{n+1}`$. Then
$$x\mathrm{\Psi }(X_0,X_1,\mathrm{},X_s)=\mathrm{\Psi }^{}(x(X_0),x(X_2),\mathrm{},x(X_s))$$
where $`\mathrm{\Psi }^{}`$ is obtained from $`\mathrm{\Psi }`$ by replacing $`_i`$ by $`_{i+1}`$ since $`x`$ is an $`\omega `$-functor from $`\mathrm{\Delta }^{n+1}`$ to $`𝒞`$. So
$$s_0x((01\mathrm{}n+1))=\mathrm{\Psi }^{}(s_0x(X_0),s_0x(X_2),\mathrm{},s_0x(X_s))=s_0x(X_0)$$
with the axioms of $`\omega `$-categories. Therefore $`P(n+1)`$ is proved. ∎
###### Definition 8.4.
Let $`𝒞`$ be a non-contracting $`\omega `$-category with exactly one initial state $`\alpha `$ and one final state $`\beta `$. Then the bilocalization $`𝒞[\alpha ,\beta ]`$ is also non-contracting and one can set $`\mathrm{\Omega }𝒞=(𝒞[\alpha ,\beta ])`$.
###### Theorem 8.5.
Let $`n1`$. Then $`\mathrm{\Omega }\mathrm{\Delta }^n=I^{n1}`$ and $`\mathrm{\Omega }I^{n1}=P^{n1}`$ where $`P^{n1}`$ is the free $`\omega `$-category generated by the composable pasting scheme of the faces of the $`(n1)`$-dimensional permutohedron.
###### Theorem 8.6.
For any $`n0`$, and any $`p>0`$, $`H_p^{gl}(I^n)=0`$.
###### Proof.
One has $`H_p^{gl}(I^n)=_{(\alpha ,\beta )𝒞_0\times 𝒞_0}H_p^{gl}(I^n[\alpha ,\beta ])`$ by Theorem 8.3. So it suffices to prove the vanishing of $`H_p^{gl}(I^n[\alpha ,\beta ])`$ as soon as $`I^n[\alpha ,\beta ]`$ contains morphisms in strictly positive dimension to prove the theorem.
Let $`\alpha `$ and $`\beta `$ be two $`0`$-morphisms of $`I^n`$ such that $`I^n[\alpha ,\beta ]`$ contains other morphisms than $`\alpha `$ and $`\beta `$. Then in particular it contains some $`1`$-morphisms from $`\alpha `$ to $`\beta `$ which is a composite of $`1`$-dimensional faces of $`I^n`$. Suppose that $`\alpha =k_1\mathrm{}k_n`$. Then $`\beta `$ is obtained from $`\alpha `$ by replacing some $`k_i`$ equal to $``$ by $`+`$. Let $`k_{\sigma _1},\mathrm{},k_{\sigma _r}`$ be these $`k_i`$. Then
$$I^n[\alpha ,\beta ]I^r[_r,+_r]$$
as $`\omega `$-category. Therefore it suffices to prove that $`H_p^{gl}(I^n[_n,+_n])`$ vanishes.
The vanishing of $`H_1^{gl}(I^n[_n,+_n])`$ is obvious. One has
$$H_p^{gl}(I^n[_n,+_n])=H_{p1}(P^n)$$
for $`p2`$ by Theorem 8.5 and $`H_{p1}(P^n)=0`$ because the simplicial nerve of a composable pasting scheme is contractible . ∎
###### Theorem 8.7.
For any $`n0`$, and any $`p>0`$, $`H_p^{gl}(\mathrm{\Delta }^n)=0`$.
###### Proof.
By proceeding as in Theorem 8.6, we see that it suffices to prove that
$$H_p^{gl}(\mathrm{\Delta }^n[(r),(s)])=0$$
for any pair $`((r),(s))`$ of $`0`$-morphisms of $`\mathrm{\Delta }^n`$ and for $`n2`$. However, $`\mathrm{\Delta }^n[(r),(s)]`$ is non-empty if and only if $`r>s`$ with our conventions and in this case,
$$\mathrm{\Delta }^n[(r),(s)]\mathrm{\Delta }^{rs}[(rs),(0)].$$
Therefore $`H_p^{gl}(\mathrm{\Delta }^n[(r),(s)])H_{p1}(I^{rs1})`$ by Theorem 8.5. ∎
More generally, as in , one sees that if $`𝒞`$ is a non-contracting $`\omega `$-category such that $`𝒞`$ is the free $`\omega `$-category generated by a composable pasting scheme in the sense of , then $`H_p^{gl}(𝒞)=0`$ for $`p1`$. This is related to the problem of the existence of the derived pasting scheme of a given composable pasting scheme .
###### Conjecture 8.8.
Let $`𝒞`$ be an $`\omega `$-category which is the free $`\omega `$-category generated by a composable pasting scheme (therefore $`𝒞`$ is non-contracting). Then for any $`p>0`$, $`H_p^{gl}(𝒞)=0`$.
## 9 Relation between the new globular homology and the old one
First of all, recall the definition of both formal corner homology theories from .
###### Definition 9.1.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Set
* $`CF_0^{}(𝒞):=𝒞_0`$
* $`CF_1^{}(𝒞):=𝒞_1`$
* $`CF_n^{}(𝒞)=𝒞_n/\{x_0y=x,x_1y=x+y,\mathrm{},x_{n1}y=x+y\text{ mod }tr^{n1}𝒞\}`$ for $`n2`$
with the differential map $`s_{n1}t_{n1}`$ from $`CF_n^{}(𝒞)`$ to $`CF_{n1}^{}(𝒞)`$ for $`n2`$ and $`s_0`$ from $`CF_1^{}(𝒞)`$ to $`CF_0^{}(𝒞)`$. This chain complex is called the formal negative corner complex. The associated homology is denoted by $`HF^{}(𝒞)`$ and is called the formal negative corner homology of $`𝒞`$. The map $`CF_{}^{}`$ (resp. $`HF_{}^{}`$) induces a functor from $`\omega Cat_1`$ to $`Comp(Ab)`$ (resp. $`Ab`$).
and symmetrically
###### Definition 9.2.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Set
* $`CF_0^+(𝒞):=𝒞_0`$
* $`CF_1^+(𝒞):=𝒞_1`$
* $`CF_n^+(𝒞)=𝒞_n/\{x_0y=y,x_1y=x+y,\mathrm{},x_{n1}y=x+y\text{ mod }tr^{n1}𝒞\}`$ for $`n2`$
with the differential map $`s_{n1}t_{n1}`$ from $`CF_n^+(𝒞)`$ to $`CF_{n1}^+(𝒞)`$ for $`n2`$ and $`t_0`$ from $`CF_1^+(𝒞)`$ to $`CF_0^+(𝒞)`$. This chain complex is called the formal positive corner complex. The associated homology is denoted by $`HF^+(𝒞)`$ and is called the formal positive corner homology of $`𝒞`$. The map $`CF_{}^+`$ (resp. $`HF_{}^+`$) induces a functor from $`\omega Cat_1`$ to $`Comp(Ab)`$ (resp. $`Ab`$).
The maps $`\mathrm{}_n^\pm `$ from $`𝒞_n`$ to $`C_n^\pm (𝒞)`$ induce a natural transformation from $`CF_{}^\pm `$ to $`CR_{}^\pm `$ and a natural transformation from $`HF_{}^\pm `$ to $`HR_{}^\pm `$.
###### Definition 9.3.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Set
* $`CF_0^{gl}(𝒞):=𝒞_0𝒞_0(𝒞_0\times 𝒞_0)`$
* $`CF_1^{gl}(𝒞):=𝒞_1`$
* $`CF_n^{gl}(𝒞)=𝒞_n/\{x_1y=x+y,\mathrm{},x_{n1}y=x+y\text{ mod }tr^{n1}𝒞\}`$ for $`n2`$
with the differential map $`s_{n1}t_{n1}`$ from $`CF_n^{gl}(𝒞)`$ to $`CF_{n1}^{gl}(𝒞)`$ for $`n2`$ and $`s_0t_0`$ from $`CF_1^{gl}(𝒞)`$ to $`CF_0^{gl}(𝒞)`$. This chain complex is called the formal globular complex. The associated homology is denoted by $`HF^{gl}(𝒞)`$ and is called the formal globular homology of $`𝒞`$.
By Theorem 7.6 and Corollary 3.6, we see that the globular folding operators induce a natural morphism of chain complex from $`CF_{}^{gl}`$ to $`CR_{}^{gl}`$, and therefore a natural transformation from $`HF_{}^{gl}`$ to $`HR_{}^{gl}`$.
###### Question 9.4.
When is the natural morphism of chain complexes $`R^{gl}`$ from $`CF_{}^{gl}(𝒞)`$ to $`CR_{}^{gl}(𝒞)`$ a quasi-isomorphism ?
###### Conjecture 9.5.
(About the thin elements of the globular complex) Let $`𝒞`$ be a globular $`\omega `$-category which is either the free globular $`\omega `$-category generated by a semi-cubical set or the free globular $`\omega `$-category generated by a globular set. Let $`x_i`$ be elements of $`C_n^{gl}(𝒞)`$ and let $`\lambda _i`$ be natural numbers, where $`i`$ runs over some set $`I`$. Suppose that for any $`i`$, $`ev(x_i)`$ is of dimension strictly lower than $`n`$ (one calls it a thin element). Then $`_i\lambda _ix_i`$ is a boundary if and only if it is a cycle.
The above conjecture is clear for $`C_2^{gl}`$ because all thin elements are degenerate. In higher dimension, there is enough room to have thin elements which are composition of degenerate elements, but which are not degenerate themselves.
The above conjecture is equivalent to claiming that the globular homology and the reduced one are equivalent for free globular $`\omega `$-categories generated by either a semi-cubical set or a globular set.
Now we are in position to give the exact statement relating the old globular homology of and the new one.
###### Definition 9.6.
Let $`(C_{}^{oldgl}(𝒞),^{oldgl})`$ be the chain complex defined as follows : $`C_0^{oldgl}(𝒞)=𝒞_0𝒞_0`$ and for $`n1`$, $`C_n^{oldgl}(𝒞)=𝒞_n`$, $`^{oldgl}(x)=(s_0x,t_0x)`$ if $`x𝒞_1`$ and for $`n1`$, $`x𝒞_{n+1}`$ implies $`^{oldgl}(x)=s_nxt_nx`$. This complex is called the old globular complex of $`𝒞`$ and its corresponding homology the old globular homology.
Instead of $`C_0^{oldgl}(𝒞)=𝒞_0𝒞_0`$, we set $`C_0^{oldgl}(𝒞)=(𝒞_0𝒞_0)`$ with the differential $`^{oldgl}(x)=s_0xt_0x`$ for $`x𝒞_1`$. This makes $`H_1^{oldgl}`$ slightly change. It does not matter because there is no influence on any potential applications. The difference appears in a situation like that of Figure 5. With $`C_0^{oldgl}(𝒞)=𝒞_0𝒞_0`$, $`u+xwv`$ is a old globular cycle. With $`C_0^{oldgl}(𝒞)=(𝒞_0𝒞_0)`$, this fake $`1`$-globular cycle is killed.
###### Theorem 9.7.
We have the following commutative diagram of natural transformations for $`0`$
where
* the map $`H_{}^{oldgl}H_{}^{gl}`$ is the canonical map induced by $`x\mathrm{}_n^{gl}(x)`$ from $`𝒞_n`$ to $`𝒩_{n1}^{gl}(𝒞)`$
* the map $`H_{}^{oldgl}HF_{}^{gl}`$ is the canonical map making all identifications like $`A_nB=A+B`$ for any $`n1`$ and any $`p`$-morphisms $`A`$ and $`B`$ with $`pn+1`$
* the map $`HF_{}^{gl}HF_{}^\pm `$ is the canonical map making the supplemental identification $`x=x_0y`$ or $`y=x_0y`$ depending on the sign $`\pm `$
* the map $`HF_{}^\pm HR_{}^\pm `$ is the canonical map induced by the folding operators $`\mathrm{}^\pm `$ of (which is likely to be an isomorphism for any strict globular $`\omega `$-category), and the map $`HF_{}^{gl}HR_{}^{gl}`$ is the canonical map induced by the folding operators $`\mathrm{}^{gl}`$ (which is also likely to be an isomorphism for any strict globular $`\omega `$-category)
* the maps $`R^{gl,\pm }`$ are the canonical maps from the globular or corner homology to the corresponding reduced homology (which are conjecturally an isomorphism for any free $`\omega `$-category generated by a semi-cubical set or a globular set).
###### Proof.
This is due to the fact that for $`n1`$, the natural map $`(h_n^\pm )^{old}`$ is induced by the set map $`\mathrm{}_n^{}`$ from $`𝒞_n`$ to $`\omega Cat(I^n,𝒞)^{}`$ ( Proposition 7.4). ∎
The difference between $`H_0^{oldgl}`$ and $`H_0^{gl}`$ is also not important. The group $`H_0^{oldgl}`$ was indeed only introduced to define the morphisms $`h^{}`$ and $`h^+`$ in dimension $`0`$. But $`H_0^{oldgl}`$ does not have any computer-scientific meaning and is not involved in any potential applications.
## 10 Globular homology and deformation of HDA
The following table summarizes how the globular nerve may be understood and compared with the two corner nerves of $`𝒞`$.
| Geometric object | Formal theory | “True” theory | Simplicial cut |
| --- | --- | --- | --- |
| Branching | formal negative corner homology | negative corner homology | $`𝒩^{}(𝒞)`$ |
| Merging | formal positive corner homology | positive corner homology | $`𝒩^+(𝒞)`$ |
| Globe | formal globular homology | globular homology | $`𝒩^{gl}(𝒞)`$ |
Intuitively, the globular nerve of $`𝒞`$ contains all achronal cuts in the middle of all globes, whereas the negative and positive corner simplicial nerves contain all achronal cuts close to respectively the negative and the positive corners of the automaton. The expression “achronal” is borrowed from and . In these papers, HDA are modeled by local pospaces, and an achronal subspace $`Y`$ of a local pospace is a topological subspace such that $`xy`$ and $`x,yY`$ imply $`x=y`$. The remarkable point is that the set of all achronal cuts of a given type can be enclosed into a simplicial set.
This could mean that the whole geometry of the free $`\omega `$-category $`𝒞`$ generated by a semi-cubical set (i.e. a HDA) would be contained in the following diagram of augmented simplicial sets
and in its temporal graph $`tr^1𝒞`$. This latter contains the information about the temporal structure of the HDA.
A problem, already mentioned in , is the question of the invariance of the globular homology of an $`\omega `$-category up to a choice of a cubification <sup>2</sup><sup>2</sup>2Some authors use the term cubicalation : this means decomposing a HDA in cubes. of the corresponding HDA. There are two types of deformations : the spatial deformations or S-deformations and the temporal deformations or T-deformations.
The globular cut is invariant by S-deformation, that is by deformations of $`p`$-morphisms with $`p2`$. This is simply due to the fact that such a deformation corresponds in the globular cut to a deformation of any simplex containing it as label. Therefore such a deformation corresponds to a deformation up to homotopy, in the usual sense, of the globular cut.
Unlike the corner homologies, the globular homology turns indeed to depend on the subdivision of time. The reason is contained in Figure 6. The obvious $`1`$-functor from the left to the right such that $`uu_1_0u_2`$ should leave the globular homology invariant. This is not the case because the first globular homology is for the left member the free $``$-module generated by $`vw`$ and $`u_0vu_0w`$, and for the right member the free $``$-module generated by $`vw`$ and $`u_2_0vu_2_0w`$ and $`u_1_0u_2_0vu_1_0u_2_0w`$. However in Figure 6, one can subdivide as many times as one wants for example $`v`$, and the globular homology will not change.
One way to overcome this problem is exposed in the last sections of , devoted to the description of a generic way to produce T-invariants starting from the globular nerve. Let us prove Claim 5.1 which enables to introduce the bisimplicial set mentioned in that paper.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Using Theorem 8.3, recall that for some $`\omega `$-functor $`x`$ from $`\mathrm{\Delta }^n`$ to $`𝒞`$, one calls $`S(x)`$ the unique element of the image of $`s_0x`$ and $`T(x)`$ the unique element of the image of $`t_0x`$. If $`(\alpha ,\beta )`$ is a pair of $`𝒩_1^{gl}(𝒞)`$, set $`S(\alpha ,\beta )=\alpha `$ and $`T(\alpha ,\beta )=\beta `$.
###### Proposition 10.1.
Let $`𝒞`$ be a non-contracting $`\omega `$-category. Let $`x`$ and $`y`$ be two $`\omega `$-functors from $`\mathrm{\Delta }^n`$ to $`𝒞`$ with $`n0`$. Suppose that $`T(x)=S(y)`$. Let $`xy`$ be the map from the faces of $`\mathrm{\Delta }^n`$ to $`𝒞`$ defined by
$$(xy)((\sigma _0\mathrm{}\sigma _r)):=x((\sigma _0\mathrm{}\sigma _r))_0y((\sigma _0\mathrm{}\sigma _r)).$$
Then the following conditions are equivalent :
1. The image of $`xy`$ is a subset of $`𝒞`$.
2. The set map $`xy`$ yields an $`\omega `$-functor from $`\mathrm{\Delta }^n`$ to $`𝒞`$ and $`_i(xy)=_i(x)_i(y)`$ for any $`0in`$.
On contrary, if for some $`(\sigma _0\mathrm{}\sigma _r)\mathrm{\Delta }^n`$, $`(xy)((\sigma _0\mathrm{}\sigma _r))`$ is $`0`$-dimensional, then $`xy`$ is the constant map $`S(x)=T(y)`$.
###### Proof.
We have to prove that Condition 1 implies Condition 2. Let us consider $`P(n)`$ : “for any non-contracting $`\omega `$-category $`𝒞`$ and any $`\omega `$-functor $`x`$ and $`y`$ from $`\mathrm{\Delta }^n`$ to $`𝒞`$ such that $`T(x)=S(y)`$ and such that the image of $`xy`$ is a subset of $`𝒞`$, then $`xy`$ yields an $`\omega `$-functor from $`\mathrm{\Delta }^n`$ to $`𝒞`$ and $`_i(xy)=_i(x)_i(y)`$ for any $`0in`$.”
Property $`P(0)`$ is obvious. Suppose $`P(n1)`$ proved for $`n1`$. For any $`0in`$, $`_i(x)_i(y)`$ is a set map from $`\mathrm{\Delta }^{n1}`$ to $`𝒞`$ satisfying the hypothesis of the proposition, so by induction hypothesis, $`_i(x)_i(y)`$ yields an $`\omega `$-functor from $`\mathrm{\Delta }^{n1}`$ to $`𝒞`$. Let $`z_i:=_i(x)_i(y)`$. For $`0j<in`$,
$`_j(z_i)`$ $`=(_j_i(x))(_j_i(y))`$ by induction hypothesis
$`=(_{i1}_j(x))(_{i1}_j(y))`$
$`=_{i1}(_j(x)_j(y))`$ by induction hypothesis
$`=_{i1}z_j`$
Therefore $`(z_i)_{0in}`$ is an $`(n1)`$-shell. So it provides a unique $`\omega `$-functor
$$z:\mathrm{\Delta }^n\backslash \{(01\mathrm{}n)\}𝒞$$
by Proposition 6.9. It remains to check that
$$z\left(s_{n1}R((01\mathrm{}n))\right)=s_n((xy)((01\mathrm{}n)))$$
and
$$z\left(t_{n1}R((01\mathrm{}n))\right)=t_n((xy)((01\mathrm{}n)))$$
to complete the proof. Let us check the first equality. One has
$$s_{n1}R((01\mathrm{}n))=\mathrm{\Psi }(X_1,\mathrm{},X_s)$$
where $`\mathrm{\Psi }`$ uses only composition laws and where $`X_1,\mathrm{},X_s`$ are faces of $`\mathrm{\Delta }^n`$ of dimension at most $`n1`$. Denote by $`\mathrm{\Psi }^{}`$ the same function as $`\mathrm{\Psi }`$ with $`_i`$ replaced by $`_{i+1}`$. Then
$`z\left(s_{n1}R((01\mathrm{}n))\right)`$
$`=z\mathrm{\Psi }(X_1,\mathrm{},X_s)`$
$`=\mathrm{\Psi }^{}(z(X_1),\mathrm{},z(X_s))`$ since $`z`$ $`\omega `$-functor
$`=\mathrm{\Psi }^{}(x(X_1)_0y(X_1),\mathrm{},x(X_s)_0y(X_s))`$ by definition of $`z`$
$`=\mathrm{\Psi }^{}(x(X_1),\mathrm{},x(X_s))_0\mathrm{\Psi }^{}(y(X_1),\mathrm{},y(X_s))`$ by interchange law
$`=\left(x\mathrm{\Psi }(X_1,\mathrm{},X_s)\right)_0\left(y\mathrm{\Psi }(X_1,\mathrm{},X_s)\right)`$ since $`x`$ and $`y`$ $`\omega `$-functors
$`=\left(xs_{n1}R((01\mathrm{}n))\right)_0\left(ys_{n1}R((01\mathrm{}n))\right)`$
$`=\left(s_nxR((01\mathrm{}n))\right)_0\left(s_nyR((01\mathrm{}n))\right)`$ since $`x`$ and $`y`$ $`\omega `$-functors
$`=s_n\left(xR((01\mathrm{}n))_0yR((01\mathrm{}n))\right)`$ by interchange law
$`=s_n((xy)((01\mathrm{}n)))`$
Now let us suppose that $`(xy)((\sigma _0\mathrm{}\sigma _r))`$ is $`0`$-dimensional in $`𝒞`$ for some $`(\sigma _0\mathrm{}\sigma _r)`$. Then
$$s_1x((\sigma _0\mathrm{}\sigma _r))_0s_1y((\sigma _0\mathrm{}\sigma _r))$$
is $`0`$-dimensional. Either $`s_0(\sigma _0\mathrm{}\sigma _r)=(n)`$ (the initial state of $`\mathrm{\Delta }^n`$) or there exists a $`1`$-morphism $`U`$ of $`\mathrm{\Delta }^n`$ such that $`s_0U=(n)`$ and $`t_0U=s_0(\sigma _0\mathrm{}\sigma _r)`$. In the first case, $`x((n))_0y((n))`$ is $`0`$-dimensional. In the second case,
$$x(t_0U)_0y(t_0U)=t_1x(U)_0t_1y(U)=t_1\left(x(U)_0y(U)\right)$$
is $`0`$-dimensional. Then $`x(U)_0y(U)`$ is $`0`$-dimensional as well as
$$x((n))_0y((n))=s_1\left(x(U)_0y(U)\right).$$
For any face $`(\tau _0\mathrm{}\tau _r)`$ of $`\mathrm{\Delta }^n\backslash \{(n)\}`$, there exists a $`1`$-morphism $`V`$ from $`((n))`$ to $`s_0(\tau _0\mathrm{}\tau _r)`$ or $`t_0(\tau _0\mathrm{}\tau _r)`$ : let us say $`s_0(\tau _0\mathrm{}\tau _r)`$. Since
$$s_1(xy)(V)=(xy)((n))$$
is $`0`$-dimensional, then $`(xy)(V)`$ is $`0`$-dimensional, as well as
$$t_1(xy)(V)=(xy)(s_0(\tau _0\mathrm{}\tau _r))=s_1(xy)((\tau _0\mathrm{}\tau _r)).$$
Therefore $`(xy)((\tau _0\mathrm{}\tau _r))`$ is $`0`$-dimensional. ∎
In the sequel, we set $`(\alpha ,\beta )(\beta ,\gamma )=(\alpha ,\gamma )`$, $`S(\alpha ,\beta )=\alpha `$ and $`T(\alpha ,\beta )=\beta `$. If $`x`$ is an $`\omega `$-functor from $`\mathrm{\Delta }^n`$ to $`𝒞`$, and if $`y`$ is the constant map $`T(x)`$ (resp. $`S(x)`$) from $`\mathrm{\Delta }^n`$ to $`𝒞_0`$, then set $`xy:=x`$ (resp. $`yx:=x`$).
###### Theorem 10.2.
Suppose that $`𝒞`$ is an object of $`\omega Cat_1`$. Then for $`n0`$, the operations $`S`$, $`T`$ and $``$ allow to define a small category $`\underset{¯}{𝒩_n^{gl}(𝒞)}`$ whose morphisms are the elements of $`𝒩_n^{gl}(𝒞)\{\text{constant maps }\mathrm{\Delta }^n𝒞_0\}`$ and whose objects are the $`0`$-morphisms of $`𝒞`$. If $`\underset{¯}{𝒩_1^{gl}(𝒞)}`$ is the small category whose morphisms are the elements of $`𝒞_0\times 𝒞_0`$ and whose objects are the elements of $`𝒞_0`$ with the operations $`S`$, $`T`$ and $``$ above defined, then one obtains (by defining the face maps $`_i`$ and degeneracy maps $`ϵ_i`$ in an obvious way on $`\{\text{constant maps }\mathrm{\Delta }^n𝒞_0\}`$) an augmented simplicial object $`\underset{¯}{𝒩_{}^{gl}}`$ in the category of small categories.
###### Proof.
Equalities $`S(x)=_iS(x)`$, $`S(x)=ϵ_iS(x)`$, $`T(x)=_iT(x)`$, $`T(x)=ϵ_iT(x)`$ are consequences of Proposition 8.3. Equality $`_i(xy)=_ix_iy`$ is proved right above. The verification of $`ϵ_i(xy)=ϵ_ixϵ_iy`$ is straightforward. ∎
By composing by the classifying space functor of small categories (cf. for example for further details), one obtains a bisimplicial set which seems to be well-behaved with respect to subdivision of time. Indeed the first total homology groups associated to both $`\omega `$-categories of Figure 6 are equal to $``$. Further explanations will be given in future papers.
To conclude, let us point out that in reasonable cases, i.e. when the $`p`$-morphisms (with $`p2`$) of a non-contracting $`\omega `$-category $`𝒞`$ are invertible with respect to the composition laws $`_i`$ of $`𝒞`$ for $`i1`$, then $`𝒞`$ becomes a globular $`\omega `$-groupoid in the sense of Brown-Higgins. And therefore in such a case, it is well-known that the globular nerve of $`𝒞`$ satisfies the Kan property (see or a generalization in ). However, this is not true in general for both corner nerves. To understand this fact, consider the $`2`$-source of $`R(000)`$ in Figure 2(c) and remove $`R(0+0)`$. Consider both inclusion $`\omega `$-functors from $`I^2`$ to respectively $`R(00)`$ and $`R(00)`$. Then the Kan condition fails because one cannot make the sum of $`R(00)`$ and $`R(00)`$ since $`R(0+0)`$ is removed.
Institut de Recherche Mathématique Avancée
ULP et CNRS
7 rue René Descartes
67084 Strasbourg Cedex
France
gaucher@irma.u-strasbg.fr
|
warning/0002/cond-mat0002093.html
|
ar5iv
|
text
|
# Statistical mechanics approach to the phase unwrapping problem
## 1 Introduction
The determination of the absolute phase from a fringe pattern is an important problem that finds applications in many areas: homomorphic signal processing , solid state physics , holographic interferometry , adaptive or compensated optics , magnetic resonance imaging and synthetic aperture radar interferometry .
In all these applications one obtains a two-dimensional fringe pattern whose spatially-varying phase is related to the physical quantity to be measured. The computation of phase by any inverse trigonometric function (e.g. arctangent) provides only principal phase values, which lie between $`\pm \pi `$ radians. The process of phase unwrapping (PU), i.e. the addition of a proper integer multiple of $`2\pi `$ to all the pixels, must be carried out before the physical quantity can be reconstructed from the phase distributions. Since many possible absolute phase fields are compatible with a given fringe pattern, phase unwrapping is ill-posed in a mathematical sense: Hadamard defined a mathematical problem to be well-posed if a unique solution exists that depends continuously on the data; in this case, the uniqueness requirement is violated.
Ill-posed problems arise frequently in many areas of science and engineering; well-known examples are analytic continuation, the Cauchy problem for differential equations, computer tomography, and many problems in image processing and machine vision that involve the reconstruction of images from noisy data.
The fact that a problem is not well-posed does not mean that it cannot be solved: rather, in order to be solved it must be first *regularized* by introducing additional constraints (prior knowledge) about the behaviour of the solution. Variational regularization corresponds to modeling the physical constraints of the problem by a suitable functional; the solution is then sought as the minimizer of this functional.
In the last years, an increasing interest has been devoted to adapt methods from Statistical Mechanics to nonconvex optimization problems arising from the variational regularization of ill-posed problems. Geman and Geman suggested that the Ising model is applicable to image restoration through the Bayesian formalism. This problem corresponds to searching the ground state of a finite-size Ising model under a non-uniform external field. Geman and Geman applied this to the recovery of corrupted images by using simulated annealing of a spin-S Ising model. After that, Gidas proposed a new method based on a combination of the renormalization group technique and the simulated annealing procedure; then, Zhang introduced Mean-Field Annealing to treat the image reconstruction problem, while Tanaka and Morita applied the cluster variation method. Methods of statistical mechanics have also been used to study combinatorial optimization problems (see, e.g., and references therein).
In a couple of recent papers, the phase unwrapping problem was handled by methods of Statistical Mechanics. Simulated annealing was applied in . In the problem is solved by the Mean-Field Annealing (MFA) technique; PU is formulated as a constrained optimization problem for the field of integer corrections to be added to the wrapped phase gradient in order to recover the true phase gradient, with the cost function consisting of second order differences, and measuring the smoothness of the reconstructed phase field. This is equivalent to finding the ground state of a locally-constrained ferromagnetic spin-1 Ising model under a non-uniform external field. The optimization problem is then solved by MFA and consistent solutions are found in difficult situations resulting from noise and undersampling.
Mean Field Annealing is closely related to Simulated Annealing . Both approaches formulate the optimization problem in terms of minimizing a cost function and defining a corresponding Gibbs distribution. Simulated Annealing then proceeds by sampling the Gibbs probability distribution as the temperature is reduced to zero, whereas MFA attempts to track an approximation of the mean of the same distribution.
The algorithm described in was constructed under the assumption that the possible values for the correction field were restricted to belong to the set $`\{1,0,1\}`$. In this paper we generalize the MFA algorithm to the case of a set $`\{L,\mathrm{},L\}`$, with $`L`$ an arbitrary integer. The corresponding statistical system is a locally-constrained spin-L Ising model.
The present generalization allows the processing of input phase patterns with arbitrary degree of undersampling; our experiments on synthetic phase fields show the effectiveness of the proposed algorithm.
The paper is organized as follows. In sect. 2 the phase unwrapping problem is introduced and its ill-posedness is highlighted. In sect. 3 the deterministic MFA algorithm is described in detail. Then, in sect. 4 some experimental results on simulated phase fields are presented. Some conclusions are then drawn in sect. 5.
## 2 Phase Unwrapping Problem
We briefly recall here the phase unwrapping terminology, and refer the reader to for a complete discussion.
Given an absolute phase pattern $`f(x,y)`$ on a two-dimensional square grid, what is actually measured is the wrapped phase field $`g(x,y)`$ which can be expressed in terms of the $`f`$ field through a wrapping operator, Wr, defined so that $`g(x,y)`$ always lies in the interval $`[\pi ,+\pi )`$:
$$g(x,y)=\mathrm{Wr}[f(x,y)]=\mathrm{arg}\left\{\mathrm{exp}[\mathrm{i}f(x,y)]\right\}.$$
(1)
Phase unwrapping means recovering the absolute phase field $`f`$, which is usually related to the physical quantity to be measured, from the knowledge of the $`g`$ field. This can be done in practice by estimating the absolute phase gradient from the wrapped phase field and integrating it throughout the 2-D grid. This simple method is effective only in absence of phase aliasing, i.e. if the phase field is correctly sampled.
In fact, if the Nyquist condition:
$$\left|f(x,y)\right|<\pi ,$$
(2)
where $``$ is the discrete gradient, is verified everywhere on the grid, the absolute phase gradient is obtained by wrapping the gradient of the wrapped phase field, according to the formula:
$$𝐀(x,y)=\mathrm{Wr}[g(x,y)].$$
(3)
As mentioned, if condition (2) is satisfied, one has:
$$f(x,y)=𝐀(x,y),$$
(4)
and the $`f`$-pattern is obtained by integrating $`𝐀`$ along any path connecting all sites on the grid.
The Nyquist condition is often violated because of undersampling of the signal from which the principal phase is extracted. This can result either from system noise, or from critical values of the slopes of the physical surface which is analyzed through interferometry. For example, in the case of SAR interferometry, the surface is the portion of Earth imaged from the sensor (usually air- or satellite-borne), while noise can arise from sensor thermal electronic motion, or from other sources of electronic signal disturbances.
If the Nyquist condition is not satisfied everywhere on the grid, then the wrapped gradient $`𝐀`$ of the wrapped phase field is not assured to equal the absolute phase gradient. In this case, a more general relation must be written, rather than (4), i.e.:
$$f(x,y)=𝐀(x,y)+2\pi 𝐤(x,y),$$
(5)
where $`𝐤(x,y)`$ is a vector field of integers. In this case, solving PU amounts to finding the correct field $`𝐤`$.
Phase aliasing conditions imply that the integration of field $`𝐀`$ depends on the path. The sources of this nonconservative behaviour are detectable by calculating the integral of the field $`𝐀`$ over every minimum closed path, i.e. the 2$`\times `$2 square having the site $`(x,y)`$ as a corner:
$$I(x,y)=\frac{1}{2\pi }[A_x(x,y)+A(y(x+1,y)A_x(x,y+1)A_y(x,y)].$$
(6)
One can show that the integral $`I(x,y)`$ will always have a value in the set $`\{1,0,1\}`$. Locations with $`I0`$ are called “residues”. In presence of residues, the field $`𝐀`$ is no more irrotational; this causes the path-dependence of the integration step previously described.
To restore the consistency of the phase gradient, then, the $`𝐤`$ field must satisfy the following consistency condition:
$$\times \left[𝐀(x,y)+2\pi 𝐤(x,y)\right]=0,$$
(7)
where $`\times `$ is the discrete curl operator. Since there are many possible $`𝐤`$ fields satisfying eq. (7), PU is an ill-posed problem according to Hadamard’s definition.
One of the most classical and widely-used algorithms for phase unwrapping is the Least Mean Squares (LMS) approach , which consists in finding the scalar field $`f`$ whose gradient is closer to $`𝐀`$ in the Least Squares sense, i.e. the minimizer of:
$$(f𝐀)^2.$$
(8)
As mentioned before, in a variational approach has been used, and the field $`𝐤`$ was “chosen” as the minimizer of the following functional:
$`R`$ $`=`$ $`{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \left[_xf(x+1,y)_xf(x,y)\right]^2}`$ (9)
$`+{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \left[_yf(x,y+1)_yf(x,y)\right]^2}`$
$`+{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \left[_xf(x,y+1)_xf(x,y)\right]^2}`$
$`+{\displaystyle \frac{1}{4\pi ^2}}{\displaystyle \left[_yf(x+1,y)_yf(x,y)\right]^2},`$
subject to constraint (7). Due to (5), $`R`$ is a functional of $`𝐤`$, i.e. $`R=R[𝐤]`$, and it measures the smoothness of the reconstructed phase surface. The optimization problem was then solved by a MFA algorithm under the assumption that the $`𝐤`$ field be restricted to take values in $`\{1,0,1\}`$. In the next section we generalize the MFA algorithm to the case of $`𝐤`$ fields belonging to $`\{L,\mathrm{},L\}`$, with $`L`$ an arbitrary integer.
## 3 The algorithm
As explained in sect. 2, we assume the solution of PU to be the minimizer of the functional (9), subject to constraint (7). Let us assume that the possible values of the $`𝐤`$ field are restricted to belong to $`\{L,\mathrm{},L\}`$. The field $`𝐤`$ may then be regarded as a system of spin-L units. We assume the Gibbs distribution:
$$P[𝐤]=\frac{\mathrm{exp}\left[\frac{R[𝐤]}{T}\right]}{_𝐤^{}\mathrm{exp}\left[\frac{R[𝐤^{}]}{T}\right]},$$
(10)
where the sum is over the $`𝐤^{}`$ fields satisfying (7), and $`T`$ is the statistical temperature. Inconsistent fields $`𝐤^{}`$ are assumed to have zero probability.
Following Mean-Field theory , we consider a probability distribution for the correction field $`𝐤`$ which treats all the variables as independent, i.e. it is the product of the marginal distributions of each variable.
Let $`\rho _x(x,y,\alpha )`$ be the probability that $`k_x(x,y)=\alpha `$, with $`\alpha =L,\mathrm{},L`$, and $`\rho _y(x,y,\alpha )`$ be the corresponding probability for $`k_y(x,y)`$. Normalization of these marginal probabilities implies a penalty functional:
$$\mathrm{\Theta }[\rho ]=\underset{(x,y)}{}\left[V_x(x,y)\left(1\underset{\alpha }{}\rho _x(x,y,\alpha )\right)+V_y(x,y)\left(1\underset{\alpha }{}\rho _y(x,y,\alpha )\right)\right],$$
(11)
where $`\{V\}`$ are Lagrange multipliers.
The entropy of the system, in the mean field approximation, is:
$$S[\rho ]=\underset{(x,y)}{}\underset{\alpha =L}{\overset{L}{}}\left[\rho _x(x,y,\alpha )\mathrm{log}\rho _x(x,y,\alpha )+\rho _y(x,y,\alpha )\mathrm{log}\rho _y(x,y,\alpha )\right].$$
(12)
It is useful to introduce the average fields $`𝐦=𝐤_\rho `$ and $`𝐐=𝐤^2_\rho `$, defined by:
$`m_x(x,y)={\displaystyle \underset{\alpha =L}{\overset{L}{}}}\alpha \rho _x(x,y,\alpha ),m_y(x,y)={\displaystyle \underset{\alpha =L}{\overset{L}{}}}\alpha \rho _y(x,y,\alpha );`$ (13)
$`Q_x(x,y)={\displaystyle \underset{\alpha =L}{\overset{L}{}}}\alpha ^2\rho _x(x,y,\alpha ),Q_y(x,y)={\displaystyle \underset{\alpha =L}{\overset{L}{}}}\alpha ^2\rho _y(x,y,\alpha ).`$ (14)
The average $`U`$ of the cost functional $`R`$ is called *internal energy*. It is easy to show that the internal energy depends only on $`𝐦`$ and $`𝐐`$:
$$U[𝐦,𝐐]=R[𝐀+2\pi 𝐤]_\rho .$$
(15)
A penalty functional is introduced to enforce constraints (7):
$`\mathrm{\Gamma }[𝐦]`$ $`={\displaystyle \underset{(x,y)}{}}\lambda (x,y)[m_x(x,y)+m_y(x+1,y)`$ (16)
$`m_x(x,y+1)m_y(x,y)+I(x,y)],`$
where $`\{\lambda \}`$ is another set of Lagrange multipliers.
Let us now introduce an effective cost functional, the *free energy*, which depends on $`T`$:
$$F[\rho ]=U[𝐦,𝐐]TS[\rho ]+\mathrm{\Gamma }[𝐦]+\mathrm{\Theta }[\rho ]$$
(17)
The free energy is the weighted sum of the internal energy (the original cost function) and the entropy functional; $`\mathrm{\Gamma }`$ and $`\mathrm{\Theta }`$ are penalty functionals to enforce the constraints of the problem. According to the variational principle of Statistical Mechanics, the best approximation to the Gibbs distribution is the minimizer of the free energy . Since $`TS`$ is a convex functional, the free energy is convex at high temperature and the global minimum can be easily attained. The solution can then be continuated as temperature is lowered, so as to reach a minimum of $`U`$. This procedure has shown to be less sensitive to local minima than conventional descent methods, and gives results close to the ones from Simulated Annealing, while requiring less computational time .
The equations for the minimum of the free energy are usually called “mean-field equations”:
$$\frac{F}{\rho _x(x,y,\alpha )}=0;\frac{F}{\rho _y(x,y,\alpha )}=0$$
(18)
After simple calculations, the solution of eqs. (18) is found to have the following form:
$`\rho _x(x,y,\alpha )`$ $`={\displaystyle \frac{\mathrm{exp}\left\{\beta \left[\frac{U}{\rho _x(x,y,\alpha )}+\frac{\mathrm{\Gamma }}{\rho _x(x,y,\alpha )}\right]\right\}}{_{\alpha ^{}=L}^L\mathrm{exp}\left\{\beta \left[\frac{U}{\rho _x(x,y,\alpha ^{})}+\frac{\mathrm{\Gamma }}{\rho _x(x,y,\alpha ^{})}\right]\right\}}},`$ (19)
$`\rho _y(x,y,\alpha )`$ $`={\displaystyle \frac{\mathrm{exp}\left\{\beta \left[\frac{U}{\rho _y(x,y,\alpha )}+\frac{\mathrm{\Gamma }}{\rho _y(x,y,\alpha )}\right]\right\}}{_{\alpha ^{}=L}^L\mathrm{exp}\left\{\beta \left[\frac{U}{\rho _y(x,y,\alpha ^{})}+\frac{\mathrm{\Gamma }}{\rho _y(x,y,\alpha ^{})}\right]\right\}}},`$ (20)
where the $`\{V\}`$ multipliers have been fixed to normalize the distributions, and $`\beta =1/T`$ is the inverse temperature. Now we observe that, for each site $`(x,y)`$ on the grid:
$$\frac{U}{\rho _x(\alpha )}=\frac{U}{m_x}\frac{m_x}{\rho _x(\alpha )}+\frac{U}{Q_x}\frac{Q_x}{\rho _x(\alpha )}=\alpha \frac{U}{m_x(x,y)}+\alpha ^2\frac{U}{Q_x(x,y)}.$$
(21)
Analogously, one can easily find:
$`{\displaystyle \frac{U}{\rho _y(\alpha )}}`$ $`=`$ $`\alpha {\displaystyle \frac{U}{m_y(x,y)}}+\alpha ^2{\displaystyle \frac{U}{Q_y(x,y)}},`$ (22)
$`{\displaystyle \frac{\mathrm{\Gamma }}{\rho _x(\alpha )}}`$ $`=`$ $`\alpha \left[\lambda (x,y)\lambda (x,y1)\right],`$ (23)
$`{\displaystyle \frac{\mathrm{\Gamma }}{\rho _y(\alpha )}}`$ $`=`$ $`\alpha \left[\lambda (x,y)+\lambda (x1,y)\right].`$ (24)
The derivatives of $`U`$ with respect to the $`\{m\}`$ and $`\{Q\}`$ variables are reported in Appendix A. From these expressions it is clear that the present formulation of PU is equivalent to finding the ground state of a finite-size, spin-L Ising model with local constraints, and under a non-uniform magnetic field.
The consistency constraints are written as equations for the $`\{\lambda \}`$ field:
$`\lambda (x,y)`$ $`=\lambda (x,y)b[m_x(x,y)+m_y(x+1,y)`$ (25)
$`m_x(x,y+1)m_y(x,y)+I(x,y)],`$
where $`b`$ is a small constant.
Equations (1920) and (25) are the mean-field equations for PU for arbitrary $`L`$. As already explained, the MFA technique consists in solving iteratively the mean-field equations at high temperature (low $`\beta `$), and then track the solution as the temperature is lowered ($`\beta `$ grows).
The algorithm can be summarized as follows. The initial distributions give the same probability to each value of the correction field, i.e. $`\rho _x(x,y,\alpha )=\rho _y(x,y,\alpha )=\frac{1}{2L+1}`$; the inverse temperature is set to $`\beta _{\mathrm{MIN}}`$ ($`\beta _{\mathrm{MAX}}`$ is the lowest temperature). Then:
1. Evaluate $`\{𝐦\}`$ and $`\{𝐐\}`$ fields by Eqs. (1314);
2. Iterate Eqs. (1920);
3. Iterate Eq. (25);
4. If Eqs. (1920) or Eq. (25) are not satisfied, goto step 1;
5. If $`\beta <\beta _{\mathrm{MAX}}`$, increase $`\beta `$ and goto step 1.
The output of this algorithm is a field $`\{𝐦_{\mathrm{OUT}}\}`$ which approximates the average of $`\{𝐤\}`$ over the global minima of the cost functional $`R`$.
We remark that the output of the algorithm described in satisfies $`m[1,1]`$ for each component of $`\{𝐦_{\mathrm{OUT}}\}`$, whereas the present algorithm satisfies the weaker constraint $`m[L,L]`$ and therefore can be used also in the case of high degree of undersampling. The estimate for the true phase gradient is $`(f)_{\mathrm{est}}=𝐀+2\pi 𝐦_{\mathrm{OUT}}`$; the phase pattern $`f`$ can then be reconstructed by $`(f)_{\mathrm{est}}`$ as described in .
## 4 Experiments
In this section we describe some experiments we performed to test the effectiveness of the proposed algorithm.
In fig. 1-(a) a synthetic phase pattern is shown, while in fig. 1-(b) the wrapped phase pattern is depicted. This test phase pattern has been constructed by the following formula:
$$\mathrm{\Phi }(x,y)=120\mathrm{exp}\left[\frac{1}{2}r^2(x,y)(\mu _1+\mu _2c(x,y))\right],1x,y128$$
(26)
with:
$`r(x,y)`$ $`=`$ $`\sqrt{(x35.5)^2+(y65.5)^2},`$
$`c(x,y)`$ $`=`$ $`{\displaystyle \frac{x35.5}{r}},`$
$`\mu _1`$ $`=`$ $`0.01,`$
$`\mu _2`$ $`=`$ $`0.0004,`$
where $`\mathrm{\Phi }`$ is in radians.
By construction, $`\mathrm{\Phi }`$ is undersampled: in fig. 2-(a) black pixels represent locations where the true correction field $`𝐤`$ is such that $`\mathrm{max}\{k_x,k_y\}=2`$, while gray pixels represent locations where $`\mathrm{max}\{k_x,k_y\}=1`$. The residue map is depicted in fig. 2-(b).
We applied the proposed algorithm to unwrap this test pattern. We used $`L=2,b=0.05`$ and the annealing schedule was established as consisting of 25 temperature values, equally spaced in the interval $`[\beta _{\mathrm{MIN}}=0.05,\beta _{\mathrm{MAX}}=1.5]`$. The output phase pattern is depicted in fig. 3. The computational time was comparable to that corresponding to . The input phase surface was perfectly reconstructed. Let us now compare the performance of the MFA algorithm with that from LMS . In fig. 4 we show the output of LMS applied to the surface of fig. 1-(a). Due to severe undersampling, the LMS performance is poor.
We also investigated the robustness of the MFA algorithm with respect to noise. In fig. 5-(a) the phase pattern obtained by adding unit-variance Gaussian noise to the surface of fig. 1-(a) is shown. The wrapped phase pattern is depicted in fig. 5-(b), while in fig. 5-(c) the inconsistencies are shown. The output of the MFA algorithm is shown in fig. 6-(a), while in fig. 6-(b) we show the phase pattern obtained by re-wrapping the MFA output. The smoothing capability of the proposed algorithm appears clearly by comparing figs. 5-(b) and 6-(b).
## 5 Conclusions
In this paper we have generalized a previously presented MFA algorithm to unwrap phase patterns. This problem is formulated as that of finding the ground state of a locally-constrained, spin-L Ising model under a non-uniform external field. The present generalization allows processing of noisy and highly undersampled input phase fields. The effectiveness of this statistical approach to PU has been demonstrated on simulated phase surfaces. Further work will be devoted to the estimation of the optimal value of $`L`$ from the observed wrapped phase data.
## Appendix A Appendix: Derivatives of the internal energy
We report here the expressions of the derivatives of the internal energy in the Mean-field approximation.
One easily finds that:
$`{\displaystyle \frac{U}{Q_x(x,y)}}=4`$
$`{\displaystyle \frac{U}{Q_y(x,y)}}=4,`$ (27)
$`{\displaystyle \frac{U}{m_x(x,y)}}`$ $`=`$ $`2m_x(x1,y)+{\displaystyle \frac{1}{\pi }}\left[A_x(x,y)A_x(x1,y)\right]+`$ (28)
$`2m_x(x+1,y)+{\displaystyle \frac{1}{\pi }}\left[A_x(x,y)A_x(x+1,y)\right]+`$
$`2m_x(x,y1)+{\displaystyle \frac{1}{\pi }}\left[A_x(x,y)A_x(x,y1)\right]+`$
$`2m_x(x,y+1)+{\displaystyle \frac{1}{\pi }}\left[A_x(x,y)A_x(x,y+1)\right].`$
$`{\displaystyle \frac{U}{m_y(x,y)}}`$ $`=`$ $`2m_y(x,y1)+{\displaystyle \frac{1}{\pi }}\left[A_y(x,y)A_y(x,y1)\right]+`$ (29)
$`2m_y(x,y+1)+{\displaystyle \frac{1}{\pi }}\left[A_y(x,y)A_y(x,y+1)\right]+`$
$`2m_y(x+1,y)+{\displaystyle \frac{1}{\pi }}\left[A_y(x,y)A_y(x+1,y)\right]+`$
$`2m_y(x1,y)+{\displaystyle \frac{1}{\pi }}\left[A_y(x,y)A_y(x1,y)\right].`$
The authors thank Dr. G. Gonnella for useful discussions on Mean-Field theory.
|
warning/0002/cond-mat0002043.html
|
ar5iv
|
text
|
# Probing Pseudogap by Josephson Tunneling
0pt0.4pt 0pt0.4pt 0pt0.4pt
## Abstract
We propose here an experiment aimed to determine whether there are superconducting pairing fluctuations in the pseudogap regime of the high-$`T_c`$ materials. In the experimental setup, two samples above $`T_c`$ are brought into contact at a single point and the differential AC conductivity in the presence of a constant applied bias voltage between the samples, $`V`$, should be measured. We argue the the pairing fluctuations will produce randomly fluctuating Josephson current with zero mean, however the current-current correlator will have a characteristic frequency given by Josephson frequency $`\omega _J=2eV/\mathrm{}`$. We predict that the differential AC conductivity should have a peak at the Josephson frequency with the width determined by the phase fluctuations time.
One of the long-standing puzzles of the high-temperature superconductivity is the nature of the so-called pseudogap regime. The pseudogap regime occurs in the wide range of temperatures above superconducting transition temperature in the underdoped cuprate superconductors. It is characterized by the suppressed quasiparticle density of states in the vicinity of the Fermi level. The similarity of the density of states in the pseudogap regime and in the superconducting state lead many to believe that that the pseudogap itself is of a superconducting origin. In this view, the long range superconducting order in the pseudogap regime is destroyed by phase fluctuations. However, locally, both electronic pairing and fluctuating regions of superconductivity should persist. Therefore, in this picture, the superconducting phase transition is believed to be a superconducting phase-ordering transition. Moreover, recent experiments by Orenstein and collaborators claim that local superfluid density is present even above $`T_c`$ in Bi2212 materials .
Whether or not the pseudogap indeed has a superconducting origin remains to be verified experimentally. Such standard experimental test as the vanishing resistivity, or the Meissner effect are bound to fail. Both these test require spatial and temporal stability of the superconducting phase on the time scale of the experiment. On the other hand, the pseudogap state can at best have superconducting order parameter that varies both in space and in time. A successful test may be possible if the fluctuations could somehow be stabilized by the proximity to a “real” superconductor . Another approach is to probe superconductivity locally in space on the time scales comparable or shorter than the characteristic time of the phase fluctuations.
In this letter we propose the first experiment of this type, which is based on the AC Josephson effect. The main point we make is that in the presence of phase fluctuations Josephson current $`j(t)`$ across the tunneling contact is a random time-dependent quantity. It has a dispersion that is the current-current correlator $`j(t)j(t^{})`$, which is related to the corrections to the conductivity across the junction. This current-current correlator “remembers” about its Josephson origin and has a scale set by Josephson frequency and phase fluctuation time. Therefore, the conductivity of a junction will have a correction due to current fluctuations,
$`\mathrm{\Delta }\sigma 1/V,`$ (1)
where $`V`$ is the applied voltage across the junction. The crucial new aspect of the proposed approach is that we will focus on the characteristic time scale of frequency fluctuations, assuming that tunneling occurs in a small region where the spatial dependence can be ignored. We will focus on the time dynamics of the phase fluctuations in our analysis.
When two pieces of a superconductor are joined by a weak link, a superconducting current begins to flow through the link in the absence of applied voltage between the superconductors. The current is related to the difference of the phases $`\varphi _1`$ and $`\varphi _2`$ of the superconductors,
$`j=j_0\mathrm{sin}(\varphi _1\varphi _2),`$ (2)
with the parameter $`j_0`$ related to the coupling strength between the two superconductors. For a superconductor in an equilibrium, evolution of the phase,
$`\varphi /t=2\mu /\mathrm{},`$ (3)
is determined by the superconductor chemical potential, $`\mu `$. Therefore, the phase as a function of time is $`\varphi (t)=2\mu t/\mathrm{}+\varphi _0`$, where $`\varphi _0`$ is the phase at time $`t=0`$. For two coupled superconductors the phase difference evolves as
$`\varphi _1(t)\varphi _2(t)=2(\mu _2\mu _1)t/\mathrm{}+\varphi _1(0)\varphi _2(0).`$ (4)
The difference of the chemical potentials equals the applied voltage, $`\mu _2\mu _1=eV`$. Hence, if $`V=0`$, both the phase difference and the current given by Eq. (2) remain constant. However, in the presence of a bias, $`V`$, the phase difference grows linearly with time and the current oscillates according to
$`j(t)=j_0(\omega _J)\mathrm{sin}(\omega _Jt+\mathrm{\Delta }\varphi ),`$ (5)
with the frequency $`\omega _J=2eV/\mathrm{}`$ and the initial phase $`\mathrm{\Delta }\varphi =\varphi _1(0)\varphi _2(0)`$. The effect of generating an alternating current by applying a constant bias to a superconducting tunnel junction is called AC Josephson effect. An important feature of this effect is that the frequency of the generated current is only a function of the applied bias voltage and is independent of the microscopic and macroscopic parameters of the system. The scale of the frequency is about 0.5 $`\mathrm{\Gamma }`$Hz per 1 microvolt. The AC Josephson effect is routinely observed in the superconducting regime. It can be observed either directly as the micro-wave emission from the oscillating current, or indirectly as “Shapiro steps” in the DC $`IV`$ curves measured in the presence of the oscillating bias voltage component.
For AC Josephson effect to be observable in the pseudogap regime, the measurement has to be local both in space and time. Suppose that above the transition temperature, superconductor can be modeled as a collection of superconducting islands of a characteristic size $`L`$, inside which phase fluctuates at a rate $`\mathrm{\Lambda }`$. Then, if the size of a contact between two superconductors is less than $`L`$ then the superconducting state is essentially uniform in the vicinity of the contact. If the applied bias voltage is such that the Josephson frequency is larger than $`\mathrm{\Lambda }`$ then the Josephson oscillations are faster than the phase fluctuations dynamics, and hence can be approximately modeled as
$`j(t)=j^{}(\omega _J,\mathrm{\Lambda },L)\mathrm{sin}(\omega _Jt+\mathrm{\Delta }\varphi (t)),`$ (6)
with the amplitude $`j^{}`$ being renormalized by spatial and temporal fluctuations of the superconducting phase. In general, the Josephson frequency can also fluctuate around its average value due to voltage fluctuations which are particularly important for small samples. However, we assume that these effects can be absorbed into the overall fluctuations of the phase, $`\mathrm{\Delta }\varphi (t)`$. It is important to note that both $`\mathrm{\Lambda }`$ and $`L`$ are functions of temperature, with L and $`1/\mathrm{\Lambda }`$ diverging at the superconducting transition. The parameter $`L`$ is related to the phase gradient correlation function $`W`$ considered by Franz and Millis . Relationship between $`\mathrm{\Lambda }`$ and $`L`$ is a subject of an active interest. In the vortex diffusion picture, where phase fluctuations are produced by moving vortices, $`L`$ is a distance vortex travels during the time $`1/\mathrm{\Lambda }`$, namely $`L=\sqrt{D/\mathrm{\Lambda }}`$. Here $`D`$ is the vortex diffusion constant. This corresponding dynamical critical exponent is $`z=2`$. Alternatively, if the phase fluctuations are governed by fast ballistic dynamics, the relation between the parameters $`L`$ and $`\mathrm{\Lambda }`$ should be $`L\mathrm{\Lambda }=v^{}`$, where $`v^{}`$ is propagation speed for the ballistic modes. This corresponds to $`z=1`$. Using different geometries in the experimental setup that we propose below may help to determine the relevant model.
A possible experimental setup that can be used to perform the measurement of the AC Josephson effect in the pseudogap regime is shown in the figure 1. The crucial aspect is that the point of contact between the superconductors be as small as possible. If the size of the contact becomes larger than $`L`$ then in addition to the temporal phase fluctuations a the point of contact one needs to include spatial fluctuations, which can lead to a significant suppression of the effect. Another desirable feature is that the two superconductors be only a few $`ab`$-planes thick. This is because the size of the superconducting “islands” is likely to be more extended in the planes, compared to across the planes. Hence, we believe that the effect we propose is more likely to be observed in the geometry of figure 1, although c-axis tunneling may also yield similar results. Finally, using very thin samples reduces the transition temperature , thereby making the pseudogap regime accessible at lower temperatures, where the thermal fluctuations are reduced.
There are several ways the oscillating super-current in the pseudogap regime can be detected. Here we consider two methods: 1) differential AC conductivity measurements in the presence of constant bias voltage, 2) detection of electro-magnetic radiation generated by the oscillating Josephson current. Although there is no coherent Josephson current in the pseudogap regime, the junction is expected to have strong response to the perturbations acting at the frequencies near $`\omega _J`$. Such super-current is also expected to generate a radiation peak at the frequency $`\omega _J`$, with a width of the peak governed by the phase fluctuations.
To make our qualitative arguments more formal we have to assign a particular form to the phase fluctuations. Here we make an assumption that the the phase difference between the superconductors, $`\mathrm{\Delta }\varphi (t)`$ follows a diffusion process, as shown in Fig. 2, with a variance
$`(\mathrm{\Delta }\varphi _t\mathrm{\Delta }\varphi _t^{})^2=2\mathrm{\Lambda }|tt^{}|.`$ (7)
and the initial phase $`\mathrm{\Delta }\varphi _0`$ distributed uniformly in the interval $`[0,2\pi ]`$. The factor of 2 appears because for a weak tunnel junction the phases of on the both sides of the junction fluctuate independently, each at a rate $`\mathrm{\Lambda }`$.
Viewing the Josephson current of Eq. (6) as a random quantity with a zero mean, we can characterize it by its dispersion and autocorrelation. The autocorrelation, according to the Kubo formula, determines the correction to the conductivity due to the fluctuating Josephson tunneling,
$`\mathrm{\Delta }\sigma (\omega )={\displaystyle \frac{1}{\omega \nu }}{\displaystyle _{\mathrm{}}^t}e^{i\omega (tt^{})}j(t)j(t^{})𝑑t^{},`$ (8)
where brackets correspond to the $`\varphi `$-averaging, and averaging over time $`t`$ is implied. Volume $`\nu `$ is necessary for normalization. Substituting expression for the current from Eq. (6), we obtain
$`\mathrm{\Delta }\sigma `$ $`=`$ $`{\displaystyle \frac{j^2}{2\omega \nu }}{\displaystyle _{\mathrm{}}^t}e^{i\omega (tt^{})}\mathrm{cos}(\omega _J(tt^{})+\mathrm{\Delta }\varphi _t\mathrm{\Delta }\varphi _t^{})`$ (10)
$`\mathrm{cos}(\omega _J(t+t^{})+\mathrm{\Delta }\varphi _t+\mathrm{\Delta }\varphi _t^{})dt^{}.`$
Since $`\mathrm{\Delta }\varphi _t+\mathrm{\Delta }\varphi _t^{}=(\mathrm{\Delta }\varphi _t\mathrm{\Delta }\varphi _0)+(\mathrm{\Delta }\varphi _t^{}\mathrm{\Delta }\varphi _0)+2\mathrm{\Delta }\varphi _0`$, after averaging over $`\mathrm{\Delta }\varphi _0`$ in the interval $`[0,2\pi ]`$, the second cosine disappears. To average over $`(\mathrm{\Delta }\varphi _t\mathrm{\Delta }\varphi _t^{})`$, we invoke the relation $`\mathrm{exp}(iu)=\mathrm{exp}(u^2/2)`$, valid for any normally distributed variable $`u`$ with a mean zero. Then after the integration we obtain
$`\mathrm{\Delta }\sigma ={\displaystyle \frac{j^2}{4\omega \nu }}\left[{\displaystyle \frac{1}{\mathrm{\Lambda }+i\omega i\omega _J}}+{\displaystyle \frac{1}{\mathrm{\Lambda }+i\omega +i\omega _J}}\right].`$ (11)
As expected, the the real part of the conductivity,
$`\mathrm{Re}(\mathrm{\Delta }\sigma )={\displaystyle \frac{j^2\mathrm{\Lambda }}{4\omega \nu }}\left[{\displaystyle \frac{1}{(\omega \omega _J)^2+\mathrm{\Lambda }^2}}+{\displaystyle \frac{1}{(\omega +\omega _J)^2+\mathrm{\Lambda }^2}}\right],`$ (12)
which corresponds to the in-phase response, has two peaks near $`\pm \omega _J`$. The divergence as $`\omega 0`$ has no physical meaning, since no superconductivity related response is expected on the time scales larger than the characteristic phase fluctuation time. This translates into the condition $`\omega \omega _J`$ for validity of Eq. (12). The imaginary part of the conductivity in the vicinity of $`\omega _J`$ is about two times smaller than the real part, and hence can be neglected in the total conductivity $`\mathrm{Abs}\sigma =\sqrt{(\mathrm{Re}\sigma )^2+(\mathrm{Im}\sigma )^2}`$.
Therefore, we predict that if the pseudogap regime is superconducting in origin there should be a peak in the differential AC conductivity at the frequency $`\omega _JV`$, with the peak value that scales as shown in Eq. (1):
$`\mathrm{\Delta }\sigma j^2/\mathrm{\Lambda }\omega _J1/V.`$ (13)
This is the main result of this paper. While this correction may be small relative to the normal (single-electron) current component, it can be extracted from the background conductivity due to its extremely high sensitivity an applied external magnetic field. As is evident, the magnitude of the correction is inversely proportional to the phase-breaking rate, $`\mathrm{\Lambda }`$, and as a consequence should be more easily observable at temperatures close to the superconducting transition. Consequently, a possible experimental approach is to start arbitrarily close to $`T_c`$ and to measure the microwave radiation from the weakly dephased Josephson current, and or to measure the differential AC conductivity as proposed above. Then, gradually incrementing the temperature one can probe how the spatial and temporal fluctuations of the order parameter phase grow with the temperature.
In fact, a similar fluctuational AC Josephson effect can be searched for even in the conventional, “low-$`T_c`$,” superconductors, in the so-called paraconductivity regime . The paraconductivity regime is characterized by superconducting order parameter fluctuations above $`T_c`$, and experimentally is associated with the rapidly decreasing (but finite) resistivity in the vicinity of $`T_c`$. Using the experimental setup we propose here, one could attempt to study the dynamics of the dephasing timescales in the close proximity of $`T_c`$ in the paraconductivity regime. The difference between the conventional paraconductivity effect and the pseudogap is that the pseudogap is believed to extend far beyond the paraconductivity range where the rapid changes in the material resistivity occur.
Let us examine now more closely the assumptions that lead to the expression for the Josephson current, Eq. (6). Within the standard theory , the Josephson current is
$`j(t)=2e\mathrm{Im}[e^{i\omega _Jt}\mathrm{\Phi }_{\mathrm{ret}}(eV)],`$ (14)
where the retarded correlation function $`\mathrm{\Phi }_{\mathrm{ret}}(eV)`$ can be obtained from the Matsubara correlation function
$`\mathrm{\Phi }(i\omega )=2{\displaystyle \underset{\mathrm{𝐤𝐩}}{}}T_{𝐤,𝐩}T_{𝐤,𝐩}{\displaystyle _{\tau \beta }^\tau }𝑑\tau ^{}F^{}(𝐤,\tau ,\tau ^{})F(𝐩,\tau ^{},\tau ),`$ (15)
via analytical continuation $`i\omega eV+i0`$. Here $`T_{𝐤,𝐩}`$ is a matrix element for tunneling from a state $`𝐤`$ on one side of the junction into a state $`𝐩`$ on the other side of the junction, and $`F(𝐤,\tau ,\tau ^{})=T_\tau c_{𝐤(\tau )}c_𝐤(\tau ^{})`$ is an anomalous time-ordered Green functions. In Eq. (14), we do not include the regular single electron contribution, proportional to $`G(𝐤,\tau ,\tau ^{})G(𝐩,\tau ^{},\tau )`$. The reason is that it does not carry the superconducting phase information and, therefore, does not produce the resonant features away from zero frequency. In the absence of phase fluctuations $`F`$ is only a function of the time difference, $`F^0(𝐤,\tau \tau ^{})=(n_F(E_𝐤)e^{E_𝐤|\tau \tau ^{}|}n_F(E_𝐤)e^{E_𝐤|\tau \tau ^{}|})/2E_𝐤`$. Phase fluctuations can be incorporated phenomenologically into the anomalous Green functions as phase factors
$`F(𝐤,\tau ,\tau ^{})F(𝐤,\tau ,\tau ^{})e^{i\varphi (\tau ,\tau ^{})}.`$ (16)
The form of the function $`\varphi (\tau ,\tau ^{})`$ depends on the model of the phase fluctuations. Here we assume that
$`\varphi (\tau ,\tau ^{})=\varphi (\tau )+\varphi ^{}(\tau \tau ^{}),`$ (17)
where $`\varphi (\tau )`$ and $`\varphi ^{}(\tau \tau ^{})`$ are uncorrelated Brownian motions. Similar statistical properties of $`F(𝐤,\tau ,\tau ^{})`$ can be obtained from a gauge transformation of the electron operators, $`c_t=\psi _te^{i\mathrm{\Theta }(t)}`$, under which $`F(\tau ,\tau ^{})=T_\tau \psi _{𝐤(\tau )}\psi _𝐤(\tau ^{})e^{i\mathrm{\Theta }(\tau )+i\mathrm{\Theta }(\tau ^{})}`$. Since $`\mathrm{\Theta }(\tau )+\mathrm{\Theta }(\tau ^{})=2\mathrm{\Theta }(\tau )(\mathrm{\Theta }(\tau )\mathrm{\Theta }(\tau ^{}))`$, and under realistic assumptions ($`\mathrm{\Theta }(\tau )\mathrm{\Theta }(\tau ^{})`$) is only a function of $`(\tau \tau ^{})`$, this approach yields an expression equivalent to Eq. (17), except for the correlations induced between the functions $`\varphi (\tau )`$ and $`\varphi ^{}(\tau \tau ^{})`$. In what follows we assume for simplicity that the correlations are absent. Then doing the average over the Brownian random process $`\varphi ^{}`$ and integrating over $`\tau ^{}`$, for the Josephson current we obtain
$`j(t)=\mathrm{Abs}[j_0(\omega _J+i\mathrm{\Lambda })]sin(\omega _Jt+\mathrm{\Delta }\varphi (t)),`$ (18)
which is identical to the form of the Josephson current conjectured in Eq. (6). The function $`j_0(z)`$ is the analytical continuation of the function $`j_0(\omega _J)`$ which determines the amplitude of the Josephson current in the absence of the phase fluctuations. In the case of s-wave superconductivity with a constant gap $`\mathrm{\Delta }`$, this function is $`j_0(eV)=(\sigma _0\mathrm{\Delta }/e)K(eV/2\mathrm{\Delta })`$, defined in terms of complete elliptic integral $`K(x)`$. In the case of d-wave superconductor $`j_0(eV)`$ is also a nontrivial function of the relative orientation between lattices in two crystals. Its specific form is not important for our discussion. Finally, we should mention that in the current-current correlator, both $`\varphi `$ and $`\varphi ^{}`$ averages should be done on the product of currents, while we have done the averaging over $`\varphi ^{}`$ independently in $`j(t)`$ and $`j(t^{})`$. The qualitative results for conductivity, however, remain the same with the two peaks at the frequencies $`\pm \omega _J`$.
In conclusion, we propose to test the relevance of the phase fluctuations scenario in the pseudogap regime of the high-$`T_c`$ superconductors by investigating fluctuating Josephson current at $`T>T_c`$. We focus on the temporal fluctuations of the phase assuming small-contact tunneling to ignore spatial dependence of the phase. We argue that although phase fluctuations will yield zero mean Josephson current, its autocorrelation function will produce finite correction to the conductivity of normal current across the junction. AC conductivity will exhibit the peak at Josephson frequency $`\omega _J=2eV/\mathrm{}`$ with the width determined by the characteristic phase fluctuation rate $`\mathrm{\Lambda }`$. possible experimental test could be to measure the junction AC conductivity $`\sigma (\omega )`$ in the presence of constant bias $`V`$ and determine if it has a peak at $`\omega _J`$. We predict specific dependence $`\delta \sigma (\omega _J)V^1`$ of the peak. Specific experimental set up is shown on Fig. 1.
We would like to thank D. Morr, M. Graff, L. Bulayevski, J. Eckstein, and M. Maley for useful discussions. This work was supported by the DOE.
|
warning/0002/hep-th0002255.html
|
ar5iv
|
text
|
# Three dimensional Conformal Field Theories from Sasakian seven-manifoldsTalk presented at the TMR conference in Paris, September 1999.
## 1 Introduction
One of the most exciting aspects of string theory that has emerged over the past decade is the deep interplay between geometry and physics. The AdS/CFT correspondence provides us the possibility of testing the connection between some geometrical features of the supergravity (SUGRA) compactifying manifold and the field content of the dual superconformal field theory (SCFT).
The purpose of , on which this talk is based, is twofold. On one side we intend to identify the gauge theory living on the boundary of $`AdS_4`$, whose IR fixed point should realize the SCFT dual to 11D SUGRA compactified on $`AdS_4\times (G/H)^7`$, where $`(G/H)^7`$ is a homogeneous seven manifold. On the other side, we want to check the correspondence by comparing the spectra of the two theories.
Especially in the case of M-theory, where we have no deep control on the fundamental dynamics, the geometrical hints are essentially the only guidelines that help us in the construction of the worldvolume theory of a collection of branes placed at a conifold singularity in transverse space, namely the vertex of the metric cone over $`(G/H)^7`$. We will see that when this cone, $`𝒞(G/H)`$, admits a toric description, it is possible to argue the gauge group and the matter field content of this theory.
We are interested in the $`𝒩2`$ supersymmetric compactifications, which were classified in 1984 . The possible internal manifolds are the sasakian and trisasakian cosets:
$`\begin{array}{cc}M^{111}& {\displaystyle \frac{SU(3)\times SU(2)\times U(1)}{SU(2)\times U(1)\times U(1)}}\\ & \\ Q^{111}& {\displaystyle \frac{SU(2)\times SU(2)\times U(2)}{U(1)\times U(1)}}\\ & \\ V_{5,2}& {\displaystyle \frac{SO(5)}{SO(3)}}\end{array}\}𝒩=2`$
$`\begin{array}{c}N^{010}{\displaystyle \frac{SU(3)}{U(1)}}\end{array}𝒩=3`$
Contrary to the type IIB case of N $`D3`$-branes at a conifold singularity , where the existence of a nontrivial superpotential provides a constraint on the observable operators of the boundary SCFT, for the gauge theory of N $`M2`$-branes, relevant to this work, it seems to be no such a superpotential to help us in the identification of the effective degrees of freedom at the strong coupling conformal point.
Despite this lack, once again geometry comes in our aid. Indeed it is possible to recognize a common algebraic structure between the pattern of the KK short multiplets and the description of the transverse cone $`𝒞(G/H)`$ as an algebraic (sub)manifold. Relying on this analogy, it is possible to associate to the generators of the polynomial ring of $`𝒞(G/H)`$ particular combinations of fundamental fields of the gauge theory. These combinations seem to be the effective degrees of freedom of the CFT, generating the chiral ring of conformal chiral operators, at the base of the spectrum.
In this way we are able to identify the CFT boundary operators dual to all the bulk short multiplets, matching not only the conformal dimensions of the first with the energies of the latter, but even the flavor and $`R`$-symmetry quantum numbers of each couple. In this sense, the richer structure of the spectra of non-maximally supersymmetric theories guarantees a less trivial test of duality than in the case of spherical compactifications.
One further highly non-trivial check is given by the SUGRA predictions on the anomalous conformal dimensions of the fundamental fields of the boundary theory. These predictions are based on the global baryonic symmetries of the latter , which reflect the existence of non-trivial homology cycles of the internal $`(G/H)^7`$ .
Finally, an interesting aspect of these non-maximally supersymmetric AdS/CFT pairs, seems to be the existence of some intriguing quantum mechanism that prevents certain long multiplets of the spectrum from acquiring anomalous dimensions. On the base of a simple consideration on the spectra of different supersymmetric compactifications, we will try to hint a possible direction towards the solution of the puzzle.
## 2 Identification of the gauge theory
We want now to describe the fundamental steps in the search for the worldvolume thoery of N $`M2`$-branes sitting at the vertex of $`𝒞(G/H)`$. We will focus on the cases where the cone is a toric manifold, namely for $`(G/H)^7=M^{111}`$ or $`Q^{111}`$. Hence we are looking for an $`𝒩=2`$ supersymmetric gauge theory in three dimensions, whose moduli space of vacua should reproduce the one of the brane system. We make a minimality hypothesis about the matter field content: we assume that the fundamental fields, apart from the gauge bosons, transform in the most basic<sup>1</sup><sup>1</sup>1in the sense that all the higher representation are realizable as tensor products of this one. representation of the $`Osp(𝒩|4)`$ superconformal group: the supersingleton, which has a field-theoretic realization as a particularly constrained chiral superfield.
The general structure of a three dimensional $`𝒩=2`$ supersymmetric gauge theory coupled to chiral matter is severly constrained. The only freedom we have is in the choice of
* the gauge group;
* the scalar Kähler manifold, namely (in the renormalizable flat case) the number and flavor representations of the matter chiral multiplets;
* the holomorphic superpotential.
Let us begin with identifying the abelian gauge theory living on a single brane. Being the gauge fields frozen at the conformal point ($`g_{_{YM}}`$ is dimensionful), we can neglect the Coulomb branch and focus on the Higgs branch of the moduli space, parametrized by the vev of the scalars in the chiral multiplets. This branch should reproduce the space of vacua of the $`M2`$-brane, which is the transverse space $`𝒞(G/H)`$.
This requirement is easily achieved when the cone admits a toric description. Roughly speaking, this means that $`𝒞(G/H)`$ can be viewed as the Kähler quotient:
$$𝒞(G/H)=\frac{\mathrm{}^p}{(\mathrm{}^{})^{(p4)}}$$
(3)
for some $`p4`$. In this case, we can take as chiral fields the $`p`$ coordinates of $`\mathrm{}^p`$. The flavor quantum numbers will be determined by the natural lifting on $`\mathrm{}^p`$, of the isometries of $`(G/H)^7`$. By choosing as abelian gauge group the $`U(1)^{(p4)}`$ compact part of the modding $`(\mathrm{}^{})^{(p4)}`$, we obtain as $`D`$-term equations its non-compact $`\mathrm{}^{{}_{}{}^{+}(p4)}`$ part, in such a way to reproduce (3) as the moduli space of vacua of the gauge theory.
Let us see how this construction is concretely implemented in the case where $`(G/H)^7`$ is the coset $`M^{111}`$, whose metric cone admits the following toric description:
$$𝒞(M^{111})=\frac{\mathrm{}^^5}{\mathrm{}^{^{}}}.$$
(4)
The quotient is defined by the equivalence:
$$(U^i,V^A)(\lambda ^2U^i,\lambda ^3V^A),\lambda \mathrm{}^{^{}},$$
(5)
where $`U^i`$ and $`V^A`$ ($`i=1,2,3`$, $`A=1,2`$) are the five coordinates of $`\mathrm{}^^5`$, transforming respectively in the fundamental representation of the $`SU(3)`$ and of the $`SU(2)`$ flavor factors. The $`\mathrm{}^^+\mathrm{}^{^{}}`$ modding action can be fixed by imposing:
$$\underset{i}{}|U^i|^2=\underset{A}{}|V^A|^2,$$
(6)
while the $`U(1)`$ part of (5) determines the charges of the corresponding chiral fields (2 and -3 respectively).
With this choice of gauge group and chiral fields, the $`D`$-term of the bosonic potential is given by:
$$𝒰(z,\overline{z})=\left(\underset{i}{}|u^i|^2\underset{A}{}|v^A|^2\right)^2,$$
(7)
whose minimization, together with the gauge equivalence:
$$(u^i,v^A)(e^{2i\theta }u^i,e^{3i\theta }v^A)$$
(8)
exactly reproduces the equation of the cone (5). For symmetry reasons, it is worth to introduce a second $`U(1)`$ group, yielding the following couples of charges for the fundamental supersingletons:
$$\{\begin{array}{cc}U^i:& (2,2)\\ V^A:& (3,3)\end{array}.$$
(9)
The diagonal factor will actually decouple, leaving the gauge group $`U(1)^2/U(1)_{diagonal}`$.
The non-abelian generalization of this gauge theory is easily obtained by promoting the $`U(1)`$ gauge groups to $`SU(N)`$, with chiral matter in the following color representations:
$$\{\begin{array}{cc}U^i:& 𝐍^{_𝐬\mathrm{𝟐}}\overline{𝐍}^{_𝐬\mathrm{𝟐}}\\ V^A:& \overline{𝐍}^{_s3}𝐍^{_𝐬\mathrm{𝟑}}\end{array}.$$
(10)
## 3 The conformal theory
Let us now consider the observable fields at the IR fixed point of this theory, where the gauge fields are integrated out. First of all they will reduce to the gauge-invariant composite operators, whose smallest holomorphic combination (in the abelian case) is given by
$$X^{ijkAB}U^iU^jU^kV^AV^B,$$
(11)
transforming in the $`(\mathrm{𝟏𝟎},\mathrm{𝟑})`$ of $`SU(3)\times SU(2)`$.
An alternative description of $`𝒞(M^{111})`$ is in terms of an algebraic submanifold of $`\mathrm{}^{30}`$, where the complex space can be parametrized by the $`10\times 3=30`$ $`X^{ijkAB}`$ of (11). The embedding is given by 325 quadratic homogeneous equations in these coordinates $`X`$. In other words, on the defining locus of the cone certain quadratic combinations of the $`X^{ijkAB}`$ identically vanish. This fact is quite general and reveals a deep connection between algebraic geometry and representation theory. Without entering into mathematical details, it is worth to stress that the embedding equations are equivalent to putting to zero certain irreducible representations spanned by the quadratic products of the $`X`$’s. Indeed the symmetric tensor product:
$$X^{ijkAB}X^{lmnCD}(\mathrm{𝟏𝟎},\mathrm{𝟑})_𝐬(\mathrm{𝟏𝟎},\mathrm{𝟑})$$
(12)
branches into several irreducible representations of the flavor group $`SU(3)\times SU(2)`$. Among these, only the highest weight survives, while the other constitute just the 325 combinations that, once equated to zero, yield the embedding equations of the cone.
In the $`M^{111}`$ case, the only surviving polynomials of homogeneous degree 6 and 4, respcetively in the $`U`$’s and the $`V`$’s, are those combinations belonging to the completely symmetric flavor representation:
$$\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}\mathrm{}_{SU(3)}\mathrm{}\mathrm{}\mathrm{}\mathrm{}_{SU(2)}.$$
(13)
If we consider higher degree polynomials in the $`X`$’s we find that, due to the vanishing of certain quadratic subfactors, the only surviving combinations are always those completely symmetrized in the flavor indices.
The mathematical structure underlying these selection rules is that of a ring. On one side we have the polynomial ring characterizing the cone $`𝒞(M^{111})`$ as an algebraic manifold:
$$\frac{\mathrm{}[X^{ijkAB}]}{_{325}},$$
(14)
where $`_{325}`$ is the ideal of the embedding equations. On the other side we have the isomorphic chiral ring of the holomorphic color singlet operator products, modded out by the proper ideal of vanishing relations.
This ring structure is easily extended to the non-abelian generalization of the gauge theory. Indeed it is possible to show that there is always only one possible way to contract the gauge indeces into a color singlet, for each homogeneous polynomial in the $`X`$’s, completely flavor symmetrized.
It is important to remark the difference between our coset compactifications and the $`T^{11}`$ case of . While for the $`Q^{111}`$ and $`M^{111}`$ the non-highest weight part of (12) is given by a considerable set of equations (respectively 9 and 325), in the $`T^{11}`$ case the only vanishing quadratic combination of the $`X`$’s corresponds to the singlet representation of the flavor group $`SU(2)\times SU(2)`$ which, in the non-abelian case, turns out to be just the superpotential of the boundary theory. In the other $`M`$-theory cases, there is no way to combine all the non-highest weight pieces of (12) into a single flavor invariant with the right dimensions for a superpotential. This is the fundamental reason why the vanishing relations, which select the observable operators at the conformal point, are to be found in a highly non-perturbative quantum mechanism, other then the minimization of the potential.
## 4 Comparison with the KK spectrum
The first test of correspondence between the CFT and the compactified SUGRA is the matching of the spectra of the two theories.
The KK compactifications of 11D SUGRA were extensively studied in the first half of the 80’s. In particular, the coset space $`M^{111}`$ was widely investigated for its seemingly possibility to explain the origin of the $`SU(3)\times SU(2)\times U(1)`$ gauge bosons of the standard model.
The developement of techniques such as harmonic analysis on coset spaces , allowed the computation of great part of the spectrum of these compactifications. But the problems concerning the fermionic spectrum (first of all the absence of chirality) and the occurence of the first string revolution rapidly distorted the attention from KK SUGRA.
Due to the renewed interest, in connection to the AdS/CFT conjecture, the KK spectra of the $`M^{111}`$, $`Q^{111}`$ and $`V_{5,2}`$ compactifications have been recently completed in a systematic way .
We will focus in particular on the CFT operators of protected conformal dimensions, corresponding to KK states belonging to short representations of the $`Osp(2|4)`$ symmetry superalgebra. The following table lists all the different kinds of $`BPS`$-saturated representations and the supercovariant differential constraint implementing the corresponding shortening conditions on the boundary superfields .
$$\begin{array}{cc}& \\ \text{ multiplet}& \text{ differential constraint on}\\ & \text{ the boundary superfield}\\ & \\ \mathrm{short}\mathrm{graviton}& 𝒟^{+\alpha }\mathrm{\Phi }_{(\alpha \beta )}(x,\theta ^\pm )=0\\ & \\ \mathrm{short}\mathrm{gravitino}& 𝒟^{+\alpha }\mathrm{\Phi }_\alpha (x,\theta ^\pm )=0\\ & \\ \mathrm{short}\mathrm{vector}& 𝒟^{+\alpha }𝒟_\alpha ^+\mathrm{\Phi }(x,\theta ^\pm )=0\\ & \\ \mathrm{hypermultiplet}& 𝒟^{+\alpha }\mathrm{\Phi }(x,\theta ^\pm )=0\\ & \\ \mathrm{massless}\mathrm{graviton}& \{\begin{array}{c}𝒟^{+\alpha }\mathrm{\Phi }_{(\alpha \beta )}(x,\theta ^\pm )=0\\ 𝒟^\alpha \mathrm{\Phi }_{(\alpha \beta )}(x,\theta ^\pm )=0\end{array}\\ & \\ \mathrm{massless}\mathrm{gravitino}& \{\begin{array}{c}𝒟^{+\alpha }\mathrm{\Phi }_\alpha (x,\theta ^\pm )=0\\ 𝒟^\alpha \mathrm{\Phi }_\alpha (x,\theta ^\pm )=0\end{array}\\ & \\ \mathrm{massless}\mathrm{vector}& \{\begin{array}{c}𝒟^{+\alpha }𝒟_\alpha ^+\mathrm{\Phi }(x,\theta ^\pm )=0\\ 𝒟^\alpha 𝒟_\alpha ^{}\mathrm{\Phi }(x,\theta ^\pm )=0\end{array}\\ & \\ \mathrm{supersingleton}& \{\begin{array}{c}𝒟^{+\alpha }\mathrm{\Phi }(x,\theta ^\pm )=0\\ 𝒟^\alpha 𝒟_\alpha ^{}\mathrm{\Phi }(x,\theta ^\pm )=0\end{array}\end{array}$$
For space reasons, it is not possible here, to make a complete list of the protected conformal superfields corresponding to all the short multiplets of the KK spectrum. We will focus on the most meaningful.
As we anticipated, the chiral ring of the SCFT is given by the following composite operators:
$$\mathrm{𝑇𝑟}\left[(U^3V^2)^k\right],$$
(15)
where, following the prescription given in the last section, the flavor indices are completely symmetrized, while the trace symbol implies the only possible contraction into a color-singlet, compatible with this flavor representation. These chiral superfields find a complete matching with the tower of hypermultiplets of the KK spectrum. The only constraint we have, is on the anomalous dimensions of the fundamental supersingletons, which we are not able to directly deduce from the CFT:
$$2=\mathrm{\Delta }[Tr(U^3V^2)]=3\mathrm{\Delta }(U)+2\mathrm{\Delta }(V).$$
(16)
In the next section we will show how $`\mathrm{\Delta }(U)`$ and $`\mathrm{\Delta }(V)`$ can be obtained from SUGRA computations, confirming the validity of our conjecture on the dual CFT.
Other important superfields are the conserved supercurrents:
$${}_{}{}^{SU(3)}J_{j}^{i}=Tr\left[U^i\overline{U}_j\frac{1}{3}\delta _j^iU^k\overline{U}_k\right]$$
(17)
and
$${}_{}{}^{SU(2)}J_{B}^{A}=Tr\left[\overline{V}^AV_B\frac{1}{2}\delta _B^A\overline{V}^CV_C\right],$$
(18)
that transform in the adjoint representation of the flavor groups, $`SU(3)`$ and $`SU(2)`$ respectively, and contain the massless vectors associated to these symmetries. Two sets of semiconserved currents, corresponding to short vector multiplets of the KK spectrum are given by composing (17) and (18) with the chiral fields (15).
Finally, two towers of KK short gravitinos are matched by the following spinorial superfields:
$$Tr\left[\left(U\overline{U}(𝒟_\alpha ^+\overline{V}V)+\overline{V}V(𝒟_\alpha ^+U\overline{U})\right)_{jB}^{iA}(U^3V^2)^k\right]$$
and
$$Tr\left[\left(U^iU^jU^kV^A𝒟_\alpha ^{}V^Bϵ_{AB}\right)(U^3V^2)^k\right].$$
All these superfields and the other which we have not listed, perfectly fit the masses, $`R`$-symmetry charges and flavor quantum numbers of the corresponding KK short multiplets.
## 5 The baryonic symmetries
The next item we want to discuss is the presence in the boundary theory of barion-like conformal operators and their interpretetion in terms of non-perturbative bulk states.
Already in the gold years of KK SUGRA, it was noted that the existence of non trivial homological cycles in the internal manyfold implies the presence of massless vectors in the spectrum of the compactified SUGRA, due to the reduction of higher degree form potential on these cycles. The corresponding Betti multiplets have been related, in the AdS/CFT perspective, to global baryonic symmetries of the dual gauge theory .
In the specific case considered in this talk, it is found that the manyfold $`M^{111}`$ has non vanishing second and fifth Betti numbers: $`b_2(M^{111})=b_5(M^{111})=1`$, implying a $`U(1)`$ baryonic symmetry. All the KK multiplets (and hence all the dual conformal operators) result discharged under this symmetry. But we have not analysed, so far, all the possible color singlet products of fundamental supersingletons at our disposal. Among these, we have the totally antisymmetrized<sup>2</sup><sup>2</sup>2implying total symmetrization in the flavor indices (in the $`SU(N)`$ color indices) product of $`N`$ singletons of the same kind, that we will abbreviate $`det(U)`$ and $`det(V)`$. The masses of these operators grow like $`N`$ in the large $`N`$ limit. Hence their SUGRA duals have to be found among non-perturbative states. Indeed we find the existence in $`M^{111}`$ of two particular families of non trivial supersymmetric 5-cycles, over which a solitonic $`M5`$-brane can be wrapped. The energy of these $`M`$-theory configurations, simply related to the volume of the cycles, has the same large $`N`$ behaviour. The exact coefficient in this linear expression can be used as a SUGRA prediction on the conformal anomalous dimension of the corresponding fundamental singleton. In this case we find:
$`\mathrm{\Delta }[det(U)]={\displaystyle \frac{4N}{9}}\mathrm{\Delta }(U)={\displaystyle \frac{4}{9}}`$
$`\mathrm{\Delta }[det(V)]={\displaystyle \frac{N}{3}}\mathrm{\Delta }(V)={\displaystyle \frac{1}{3}}`$
These numbers perfectly fit the only constraint (16) we have from the SCFT side:
$$3\mathrm{\Delta }(U)+2\mathrm{\Delta }(V)=2.$$
(19)
Furthermore, even the flavor irreducible representation of these configurations (deducible from the action of the $`M^{111}`$ isometries on the cycles of each family) perfectly agree with the flavor quantum numbers of the corresponding determinant operators, yielding a highly non trivial check of duality.
## 6 Comments about the rational long multiplets
An interesting feature common to both type IIB and 11D SUGRA coset compactifications, first noted in for the $`T^{11}`$ space, is the presence in the KK spectrum of long multiplets with rational energies (see also the talk by G. Dall’Agata about this point). This fact does not seem to be a mere coincidence. Indeed it has been recognized that the dual conformal operators follow a precise pattern: they are products of two or more quantities separately protected that, nevertheless, have no a priori reason to be globally protected. The conformal dimension of these operator products, as it is deduced by the energy of the corresponding SUGRA states, is instead given by the naive sum of the dimensions of the single factors.
To make a concrete example, both the $`M^{111}`$ and $`Q^{111}`$ bulk spectra contain a tower of long graviton multiplets, associated to boundary vector superfields of the form:
$$\mathrm{\Phi }_{(\alpha \beta )}T_{(\alpha \beta )}\times J\times \varphi ,$$
(20)
where $`T_{(\alpha \beta )}`$, $`J`$ and $`\varphi `$ are respectively the stress energy tensor, a conserved vector current and a chiral operator, and the dimension of $`\mathrm{\Phi }`$ is precisely given by:
$$\mathrm{\Delta }(\mathrm{\Phi })=\mathrm{\Delta }(T)+\mathrm{\Delta }(J)+\mathrm{\Delta }(\varphi ).$$
(21)
We have not yet a definitive explaination of the quantum mechanism that seems to protect these operators from acquiring anomalous dimensions. But a simple consideration has emerged while studying the $`𝒩=3`$ $`SCFT`$ corresponding to the SUGRA compactification on the trisasakian manifold $`N^{010}`$ .
From the analysis of the $`Osp(3|4)`$ supermultiplets and of their decomposition into $`𝒩=2`$ irreducible representations , one can realize that the short $`𝒩=3`$ multiplets always contain long multiplets of the lower superalgebra. Furthermore, from a careful identification of the corresponding boundary conformal operators, we have recognized for some of these the same structure of (20). Obviously, belonging to the same supermultiplet of the higher supersymmetry algebra, these long $`𝒩=2`$ multiplets have energies that differ from those of the other states only by (half)-integers. Hence they are necessarily rational, despite the fact that, from the $`𝒩=2`$ viewpoint, they are long.
This simple consideration seems to hint that the quantum protection mechanism previously advocated may be found in a residual form of higher supersymmetry that the SCFT could reflect, such as a spontaneous supersymmetry breaking.
### Acknowledgements.
I would like to thank all the authors of the paper , on which this talk is based and also P. Termonia for his collaboration in the preliminary results on harmonic analysis. This work is supported by the European Commission TMR programme ERBFMRX-CT96-0045.
|
warning/0002/nucl-ex0002002.html
|
ar5iv
|
text
|
# Search for the electric dipole excitations to the 3𝑠_{1/2}⊗[2⁺₁⊗3⁻₁] multiplet in 117Sn
## I Introduction
Low-lying electric dipole excitations have been studied extensively in a variety of spherical and deformed nuclei over the last decade. A survey on the systematics of observed electric dipole excitations in the A = 130–200 mass region is given in Ref. . Systematic nuclear resonance fluorescence (NRF) experiments performed within the chains of the N = 82 isotones (<sup>138</sup>Ba, <sup>140</sup>Ce, <sup>142</sup>Nd and <sup>144</sup>Sm) and the Z = 50 isotopes (<sup>116-124</sup>Sn) showed that the low-lying electric dipole strength $`B(E`$1)$``$ is mainly concentrated in the first $`J^\pi =1_1^{}`$ state. Some uniform properties of these $`1_1^{}`$ states were observed in both chains. Their excitation energies are lying close to the summed energies of the first quadrupole and octupole collective vibrational states and they are populated by an enhanced electric dipole excitation (two to three orders of magnitude larger than other low-lying $`1^{}`$ states). Both arguments, strongly suggest an underlying quadrupole-octupole coupled $`[2_1^+3_1^{}]`$ two-phonon structure. Indeed, a detailed microscopic study within the framework of the Quasi-particle phonon model revealed a practically pure two-phonon $`[2_1^+3_1^{}]_1^{}`$ configuration in the wave function of these $`1_1^{}`$ states. The observed enhanced electric dipole strength of the “forbidden” $`E1`$ transitions can be reproduced from a consideration of the internal fermion structure of the phonons and taking into account a delicate destructive interference with the GDR $`1^{}`$ one-phonons . The most direct experimental proof for an underlying two-phonon structure can be obtained from a measurement of the decay pattern of the two-phonon states to their one-phonon components. This has been achieved up to now only in very few cases: <sup>142</sup>Nd, <sup>144</sup>Nd and <sup>144</sup>Sm . In each of these nuclei, an enhanced $`B(E2)`$ strength in the decay of $`1_1^{}3_1^{}`$ could be measured consistent with a two-phonon picture. It is still a tough challenge to observe the other members of the $`[2_1^+3_1^{}]`$ two-phonon quintuplet as illustrated recently in Ref. .
As a natural extension of the systematic investigations on the even-even spherical nuclei, the question arises how the observed enhanced electric dipole excitation strength of the two-phonon $`[2_1^+3_1^{}]_1^{}`$ state fragments over several levels of a particle two-phonon coupled multiplet in the odd-mass adjacent nuclei. In such a study, the experimental technique and the theoretical model should meet some requirements. In the first place, the experimental probe should be very selective in the excitation of levels because of the high level density in odd-mass nuclei. The number of the involved levels in the odd-mass nucleus increases drastically compared to the even-even neighbouring nucleus: in the odd-mass nucleus levels with a spin equal $`J_01`$, $`J_0`$ and $`J_0+1`$ can be populated via dipole excitations from the ground state with spin $`J_0`$. Quadrupole transitions will excite levels with a spin between $`J_02`$ and $`J_0+2`$. The real photon probe is such a selective experimental tool. Using an intensive bremsstrahlung source only dipole and to a much lesser extent also electric quadrupole excitations will be induced. Secondly, the theoretical model should be able to distinguish between the degrees of freedom of the collective phonons and the additional particle degrees of freedom which open extra possible excitation channels and which have no counterparts in the even-even nuclei.
Electric dipole excitations to a particle two-phonon multiplet were for the first time identified in <sup>143</sup>Nd . In the energy region between 2.8 and 3.8 MeV, 13 levels were observed for which an underlying particle two-phonon $`f_{7/2}[2_1^+3_1^{}]`$ structure was suggested. The observed fragmentation and $`B(E`$1)$``$ strength distribution could be reproduced in a phenomenological simple core coupling model based on quadrupole-quadrupole coupling . Moreover, the summed experimental $`B(E`$1)$``$ strength between 2.8 and 3.8 MeV agrees within the statistical error with the known $`B(E`$1)$``$ strength of the two-phonon $`[2_1^+3_1^{}]_1^{}`$ state in the neighbouring even-mass <sup>142</sup>Nd nucleus. It was concluded that the unpaired neutron in its $`f_{7/2}`$ orbital, outside the closed major N = 82 shell, couples extremely weakly to the two-phonon $`[2_1^+3_1^{}]`$ quintuplet and plays the role of a pure spectator. Later on, similar NRF-experiments performed on <sup>139</sup>La and <sup>141</sup>Pr revealed also a large fragmentation of the electric dipole strength, but in both cases less than 40% of the two-phonon $`B(E`$1)$``$ strength in the neighbouring even-mass <sup>138</sup>Ba, <sup>140</sup>Ce and <sup>142</sup>Nd nuclei was observed. In <sup>139</sup>La and <sup>141</sup>Pr the odd proton in the partly filled shell couples more strongly to the two-phonon quintuplet. Intermediate cases have been observed in the open shell nuclei <sup>113</sup>Cd and <sup>133</sup>Cs .
The odd-mass <sup>117</sup>Sn nucleus was chosen to investigate the fragmentation of the well-known two-phonon $`B(E`$1)$``$ strength from its even-mass neighbours <sup>116</sup>Sn and <sup>118</sup>Sn. The NRF technique was used for obvious reasons. In <sup>117</sup>Sn, the unpaired neutron is situated halfway the major N = 50 and 82 shells. As an interesting property, the ground state spin $`J_0^\pi =1/2^+`$ limits the possible dipole excitations to levels with a spin $`J=1/2`$ or $`3/2`$ and electric quadrupole excitations can only occur to states with a spin and parity $`J^\pi =3/2^+`$ and $`5/2^+`$. For the first time, calculations within the QPM are carried out in a complete configuration space for an odd-mass spherical nucleus. Our first results on the experimentally observed fragmentation and on the performed calculations were described in a previous letter paper . In the present paper a comprehensive discussion of the experimental and theoretical aspects of our work will be presented.
## II Experimental method and setup
The nuclear resonance fluorescence technique or the resonant scattering of real photons off nuclei as described in many reviewing articles e.g. , was applied to investigate the <sup>117</sup>Sn nucleus. The main advantages of this real photon probe consist of the highly selective excitation of levels, as pointed out in the introduction, and the possibility of a completely model independent analysis of the data. The use of HP Ge detectors to detect the resonantly scattered photons, allows the observation of individual levels and the determination of their excitation energies $`E_x`$ with a precision better than 1 keV. The total elastic scattering cross section $`I_S`$, energy integrated over a single resonance and integrated over the full solid angle equals :
$$I_S=g\left(\frac{\pi \mathrm{}c}{E_x}\right)^2\frac{\mathrm{\Gamma }_0^2}{\mathrm{\Gamma }}$$
(1)
with $`\mathrm{\Gamma }_0`$ and $`\mathrm{\Gamma }`$ the ground state and total transition width and $`g`$ a statistical factor depending on the ground state spin $`J_0`$ and the spin $`J`$ of the excited level:
$$g=\frac{2J+1}{2J_0+1}.$$
(2)
In our experiments, the scattering cross sections of the observed levels in <sup>117</sup>Sn are determined relative to the <sup>27</sup>Al calibration standard. The spectral shape of the incoming bremsstrahlung flux is fitted using the Schiff formula for thin targets and the cross sections for the transitions in <sup>117</sup>Sn are extracted relatively to the cross sections of well-known transitions in <sup>27</sup>Al . For even-even nuclei, the spin $`J`$ of the excited level is easily obtained from the measured angular distribution of the resonantly scattered photons. A clear distinction between dipole and quadrupole transitions can be made by comparing the observed $`\gamma `$ intensities at the scattering angles of $`90^{}`$ and $`127^{}`$ . However, for odd-mass nuclei the extraction of limited spin information is only possible for nuclei as <sup>117</sup>Sn with a ground state spin $`J_0`$ of 1/2 . Higher half integer ground state spins lead to nearly isotropic angular distributions. From an evaluation of the angular distribution function $`W`$ for the involved spin sequence, scattering angle and mixing ratio $`\delta `$, the following results are obtained:
$`\frac{1}{2}\frac{1}{2}\frac{1}{2}`$
$$W(90^{})=W(127^{})=W(150^{})=1$$
(3)
$`\frac{1}{2}\frac{3}{2}\frac{1}{2}`$
$$\frac{W(90^{})}{W(127^{})}=0.866;\frac{W(90^{})}{W(150^{})}=0.757;\delta =0,\pm \mathrm{}$$
(4)
$`\frac{1}{2}\frac{5}{2}\frac{1}{2}`$
$$\frac{W(90^{})}{W(127^{})}=1.168;\frac{W(90^{})}{W(150^{})}=0.842;\delta =0.$$
(5)
The scattering angles correspond to those used in our experiments. The angular distribution functions $`W`$ are independent of the parity of the excited level. In Fig. 1 the expected angular correlation function ratios $`W(90^{})/W(150^{})`$ are plotted versus the ratio $`W(90^{})/W(127^{})`$ for different values of the mixing ratio $`\delta `$ and for the three possible induced spin sequences. A level with spin 1/2 can only be excited from the ground state via a dipole transition and hence no mixing of multipolarities is possible in a 1/2 – 1/2 – 1/2 spin cascade. Such a cascade has an isotropic angular distribution function $`W`$. The square in Fig. 1 represents the unique location in this figure where 1/2 – 1/2 – 1/2 spin sequences can be found. In the case of a 1/2 – 3/2 – 1/2 spin sequence, assuming a positive parity for the 1/2 state (as is the case in <sup>117</sup>Sn), a mixing between $`E1`$ and $`M2`$ or $`M1`$ and $`E2`$ transitions is possible depending on the parity of the excited 3/2 level. However, $`M2`$ transitions can not be observed in NRF experiments. For a pure $`E1`$ ($`M1`$) transition, corresponding to a mixing ratio $`\delta `$ equal 0, or a pure $`M2`$ ($`E2`$) transition, with mixing ratio $`\delta `$ equal $`\pm \mathrm{}`$, to the 3/2 level with a negative (positive) parity, the same two values for the ratios $`W(90^{})/W(127^{})`$ and $`W(90^{})/W(150^{})`$ are obtained. This is represented by the triangle in Fig. 1. For mixing ratios varying between 0 and $`\pm \mathrm{}`$ the point representing the couple of ratios $`W(90^{})/W(127^{})`$ and $`W(90^{})/W(150^{})`$ moves along the solid line in Fig. 1. For a 1/2 – 5/2 – 1/2 spin sequence, transitions with multipolarities L = 2 and L = 3 are theoretically possible. However, $`M2`$, $`E3`$ and $`M3`$ excitations can be excluded to be observed in our NRF-experiments because they have scattering cross sections far below the sensitivity of our setup. Therefore, only a pure $`E2`$ excitation to a 5/2 level can be observed. The star in Fig. 1 shows where these 1/2 – 5/2 – 1/2 spin sequences can be expected. For the strongest transitions observed with a high statistical precision, the experimental uncertainties on the ratios $`W(90^{})/W(127^{})`$ and $`W(90^{})/(150^{})`$ will be small enough to suggest the induced half integer spin sequence. In most cases the statistical accuracy will not allow to determine the spin sequence and hence the statistical spin factor $`g`$ can not be determined.
Compton polarimetry and the scattering of linearly polarized off-axis bremsstrahlung represent two useful techniques to determine the parities of the excited levels in even-even nuclei . For odd-mass nuclei, the measured azimuthal asymmetry of the resonantly scattered photons in both methods will nearly vanish because of the half integer spin sequences and strongly depend on the mixing ratio $`\delta `$. The lower statistics in these experiments do not allow to distinguish between the different possible cases. For a further analysis of the data, it will be assumed that all observed levels are populated via pure electric dipole excitations and that they do not have any decay branchings (unless observed otherwise) to intermediate lower-lying levels ($`\mathrm{\Gamma }_0/\mathrm{\Gamma }=1`$). In this case, the product of the ground state transition width and the spin factor g can immediately be extracted from the measured scattering cross section $`I_S`$ (Eq. 1) and the reduced electric dipole excitation probability is given by:
$$B(E1)=\frac{2.866}{3}\frac{g\mathrm{\Gamma }_0}{E_x^3}(10^3e^2fm^2)$$
(6)
with the ground state transition width $`\mathrm{\Gamma }_0`$ in meV and the excitation energy $`E_x`$ in MeV. The deduced $`B(E`$1)$``$ strength in our NRF-experiments, includes automatically the statistical factor g and hence can immediately be compared with the observed $`B(E`$1)$``$ strength in other nuclei or calculated from theoretical models.
The experiments were performed at the NRF-facility of the 4.3 MV Dynamitron accelerator of the Stuttgart University. The energy of the electron beam was 4.1 MeV and the beam current was limited to about 250 $`\mu `$A due to the thermal characteristics of the bremsstrahlung production target and to avoid too high count rates in the detectors. A setup consisting of 3 HP Ge detectors, installed at scattering angles of 90, 127 and 150 each with an efficiency $`ϵ`$ of 100% (relative to a 3” x 3” NaI(Tl) detector), was used to measure the total elastic scattering cross sections of the levels in <sup>117</sup>Sn. These HP Ge detectors allow to detect the resonantly scattered photons with a high sensitivity and a very good energy resolution. Two metallic Sn disks with a diameter of 1 cm, a total weight of 1.649 g, and an isotopic enrichment of 92.10% in <sup>117</sup>Sn were irradiated during five days. Two <sup>27</sup>Al disks with a total amount of 0.780 g were alternated with the two <sup>117</sup>Sn disks for calibration purposes.
## III Results
Part of the recorded <sup>117</sup>Sn ($`\gamma `$,$`\gamma ^{}`$) spectrum (2.6 $``$ $`E_\gamma `$ $``$ 3.6 MeV) is shown in Fig. 2 together with the spectra of its even-even <sup>116</sup>Sn and <sup>118</sup>Sn neighbours. In our previous studies on <sup>116</sup>Sn and <sup>118</sup>Sn , we found that the $`E1`$ strength in this energy region below 4 MeV is mostly concentrated in the two-phonon $`[2_1^+3_1^{}]_1^{}`$ state. In the spectra of the even-even nuclei, the dominating peak at an energy of about 3.3 MeV corresponds to the deexcitation of this two-phonon 1<sup>-</sup> state into the ground state. In comparison to the spectra of the even-even Sn nuclei, the <sup>117</sup>Sn spectrum contains a lot of $`\gamma `$ transitions superimposed on the smooth background in the vicinity of the two-phonon 1<sup>-</sup> states . The peaks stemming from the deexcitation of the two-phonon $`[2_1^+3_1^{}]_1^{}`$ states in <sup>116,118,120</sup>Sn have also been observed in the spectrum due to the small admixtures of 0.86%, 5.81% and 0.76% in the target. In Fig. 2b only the peak for <sup>118</sup>Sn can be clearly observed due to the scale used. All observed $`\gamma `$ transitions in <sup>117</sup>Sn are listed in Table I with the corresponding level excitation energies, the measured total elastic scattering cross sections $`I_S`$, the extracted transition width ratios $`g\frac{\mathrm{\Gamma }_0^2}{\mathrm{\Gamma }}`$ (depending on the statistical spin-factor g) and deduced electric excitation probabilities $`B(E`$1)$``$. The total elastic scattering cross sections has been determined from a summed spectrum over the three scattering angles. The three HP Ge-detectors have nearly the same efficiency and the summed angular distributions over the three detectors equals three for all possible spin cascades. This method allows us to observe also some weaker lines.
In some cases, the observed angular distribution ratios provide an indication of the spin $`J`$ of the photo-excited levels. These results are summarized in Table II. In the first column the energies of these levels are given. In the next columns the observed angular distribution ratios $`W(90^{})/W(127^{})`$ and $`W(90^{})/W(150^{})`$ and the suggested spin are presented. The experimentally observed angular distribution ratios $`W(90^{})/W(150^{})`$ versus $`W(90^{})/(127^{})`$ are included in Fig. 1 (points with error bars). Two groups of $`\gamma `$ transitions can clearly be observed. For a first group of 5 levels, located around the triangle, a probable spin assignment of 3/2 can be given. A second group of 6 levels can be found around the square. Here an assignment of 1/2 (3/2) can be given. We prefer $`J=1/2`$ for these states because we suppose that strong transitions should have an $`E1`$ character. These $`E1`$ transitions have an isotropic distribution only for excited states with $`J=1/2`$. However we can not exclude $`J=3/2`$ (negative or positive parity) as transitions with a mixing ratio $`\delta `$ around -3.73 or 0.27 also lead to an isotropic distribution. All the spins given in Table II were assigned at least at a statistical significance level of $`1\sigma `$. Under these conditions, no levels with a spin of 5/2 were found.
Five of the observed $`\gamma `$ transitions are probably due to inelastic deexcitations of a level to the well-known first $`3/2^+`$ state in <sup>117</sup>Sn at 158.562(12) keV . They are summarized in Table III. In this Table the level excitation energies $`E_x`$, the energies $`E_\gamma `$ of the deexciting $`\gamma `$ transitions, the relative $`\gamma `$ intensities $`I_\gamma `$, the ground state transition widths $`g\mathrm{\Gamma }_0`$ and reduced electric dipole excitation probabilities corrected for the observed inelastic decays, are given. All levels are included for which holds:
$$E_i(E_xE_{158})<\mathrm{\Delta }E$$
(7)
with:
$$\mathrm{\Delta }E=\sqrt{(\mathrm{\Delta }E_x)^2+(\mathrm{\Delta }E_i)^2+(\mathrm{\Delta }E_{158})^2}$$
(8)
and $`E_x`$, $`E_i`$, $`E_{158}`$ and $`\mathrm{\Delta }E_x`$, $`\mathrm{\Delta }E_i`$, $`\mathrm{\Delta }E_{158}`$ the excitation energies of the level, the energy of the inelastic $`\gamma `$ transition and the first $`3/2^+`$ state and their respective uncertainties. Using the above mentioned rule for detecting inelastic decays two other candidates were found. The $`\gamma `$ ray with the energy of 3560.5 keV can be due to an inelastic transition of one line from the multiplet at 3719.8 keV to the first $`3/2^+`$ state. Also the $`\gamma `$ ray with the energy of 2986.7 keV fits into the energy relation with the observed level at 3144.9 keV. However, when this $`\gamma `$ ray is seen as completely inelastic, an unreasonable low branching ratio $`\mathrm{\Gamma }_0/\mathrm{\Gamma }`$ of 12% for the 3144.9 keV level is obtained. Both cases are not considered in Table III. Certainty about the observed probable inelastic $`\gamma `$ transitions requires time coincidence measurements. With the above mentioned method, our experimental results show no further candidates for inelastic transitions to other intermediate observed levels.
## IV Discussion
According to the phenomenological core coupling model, the level scheme for <sup>117</sup>Sn can be obtained from the coupling of the odd 3s<sub>1/2</sub> neutron to the <sup>116</sup>Sn core. This is schematically shown in Fig. 3. Each level in <sup>116</sup>Sn (except for $`J=0`$ levels) gives rise to two new levels due to the spin 1/2 of the odd neutron. The low-lying level scheme of <sup>116</sup>Sn is dominated by the strong quadrupole ($`2^+`$) and octupole ($`3^{}`$) vibrational states, typical for a spherical semi-magic nucleus. The coupling of the 3s<sub>1/2</sub> neutron to the first $`2^+`$ state in <sup>116</sup>Sn results in two new levels $`[3s_{1/2}2_1^+]_{3/2^+}`$ and $`[3s_{1/2}2_1^+]_{5/2^+}`$ which can be excited in NRF via $`M1`$ and $`E2`$ transitions (solid lines in Fig. 3). The similar doublet consisting of the $`[3s_{1/2}3_1^{}]_{5/2^{}}`$ and $`[3s_{1/2}3_1^{}]_{7/2^{}}`$ states requires $`M2`$ and $`E3`$ excitations which are not observable in NRF (dashed lines). The main aim of our NRF-experiments was to search for levels belonging to the $`3s_{1/2}[2_1^+3_1^{}]`$ multiplet which can be populated via electric dipole transitions. When the quadrupole-octupole coupled two-phonon quintuplet is built on top of the 1/2<sup>+</sup> ground state, a multiplet of 10 negative parity states is obtained of which 3 levels can be excited via $`E1`$ transitions (solid lines in Fig. 3). In this simple model, these three transitions carry the complete $`B(E`$1)$``$ strength. In Fig. 4b the obtained total scattering cross sections $`I_S`$ for the photo-excited levels in <sup>117</sup>Sn (with exclusion of the lines which are probable due to inelastic scattering given in Table III) and in its even-even neighbours <sup>116</sup>Sn and <sup>118</sup>Sn are plotted. The total scattering cross section for the excitation of the two-phonon $`[2_1^+3_1^{}]`$ states in <sup>116</sup>Sn and <sup>118</sup>Sn has been reduced by a factor of 3. A strong fragmentation of the strength has been observed in <sup>117</sup>Sn compared to its even-even neighbours. It is already clear from the observed fragmentation of the strength that a phenomenological core coupling model, which was successful in describing the observed strength in <sup>143</sup>Nd, will be insufficient in our case. Due to a lack of spin and parity information of the photo-excited levels in <sup>117</sup>Sn in our NRF-experiment and due to a lack of experimental data from other investigations , we need to turn to a more elaborated theoretical interpretation to get more insight.
### A QPM formalism for odd-mass nuclei
The Quasiparticle phonon model (QPM) was already successful in describing collective properties in even-even mass nuclei . Recently, the QPM has been applied to describe the position and the $`E1`$ excitation probability of the lowest $`1^{}`$ state in the even-even <sup>116-124</sup>Sn isotopes . This state has a two-phonon character with a contribution of the $`[2_1^+3_1^{}]_1^{}`$ configuration of 96-99%. For odd-mass nuclei, this model was used to describe the fragmentation of deep-lying hole and high-lying particle states and the photo-production of isomers . It has already been applied to calculate the absolute amount of strength in <sup>115</sup>In , but up till now it has not been extended to describe and understand the high fragmentation of the strength and the distribution of the $`B(E`$1)$``$, $`B(M`$1)$``$, and $`B(E`$2)$``$ strength in the energy region below 4 MeV.
General ideas about the QPM and its formalism to describe the excited states in odd-mass spherical nuclei with a wave function which includes up to “quasiparticle $``$ two-phonon” configurations are presented in review articles . It is extended here by including “quasiparticle $``$ three-phonon” configurations as well. A Woods-Saxon potential is used in the QPM as an average field for protons and neutrons. Phonons of different multipolarities and parities are obtained by solving the RPA equations with a separable form of the residual interaction including a Bohr-Mottelson form factor. The single-particle spectrum and phonon basis are fixed from calculations in the neighboring even-even nuclear core, i.e. in <sup>116</sup>Sn when <sup>117</sup>Sn nucleus is considered.
In our present calculations the wave functions of the ground state and the excited states are mixtures of different “quasiparticle $``$ $`N`$-phonon” ($`[qpNph]`$) configurations, where $`N`$ =0, 1, 2, 3:
$`\mathrm{\Psi }^\nu (JM)=\{C^\nu (J)\alpha _{JM}^++{\displaystyle \underset{j\beta _1}{}}S_{j\beta _1}^\nu (J)[\alpha _j^+Q_{\beta _1}^+]_{JM}`$ (9)
$`+`$ $`{\displaystyle \underset{j\beta _1\beta _2}{}}{\displaystyle \frac{D_{j\beta _1\beta _2}^\nu (J)[\alpha _j^+Q_{\beta _1}^+Q_{\beta _2}^+]_{JM}}{\sqrt{1+\delta _{\beta _1\beta _2}}}}`$ (10)
$`+`$ $`{\displaystyle \underset{j\beta _1\beta _2\beta _3}{}}{\displaystyle \frac{T_{j\beta _1\beta _2\beta _3}^\nu (J)[\alpha _j^+Q_{\beta _1}^+Q_{\beta _2}^+Q_{\beta _3}^+]_{JM}}{\sqrt{1+\delta _{\beta _1\beta _2}+\delta _{\beta _1\beta _3}+\delta _{\beta _2\beta _3}+2\delta _{\beta _1\beta _2\beta _3}}}}\}_{g.s.}`$ (11)
where the coefficients $`C`$, $`S`$, $`D`$ and $`T`$ describe a contribution of each configuration to a norm of the wave function. We use the following notations $`\alpha ^+`$ and $`Q^+`$ for the coupling between the creation operators of quasiparticles and phonons.
$`[\alpha _j^+Q_{\lambda i}^+]_{JM}={\displaystyle \underset{m\mu }{}}C_{jm\lambda \mu }^{JM}\alpha _{jm}^+Q_{\lambda \mu i}^+,`$
$`[\alpha _j^+Q_{\beta _1}^+Q_{\beta _2}^+Q_{\beta _3}^+]_{JM}={\displaystyle \underset{\lambda _1\lambda _2}{}}[\alpha _j^+[Q_{\beta _1}^+[Q_{\beta _2}^+Q_{\beta _3}^+]_{\lambda _1}]_{\lambda _2}]_{JM},`$
$$[Q_{\lambda _1i_1}^+Q_{\lambda _2i_2}^+]_{\lambda \mu }=\underset{\mu _1\mu _2}{}C_{\lambda _1\mu _1\lambda _2\mu _2}^{\lambda \mu }Q_{\lambda _1\mu _1i_1}^+Q_{\lambda _2\mu _2i_2}^+$$
(12)
where $`C`$ are Clebsh-Gordon coefficients. Quasiparticles are characterized by their shell quantum numbers $`jm|nljm>`$ with a semi-integer value of the total angular momenta $`j`$. They are the result of a Bogoliubov transformation from particle creation (annihilation) $`a_{jm}^+`$ ($`a_{jm}`$) operators:
$$a_{jm}^+=u_j\alpha _{jm}^++(1)^{jm}v_j\alpha _{jm}.$$
(13)
The quasiparticle energy spectrum and the occupation number coefficients $`u_j`$ and $`v_j`$ in Eq. (13) are obtained in the QPM by solving the BCS equations separately for neutrons and protons.
Phonons with quantum numbers $`\beta |\lambda \mu i>`$ are linear superpositions of two-quasiparticle configurations:
$`Q_{\lambda \mu i}^+`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\tau }{\overset{n,p}{}}}{\displaystyle \underset{jj^{}}{}}\{\psi _{jj^{}}^{\lambda i}[\alpha _j^+\alpha _j^{}^+]_{\lambda \mu }`$ (14)
$``$ $`(1)^{\lambda \mu }\phi _{jj^{}}^{\lambda i}[\alpha _j^{}\alpha _j]_{\lambda \mu }\}.`$ (15)
A spectrum of phonon excitations is obtained by solving the RPA equations for each multipolarity $`\lambda `$ which is an integer value. The RPA equations also yield forward (backward) $`\psi _{jj^{}}^{\lambda i}`$ ($`\phi _{jj^{}}^{\lambda i}`$) amplitudes in definition (15):
$$\left(\genfrac{}{}{0pt}{}{\psi }{\phi }\right)_{jj^{}}^{\lambda i}(\tau )=\frac{1}{\sqrt{𝒴_\tau ^{\lambda i}}}\frac{f_{jj^{}}^\lambda (\tau )(u_jv_j^{}+u_j^{}v_j)}{\epsilon _j+\epsilon _j^{}\omega _{\lambda i}}$$
(16)
where $`\epsilon _j`$ is a quasiparticle energy, $`\omega _{\lambda i}`$ is the energy needed for the excitation of an one-phonon configuration, $`f_{jj^{}}^\lambda `$ is a reduced single-particle matrix element of residual forces, and the value $`𝒴_\tau ^{\lambda i}`$ is determined from a normalization condition for the phonon operators:
$$|Q_{\lambda \mu i}Q_{\lambda \mu i}^+|_{ph}=\underset{\tau }{\overset{n,p}{}}\underset{jj^{}}{}\{(\psi _{jj^{}}^{\lambda i})^2(\phi _{jj^{}}^{\lambda i})^2\}=1.$$
(17)
The phonon’s index $`i`$ is used to distinguish between phonon excitations with the same multipolarity but with a difference in energy and structure. The RPA equations yield both, collective- (e.g. $`2_1^+`$ and $`3_1^{}`$), and weakly-collective phonons. The latter correspond to phonons for which some specific two-quasiparticle configuration is dominant in Eq. (15) while for other configurations $`\psi _{jj^{}}^{\lambda i},\phi _{jj^{}}^{\lambda i}0`$.
When the second, third, etc. terms in the wave function of Eq. (10) are considered, phonon excitations of the core couple to a quasiparticle at any level of the average field, not only at the ones with the quantum numbers $`J^\pi `$ as for a pure quasiparticle configuration. It is only necessary that all configurations in Eq. (10) have the same total spin and parity. To achieve a correct position of the $`[qp2ph]`$ configurations, in which we are especially interested in these studies, $`[qp3ph]`$ configurations are important. The excitation energies and the contribution of the different components from the configuration space to the structure of each excited state (i.e. coefficients $`C`$, $`S`$, $`D`$ and $`T`$ in Eq. (10)) are obtained by a diagonalization of the model Hamiltonian on a set of employed wave functions. The coupling matrix elements between the different configurations in the wave functions of Eq. (10) in odd-mass nuclei are calculated on a microscopic footing, making use of the internal fermion structure of the phonons and the model Hamiltonian. For example, the interaction matrix element between the $`[qp1ph]`$ and the $`[qp2ph]`$ configurations has the form (see, Ref. ):
$`<[\alpha _{jm}Q_{\lambda \mu i}]_{JM}|H|[\alpha _{j^{}m^{}}^+[Q_{\lambda _1\mu _1i_1}^+Q_{\lambda _2\mu _2i_2}^+]_{IM^{}}]_{JM}>`$ (18)
$`=\delta _{jj^{}}\delta _{\lambda I}U_{\lambda _1i_1}^{\lambda _2i_2}(\lambda i)()^{j^{}+\lambda +I}2\sqrt{(2j+1)(2I+1)}`$ (19)
$`\times [()^{\lambda _1}\delta _{\lambda \lambda _1}\left\{\begin{array}{ccc}\lambda _2& \lambda _1& I\\ J& j& j^{}\end{array}\right\}\mathrm{\Gamma }(jj^{}\lambda _2i_2)`$ (22)
$`+()^{\lambda _2}\delta _{\lambda \lambda _2}\left\{\begin{array}{ccc}\lambda _1& \lambda _2& I\\ J& j& j^{}\end{array}\right\}\mathrm{\Gamma }(jj^{}\lambda _1i_1)]`$ (25)
where $`H`$ is a model Hamiltonian, $`U_{\lambda _1i_1}^{\lambda _2i_2}(\lambda i)`$ is an interaction matrix element between one- and two-phonon configurations in the neighbouring even-mass nucleus ($`U`$ is a complex function of phonon’s amplitudes $`\psi `$ and $`\phi `$ and $`f_{jj^{}}^\lambda `$; its explicit form can be found in Ref. ) and $`\mathrm{\Gamma }`$ is an interaction matrix element between quasiparticle $`\alpha _{JM}^+`$ and quasiparticle-phonon $`[\alpha _{jm}^+Q_{\lambda \mu i}^+]_{JM}`$ configurations, it is equal to:
$$\mathrm{\Gamma }(Jj\lambda i)=\sqrt{\frac{2\lambda +1}{2J+1}}\frac{f_{Jj}^\lambda (u_Ju_jv_jv_J)}{\sqrt{𝒴_\tau ^{\lambda i}}}.$$
(26)
Equations (25,26) are obtained by applying the exact commutation relations between the phonon and quasiparticle operators:
$`[\alpha _{jm},Q_{\lambda \mu i}^+]_\mathrm{\_}`$ $`=`$ $`{\displaystyle \underset{j^{}m^{}}{}}\psi _{jj^{}}^{\lambda i}C_{jmj^{}m^{}}^{\lambda \mu }\alpha _{j^{}m^{}}^+,`$ (27)
$`[\alpha _{jm}^+,Q_{\lambda \mu i}^+]_\mathrm{\_}`$ $`=`$ $`(1)^{\lambda \mu }{\displaystyle \underset{j^{}m^{}}{}}\phi _{jj^{}}^{\lambda i}C_{jmj^{}m^{}}^{\lambda \mu }\alpha _{j^{}m^{}}.`$ (28)
The exact commutation relations between the phonon operators $`Q_{\lambda \mu i}`$ and $`Q_{\lambda ^{}\mu ^{}i^{}}^+`$
$`[Q_{\lambda \mu i},Q_{\lambda ^{}\mu ^{}i^{}}^+]_\mathrm{\_}=\delta _{\lambda \lambda ^{}}\delta _{\mu \mu ^{}}\delta _{ii^{}}`$ (29)
$``$ $`{\displaystyle \underset{\genfrac{}{}{0pt}{}{jj^{}j_2}{mm^{}m_2}}{}}\alpha _{jm}^+\alpha _{j^{}m^{}}\times \{\psi _{j^{}j_2}^{\lambda i}\psi _{jj_2}^{\lambda ^{}i^{}}C_{j^{}m^{}j_2m_2}^{\lambda \mu }C_{jmj_2m_2}^{\lambda ^{}\mu ^{}}`$ (30)
$``$ $`()^{\lambda +\lambda ^{}+\mu +\mu ^{}}\phi _{jj_2}^{\lambda i}\phi _{j^{}j_2}^{\lambda ^{}i^{}}C_{jmj_2m_2}^{\lambda \mu }C_{j^{}m^{}j_2m_2}^{\lambda ^{}\mu ^{}}\}`$ (31)
are used to calculate the interaction matrix elements $`U`$ in even-even nuclei.
The interaction matrix elements between the $`[qp2ph]`$ and the $`[qp3ph]`$ configurations have a structure similar to (25). We do not provide them here because of their complexity. But even Eq. (25) shows that an unpaired quasiparticle does not behave as a spectator but modifies the interaction between the complex configurations compared to an even-mass nucleus (see second term in this equation). This takes place because the phonons possess an internal fermion structure and the matrix elements $`\mathrm{\Gamma }`$ correspond to an interaction between an unpaired quasiparticle and the two-quasiparticle configurations composing the phonon operator.
It should be pointed out that in the present approach interaction matrix elements are calculated in first order perturbation theory. This means that any $`[qpNph]`$ configuration interacts with the $`[qp(N\pm 1)ph]`$) ones, but its coupling to $`[qp(N\pm 2)ph]`$ configurations is not included in this theoretical treatment. The omitted couplings have non-vanishing interaction matrix elements only in second order perturbation theory. They are much smaller than the ones taken into account and they are excluded from our consideration for technical reasons. An interaction with other $`[qpNph]`$ configurations is taken into account while treating the Pauli principle corrections. In a calculation of the self-energy of the complex configurations we employ a model Hamiltonian written in terms of quasiparticle operators and exact commutation relations (27,31) between quasiparticle and phonon operators. In this case, we obtain a “Pauli shift correction” for the energy of a complex configuration from the sum of the energies of its constituents. Also, when considering complex configurations their internal fermion structure is analyzed and the ones which violate the Pauli principle are excluded from the configuration space. Pauli principle corrections have been treated in a diagonal approximation (see, Ref. for details).
In the actual calculations, the phonon basis includes the phonons with multipolarity and parity $`\lambda ^\pi =1^\pm ,2^+,3^{}`$ and $`4^+`$. Several low-energy phonons of each multipolarity are included in the model space. The most important ones are the first collective $`2^+,3^{}`$ and $`4^+`$ phonons and the ones which form the giant dipole resonance (GDR). Non-collective low-lying phonons of an unnatural parity and natural parity phonons of higher multipolarities are of a marginal importance. To make realistic calculations possible one has to truncate the configuration space. We have done this on the basis of excitation energy arguments. All $`[qp1ph]`$ and $`[qp2ph]`$ with $`E_x6`$ MeV, and $`[qp3ph]`$ with $`E_x8`$ MeV configurations are included in the model space. The only exceptions are $`[J_{g.s.}1^{}]`$ configurations which have not been truncated at all to treat a core polarization effect due to the coupling of low-energy dipole transitions to the GDR on a microscopic level. Thus, for electric dipole transitions we have no renormalized effective charges and used $`e^{\mathrm{eff}}(p)=(N/A)e`$ and $`e^{\mathrm{eff}}(n)=(Z/A)e`$ values to separate the center of mass motion. For $`M1`$ transitions we use $`g_s^{\mathrm{eff}}=0.64g_s^{\mathrm{free}}`$ as recommended in Ref. . By doing this all the important configurations for the description of low-lying states up to 4 MeV are included in the model space. The dimension of this space depends on the total spin of the excited states, and it varies between 500 and 700 configurations.
### B Comparison between experimental data and QPM calculations
Since only $`E1`$, $`M1`$ and $`E2`$ transitions can be observed in the present experiment, the discussion of the properties of the excited states will be restricted to states with $`J^\pi =1/2^\pm ,3/2^\pm `$ and $`5/2^+`$. As the parities of the decaying levels are unknown and the spin could be assigned for only a few levels, the best quantity for the comparison between the theoretical predictions and the experimental results are the total integrated cross sections $`I_s`$. The theoretical reduced excitation probabilities $`B(\pi L`$)$``$ can be transformed into $`I_S`$ values via the following relation:
$$I_S=\frac{8\pi ^3(L+1)}{L[(2L+1)!!]^2}\left(\frac{E_x}{\mathrm{}c}\right)^{2L1}B(\pi L)\frac{\mathrm{\Gamma }_0}{\mathrm{\Gamma }_{tot}},$$
(32)
where $`E_x`$ is the excitation energy, $`L`$ the multipolarity of the transition and $`\mathrm{\Gamma }_0`$ denotes the partial ground state decay width. The obtained $`I_S`$ values for the elastic transitions are plotted in Fig. 4c and compared with the results of our $`(\gamma ,\gamma ^{})`$ experiments given in Fig. 4b. The inelastic decays are accounted for in the total decay widths $`\mathrm{\Gamma }_{tot}`$. Details concerning the calculations and branching ratios will be discussed below. Supporting the experimental findings our calculations also produce a strong fragmentation of the electromagnetic strength. The strongest transitions have $`E1`$ character, but also $`E2`$ and $`M1`$ excitations yield comparable cross sections. The total cross section $`I_S`$ is disentangled into its $`E1`$, $`M1`$ and $`E2`$ components in Fig. 5b,c,d, and compared to the calculated $`I_s`$ values of the core nucleus, <sup>116</sup>Sn (Fig. 5a). The calculated sum of the total cross sections of the plotted $`E1`$, $`M1`$ and $`E2`$ transitions in Fig 5b-d equals 73, 37 and 42 eVb. The summed experimental elastic cross sections, shown in Fig. 4b, equals 133 (22) eVb and agrees within 15% with the theoretically predicted value of 152 eVb.
Although the experimentally observed levels do not match in detail with the calculated level scheme one by one, some interesting general conclusions can be drawn. The most essential differences in the electromagnetic strength distribution over low-lying states in even-even <sup>116,118</sup>Sn and odd-mass <sup>117</sup>Sn take place for the electric dipole transitions. The reason becomes clear by considering which states can be excited from the ground state by $`E1`$ transitions. In the even-even core there is only one $`1^{}`$ configuration with an excitation energy below 4 MeV (thick line with triangle in Fig. 5a). It has a $`[2_1^+3_1^{}]_1^{}`$ two-phonon nature . This is a general feature in heavy semi-magic even-even nuclei . All other $`1^{}`$ configurations have excitation energies more than 1 MeV higher. Therefore, the $`1_1^{}`$ state has an almost pure two-phonon character in semi-magic nuclei. In contrast, there are many $`[qp1ph]`$ and $`[qp2ph]`$ configurations with the same spin and parity close to the two corresponding configurations $`[3s_{1/2}[2_1^+3_1^{}]_1^{}]_{1/2^{},3/2^{}}`$ in <sup>117</sup>Sn. Interactions lead to a strong fragmentation of these two main configurations (see, Table IV). The resulting states are carrying a fraction of the $`E1`$ excitation strength from the ground state.
The predicted properties of some states with spin and parity $`J^\pi =1/2^{}`$ and $`3/2^{}`$ which can be excited from the $`1/2^+`$ ground state in <sup>117</sup>Sn by electric dipole transition are presented in Table IV. A large part of the $`[3s_{1/2}[2_1^+3_1^{}]_1^{}]_{1/2^{},3/2^{}}`$ configurations is concentrated in the 3/2<sup>-</sup> states with an excitation energy of 3.04, 3.55 and 3.56 MeV and in the 1/2<sup>-</sup> states at 3.00 and 3.63 MeV (see, fifth column of this table). These states are marked with triangles in Fig. 5b (as well as four other states with a smaller contribution of these configurations). The $`E1`$ strength distribution among low-lying levels is even more complex because 3/2<sup>-</sup> states at 2.13, 2.33 and 3.93 MeV have a noticeable contribution from the $`3p_{3/2}`$ one-quasiparticle configuration (indicated in the forth column of Table IV) with a large reduced excitation matrix element $`<3p_{3/2}||E1||3s_{1/2}>`$ for which there is no analogue in the even-even core <sup>116</sup>Sn. Also the coupling to $`[3s_{1/2}1_{\text{GDR}}^{}]`$, which treats the core polarization effect, is somewhat different than in the core nucleus, because the blocking effect plays an important role in the interaction with other configurations (see, also Ref. , where only the last type of transitions have been accounted for). The calculated total $`B(E`$1)$``$ strength in the energy region from 2.0 to 4.0 MeV is $`7.210^3`$ e<sup>2</sup>fm<sup>2</sup>. It agrees well with the calculated $`B(E1,0_{g.s.}^+[2^+3^{}]_1^{})=8.210^3`$ e<sup>2</sup>fm<sup>2</sup> in the neighboring <sup>116</sup>Sn nucleus .
The calculations indicate that among the negative parity states in <sup>117</sup>Sn which are relatively strongly excited from the ground state, a few are characterized by a visible $`E1`$-decay into the low-lying $`3/2_1^+`$ state. These are $`3/2^{}`$ states at 2.33, 3.65 and 3.93 MeV and $`1/2^{}`$ state at 3.63 MeV. The state at 2.33 MeV decays into the $`3/2_1^+`$ state due to single-particle transition with a large reduced excitation matrix element $`<2d_{3/2}||E1||3p_{3/2}>`$. The states at higher energies decay into the $`3/2_1^+`$ state because of an admixture of $`[3p_{1/2}[2_1^+3_1^{}]_1^{}]_{1/2^{},3/2^{}}`$ configurations in their wave functions.
Positive parity states in <sup>117</sup>Sn are deexciting to the 1/2<sup>+</sup> ground state by $`M1`$ or $`E2`$ or mixed $`M1/E2`$ transitions. The predicted properties of the $`1/2^+`$, $`3/2^+`$ and $`5/2^+`$ states in <sup>117</sup>Sn are presented in Table V. The $`B(E`$2)$``$ strength distribution is dominated by the excitation of the 3/2<sup>+</sup> state at 1.27 MeV and the 5/2<sup>+</sup> state at 1.49 MeV. The wave functions of these states carry 85% and 60% of the $`[3s_{1/2}2_1^+]`$ configuration, respectively. These two states correspond with a high probability to the experimentally observed levels at 1447 and 1510 keV. A smaller fraction of the above mentioned configuration can be found in the 3/2<sup>+</sup> state at 2.32 MeV (5%) and the 5/2<sup>+</sup> state at 2.23 MeV (6%). The rather fragmented $`E2`$ strength at higher energies (Fig. 5c) is mainly due to $`[3s_{1/2}2_{4,5}^+]`$ configurations which are much less collective than the first one. Fragmented $`E2`$ strength between 2.0 and 4.0 MeV originating from the excitation of the $`2_{4,5}^+`$ phonons has also been observed in NRF experiments on the even-mass <sup>116</sup>Sn nucleus . It could be well reproduced by theoretical calculations (see, thin lines in Fig. 5a). In the odd-mass <sup>117</sup>Sn nucleus the corresponding strength is even more fragmented because of the higher density of the configurations. Nevertheless, these $`E2`$ excitations at high energies contribute appreciably to the reaction cross section, because the $`E2`$ photon scattering cross section is a cubic function of the excitation energy (see, Eq. (32)).
The $`B(M`$1)$``$ strength in the calculations is concentrated mainly above 3.5 MeV as can be seen in Fig. 5d. The wave functions of the 1/2<sup>+</sup> and 3/2<sup>+</sup> states at these energies are very complex. The main configurations, responsible for the $`M1`$ strength, are the $`[2d_{5/2,3/2}2_i^+]`$ ones which are excited because of the internal fermion structure of the phonons (similar to $`E1`$ $`0_{g.s.}^+[2_1^+3_1^{}]_1^{}`$ excitations). They have no analogous transitions in even-even nuclei. The configuration $`[3s_{1/2}1_1^+]`$ has an excitation energy of about 4.2 MeV but its contribution to the structure of states below 4 MeV is rather weak. Most of the states with the largest $`B(M`$1)$``$ values have $`J^\pi =1/2^+`$ (see, Table V).
The QPM calculations show that the two-phonon $`B(E1)`$ strength from the even-even nuclei is fragmented over several states. Even with the present sensitivity of our NRF-setup, it is impossible to resolve all of these details. Nevertheless, when all experimentally observed transitions between 2.7 and 3.6 MeV are considered to be $`E1`$ transitions, the total summed $`B(E1`$)$``$ strength amounts to 5.93 (75) $`10^3e^2fm^2`$ or 91(9)–82(10)% of the two-phonon $`B(E1)`$ strength in the neighbouring nuclei <sup>116</sup>Sn and <sup>118</sup>Sn. This value is considerably higher than in the case of <sup>139</sup>La and <sup>141</sup>Pr where less than 40% was observed. It shows that the $`1/2^+`$ ground state spin of <sup>117</sup>Sn limits the possible fragmentation and hence a larger amount of the particle two-phonon coupled $`B(E1)`$ strength could be resolved in this NRF-experiment.
## V Conclusions
Nuclear resonance fluorescence experiments performed on the odd-mass spherical nucleus <sup>117</sup>Sn revealed a large fragmentation of the electromagnetic strength below an excitation energy of 4 MeV. The search for the fragments of the $`3s_{1/2}[2_1^+3_1^{}]`$ multiplet carrying the $`B(E1)`$ strength of the adjacent even-even nuclei is complicated by the limited spin information. QPM calculations carried out for the first time in a complete configuration space can explain the fragmentation of the excitation strength and shed light on how the $`B(E1)`$, $`B(M1)`$ and $`B(E2)`$ strength is distributed over this energy region.
## Acknowledgements
This work is part of the Research program of the Fund for Scientific Research Flanders. The support by the Deutsche Forschungsgemeinschaft (DFG) under contracts Kn 154-30 and Br 799/9-1 is gratefully acknowledged. V. Yu. P. acknowledges a financial support form the Research Council of the University of Gent and NATO.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
|
warning/0002/hep-th0002245.html
|
ar5iv
|
text
|
# Contents
## 1 Conventions and notation
##### Spacetime.
Greek indices from the middle of the alphabet generally are spacetime indices, $`\mu =0,1,\mathrm{},n1`$. We work in $`n`$-dimensional Minkowskian space with metric $`\eta _{\mu \nu }=\mathrm{diag}(1,+1,\mathrm{},+1)`$. The spacetime coordinates are denoted by $`x^\mu `$. The Levi-Civita tensor $`ϵ^{\mu _1\mathrm{}\mu _n}`$ is completely antisymmetric in all its indices with $`ϵ^{01\mathrm{}n1}=1`$. Its indices are lowered with the metric $`\eta _{\mu \nu }`$. The differentials $`dx^\mu `$ anticommute, the volume element is denoted by $`d^nx`$,
$$dx^\mu dx^\nu =dx^\mu dx^\nu =dx^\nu dx^\mu ,d^nx=dx^0\mathrm{}dx^{n1}.$$
##### Gauge theories of the Yang-Mills type.
The gauge group is denoted by $`G`$, its Lie algebra by $`𝒢`$. Capital Latin indices from the middle of the alphabet $`I,J,\mathrm{}`$ generally refer to a basis for $`𝒢`$. The structure constants of the Lie algebra in that basis are denoted by $`f_{IJ}^{}{}_{}{}^{K}`$.
The gauge coupling constant(s) are denoted by $`e`$ and are explicitly displayed in the formulae. However, in most formulae we use a collective notation which does not distinguish the different gauge coupling constants when the gauge group is the direct product of several (abelian or simple) factors.
The covariant derivative is defined by $`D_\mu =_\mu eA_\mu ^I\rho (e_I)`$ where $`\{e_I\}`$ is a basis of $`𝒢`$ and $`\rho `$ a representation of $`𝒢`$. The field strength is defined by $`F_{\mu \nu }^I=_\mu A_\nu ^I_\nu A_\mu ^I+ef_{JK}^{}{}_{}{}^{I}A_\mu ^JA_\nu ^K`$, the corresponding 2-form by $`F^I=\frac{1}{2}F_{\mu \nu }^Idx^\mu dx^\nu `$.
##### General.
The Einstein summation convention over repeated upper or lower indices generally applies.
Complete symmetrization is denoted by ordinary brackets $`(\mathrm{})`$, complete antisymmetrization by square brackets $`[\mathrm{}]`$ including the normalization factor:
$$M_{(\mu _1\mathrm{}\mu _k)}=\frac{1}{k!}\underset{\sigma 𝒮^k}{}M_{\mu _{\sigma (1)}\mathrm{}\mu _{\sigma (k)}},M_{[\mu _1\mathrm{}\mu _k]}=\frac{1}{k!}\underset{\sigma 𝒮^k}{}()^\sigma M_{\mu _{\sigma (1)}\mathrm{}\mu _{\sigma (k)}},$$
where the sums run over all elements $`\sigma `$ of the permutation group $`𝒮^k`$ of $`k`$ objects and $`()^\sigma `$ is $`1`$ for an even and $`1`$ for an odd permutation.
Dependence of a function $`f`$ on a set of fields $`\varphi ^i`$ and a finite number of their derivatives is collectively denoted by $`f([\varphi ])`$,
$$f([\varphi ])f(\varphi ^i,_\mu \varphi ^i,\mathrm{},_{(\mu _1\mathrm{}\mu _r)}\varphi ^i).$$
All our derivatives are left derivatives.
The antifield of a field $`\varphi ^i`$ is denoted by $`\varphi _i^{}`$; for instance the antifield corresponding to the Yang-Mills gauge potential $`A_\mu ^I`$ is $`A_I^\mu `$.
The Hodge dual of a $`p`$-form $`\omega `$ is denoted by $`\omega `$,
$$\omega =\frac{1}{p!}dx^{\mu _1}\mathrm{}dx^{\mu _p}\omega _{\mu _1\mathrm{}\mu _p},\omega =\frac{1}{p!(np)!}dx^{\mu _1}\mathrm{}dx^{\mu _{np}}ϵ_{\mu _1\mathrm{}\mu _n}\omega ^{\mu _{np+1}\mathrm{}\mu _n}.$$
## 2 Introduction
### 2.1 Purpose of report
Gauge symmetries underlie all known fundamental interactions. While the existence of the gravitational force can be viewed as a consequence of the invariance of the laws of physics under arbitrary spacetime diffeomorphisms, the non-gravitational interactions are dictated by the invariance under an internal non-Abelian gauge symmetry.
It has been appreciated in the last twenty years or so that many physical questions concerning local gauge theories can be powerfully reformulated in terms of local BRST cohomology. The BRST differential was initially introduced in the context of perturbative quantum Yang-Mills theory in four dimensions . One of the aims was to relate the Slavnov-Taylor identities underlying the proof of power-counting renormalizability to an invariance of the gauge-fixed action (for a recent historical account, see ). However, it was quickly realized that the scope of BRST theory is much wider. It can not only be formulated for any theory with a gauge freedom, but also it is quite useful at a purely classical level.
The purpose of this report is to discuss in detail the local BRST cohomology for gauge theories of the Yang-Mills type. We do so by emphasizing whenever possible the general properties of the BRST cohomology that remain valid in other contexts. At the end of the report, we give some references to papers where the local BRST cohomology is computed for other gauge theories by means of similar techniques.
### 2.2 Gauge theories of the Yang-Mills type
We first define the BRST differential in the Yang-Mills context. The Yang-Mills gauge potential is a one-form, which we denote by $`A^I=dx^\mu A_\mu ^I`$. The gauge group can be any finite dimensional group of the form $`G=G_0\times G_1`$, where $`G_0`$ is Abelian and $`G_1`$ is semi-simple. In practice, $`G`$ is compact so that $`G_0`$ is a product of $`U(1)`$ factors while $`G_1`$ is compact and semi-simple. This makes the standard Yang-Mills kinetic term definite positive. However, it will not be necessary to make this assumption for the cohomological calculation. This is important as non-compact gauge groups arise in gravity or supergravity. The Lie algebra of the gauge group is denoted by $`𝒢`$.
The matter fields are denoted by $`\psi ^i`$ and can be bosonic or fermionic. They are assumed to transform linearly under the gauge group according to a completely reducible representation. The corresponding representation matrices of $`𝒢`$ are denoted by $`T_I`$ and the structure constants of $`𝒢`$ in that basis are written $`f_{IJ}^{}{}_{}{}^{K}`$,
$$[T_I,T_J]=f_{IJ}^{}{}_{}{}^{K}T_K.$$
(2.1)
The field strengths and the covariant derivatives of the matter fields are denoted by $`F_{\mu \nu }^I`$ and $`D_\mu \psi ^i`$ respectively,
$`F_{\mu \nu }^I`$ $`=`$ $`_\mu A_\nu ^I_\nu A_\mu ^I+ef_{JK}^{}{}_{}{}^{I}A_\mu ^JA_\nu ^K,`$ (2.2)
$`D_\mu \psi ^i`$ $`=`$ $`_\mu \psi ^i+eA_\mu ^IT_{Ij}^i\psi ^j.`$ (2.3)
Here $`e`$ denotes the gauge coupling constant(s) (one for each simple or abelian factor).
The (infinitesimal) gauge transformations read
$$\delta _ϵA_\mu ^I=D_\mu ϵ^I,\delta _ϵ\psi ^i=eϵ^IT_{Ij}^i\psi ^j,$$
(2.4)
where the $`ϵ^I`$ are the ”gauge parameters” and
$$D_\mu ϵ^I=_\mu ϵ^I+ef_{JK}^{}{}_{}{}^{I}A_\mu ^Jϵ^K.$$
(2.5)
In the BRST formalism, the gauge parameters are replaced by anticommuting fields $`C^I`$; these are the “ghost fields” of .
The Lagrangian $`L`$ is a function of the fields and their derivatives up to a finite order (“local function”),
$$L=L([A_\mu ^I],[\psi ^i])$$
(2.6)
(see section 1 for our notation and conventions). It is invariant under the gauge transformations (2.4) up to a total derivative, $`\delta _ϵL=_\mu k^\mu `$ for some $`k^\mu `$ that may be zero. The detailed form of the Lagrangian is left open at this stage except that we assume that the matter sector does not carry a gauge-invariance of its own, so that (2.4) are the only gauge symmetries. This requirement is made for definiteness. The matter fields could carry further gauge symmetries (e.g., $`p`$-form gauge symmetries) which would bring in further ghosts; these could be discussed along the same lines but for definiteness and simplicity, we exclude this possibility.
A theory with the above field content and gauge symmetries is said to be of the “Yang-Mills type”. Specific forms of the Lagrangian are $`L=(1/4)\delta _{IJ}F_{\mu \nu }^IF^{J\mu \nu }`$, for which $`\delta _ϵL=0`$ (Yang-Mills original theory ) or $`d^3xL=\mathrm{Tr}(AdA+(e/3)A^3)`$ which is invariant only up to a non-vanishing surface term, $`\delta _ϵL=_\mu k^\mu `$ with $`k^\mu 0`$ (Chern-Simons theory in $`3`$ dimensions ).
We shall also consider “effective Yang-Mills theories” for which the Lagrangian contains all possible terms compatible with gauge invariance and thus involves derivatives of arbitrarily high order.
In physical applications one usually assumes, of course, that $`L`$ is in addition Lorentz or Poincaré invariant. The cohomological considerations actually go through without this assumption.
The BRST differential acts in an enlarged space that contains not only the original fields and the ghosts, but also sources for the BRST variations of the fields and the ghosts. These sources are denoted by $`A_I^\mu `$, $`\psi _i^{}`$ and $`C_I^{}`$ respectively, and have Grassmann parity opposite to the one of the corresponding fields. They have been introduced in order to control how the BRST symmetry gets renormalized . They play a crucial role in the BV construction where they are known as the antifields; for this reason, they will be indifferently called antifields or BRST sources here.
The BRST-differential decomposes into the sum of two differentials
$$s=\delta +\gamma $$
(2.7)
with $`\delta `$ and $`\gamma `$ acting as
$$\begin{array}{ccc}Z& \delta Z& \gamma Z\\ & & \\ \text{}A_\mu ^I& 0& D_\mu C^I\\ \text{}\psi ^i& 0& eC^IT_{Ij}^i\psi ^j\\ \text{}C^I& 0& \frac{1}{2}ef_{KJ}^{}{}_{}{}^{I}C^JC^K\\ \text{}C_I^{}& D_\mu A_I^\mu e\psi _i^{}T_{Ij}^i\psi ^j& ef_{JI}^{}{}_{}{}^{K}C^JC_K^{}\\ \text{}A_I^\mu & L_I^\mu & ef_{JI}^{}{}_{}{}^{K}C^JA_K^\mu \\ \text{}\psi _i^{}& L_i& eC^I\psi _j^{}T_{Ii}^j\end{array}$$
(2.8)
where
$`D_\mu C^I`$ $`=`$ $`_\mu C^I+ef_{JK}^{}{}_{}{}^{I}A_\mu ^JC^K`$ (2.9)
$`D_\mu A_I^\mu `$ $`=`$ $`_\mu A_I^\mu ef_{JI}^{}{}_{}{}^{K}A_\mu ^JA_K^\mu `$ (2.10)
$`L_I^\mu `$ $`=`$ $`{\displaystyle \frac{\delta L}{\delta A_\mu ^I}},L_i=()^{ϵ_i}{\displaystyle \frac{\delta L}{\delta \psi ^i}}.`$ (2.11)
Here $`\delta L/\delta \varphi `$ denotes the Euler-Lagrange derivative of $`L`$ with respect to $`\varphi `$ (taken from the left), and $`ϵ_i`$ is the Grassmann parity of $`\psi ^i`$ ($`ϵ_i=0`$ for a boson, $`ϵ_i=1`$ for a fermion). The parity dependent sign in $`L_i`$ originates from the convention that $`\delta `$ is defined through Euler-Lagrange right-derivatives of $`L`$.
$`\delta `$, $`\gamma `$ (and thus $`s`$) are extended to the derivatives of the variables by the rules $`\delta _\mu =_\mu \delta `$, $`\gamma _\mu =_\mu \gamma `$. These rules imply $`\delta d+d\delta =\gamma d+d\gamma =sd+ds=0`$, where $`d`$ is the exterior spacetime derivative $`d=dx^\mu _\mu `$, because $`dx^\mu `$ is odd. Furthermore, $`\delta `$, $`\gamma `$, $`s`$ and $`d`$ act as left (anti-)derivations, e.g., $`\delta (ab)=(\delta a)b+(1)^{ϵ_a}a\delta b`$, where $`ϵ_a`$ is the Grassmann parity of $`a`$.
The BRST differential is usually not presented in the above manner in the literature; we shall make contact with the more familiar formulation in the appendix 2.A below. However, we point out already that on the fields $`A_\mu ^I`$, $`\psi ^i`$ and the ghosts $`C^I`$, $`s`$ reduces to $`\gamma `$ and the BRST transformation takes the familiar form of a “gauge transformation in which the gauge parameters are replaced by the ghosts” (for the fields $`A_\mu ^I`$ and $`\psi ^i`$), combined with $`sC^I=\frac{1}{2}ef_{KJ}^{}{}_{}{}^{I}C^JC^K`$ in order to achieve nilpotency. The differential $`\delta `$, known as the Koszul-Tate differential, acts non-trivially only on the antifields. It turns out to play an equally important rôle in the formalism.
The decomposition (2.7) is related to the various gradings that one introduces in the algebra generated by the fields, ghosts and antifields. The gradings are the pure ghost number $`\mathrm{𝑝𝑢𝑟𝑒𝑔ℎ}`$, the antifield number $`\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}`$ and the (total) ghost number $`\mathrm{𝑔ℎ}`$. They are not independent but related through
$$\mathrm{𝑔ℎ}=\mathrm{𝑝𝑢𝑟𝑒𝑔ℎ}\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}.$$
(2.12)
The antifield number is also known as “antighost number”. The gradings of the basic variables are given by
$$\begin{array}{cccc}Z& \mathrm{𝑝𝑢𝑟𝑒𝑔ℎ}(Z)& \mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(Z)& \mathrm{𝑔ℎ}(Z)\\ & & & \\ \text{}A_\mu ^I& 0& 0& 0\\ \text{}\psi ^i& 0& 0& 0\\ \text{}C^I& 1& 0& 1\\ \text{}C_I^{}& 0& 2& 2\\ \text{}A_I^\mu & 0& 1& 1\\ \text{}\psi _i^{}& 0& 1& 1\end{array}$$
(2.13)
The BRST differential and the differentials $`\delta `$ and $`\gamma `$ all increase the ghost number by one unit. The differential $`\delta `$ does so by decreasing the antifield number by one unit while leaving the pure ghost number unchanged, while $`\gamma `$ does so by increasing the pure ghost number by one unit while leaving the antifield number unchanged. Thus, one has $`\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(\delta )=1`$ and $`\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(\gamma )=0`$. The decomposition (2.7) corresponds to an expansion of $`s`$ according to the antifield number, $`s=_{k1}s_k`$, with $`\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(s_k)=k`$, $`s_1=\delta `$ and $`s_0=\gamma `$. For Yang-Mills gauge models, the expansion stops at $`\gamma `$. For generic gauge theories, in particular theories with open algebras, there are higher-order terms $`s_k`$, $`k1`$ (when the gauge algebra closes off-shell, $`s_1,s_2,\mathrm{}`$ vanish on the fields, but not necessarily on the antifields).
### 2.3 Relevance of BRST cohomology
As stated above, the BRST symmetry provides an extremely efficient tool for investigating many aspects of a gauge theory. We review in this subsection the contexts in which it is useful. We shall only list the physical questions connected with BRST cohomological groups without making explicitly the connection here. This connection can be found in the references listed in the course of the discussion. \[Some comments on the link between the BRST symmetry and perturbative renormalization are surveyed – very briefly – in appendix 2.A below to indicate the connection with the gauge-fixed formulation in which these questions arose first.\]
A crucial feature of the BRST differential $`s`$ is that it is a differential, i.e., it squares to zero,
$$s^2=0.$$
(2.14)
In Yang-Mills type theories (and in other gauge theories for which $`s=\delta +\gamma `$) this is equivalent to the fact that $`\delta `$ and $`\gamma `$ are differentials that anticommute,
$$\delta ^2=0,\delta \gamma +\gamma \delta =0,\gamma ^2=0.$$
(2.15)
The BRST cohomology is defined as follows. First, the BRST cocycles $`A`$ are objects that are “BRST -closed”, i.e., in the kernel of $`s`$,
$$sA=0.$$
(2.16)
Because $`s`$ squares to zero, the BRST-coboundaries, i.e., the objects that are BRST-exact (in the image of $`s`$) are automatically closed,
$$A=sBsA=0.$$
(2.17)
The BRST-cohomology is defined as the quotient space Ker$`s/`$Im$`s`$,
$$H(s)=\frac{\text{Ker}s}{\text{Im}s}.$$
(2.18)
An element in $`H(s)`$ is an equivalence class of BRST-cocycles, where two BRST-cocycles are identified if they differ by a BRST-coboundary. One defines similarly $`H(\delta )`$ and $`H(\gamma )`$.
One can consider the BRST cohomology in the space of local functions or in the space of local functionals. In the first instance, the “cochains” are local functions, i.e., are functions of the fields, the ghosts, the antifields and a finite number of their derivatives. In the second instance, the cochains are local functionals i.e., integrals of local volume-forms, $`A=a`$, $`a=fd^nx`$, where $`f`$ is a local function in the above sense ($`n`$ denotes the spacetime dimension). When rewritten in terms of the integrands (see subsection 4.4), the BRST cocycle and coboundary conditions $`sa=0`$ and $`a=sb`$ become respectively
$`sa+dm=0`$ $`\text{(cocycle condition)},`$ (2.19)
$`a=sb+dn`$ $`\text{(coboundary condition)},`$ (2.20)
for some local $`(n1)`$-forms $`m`$ and $`n`$. For this reason, we shall reserve the notation $`H(s)`$ for the BRST cohomology in the space of local functions and denote by $`H(s|d)`$ the BRST cohomology in the space of local functionals. So $`H(s|d)`$ is defined by (2.19) and (2.20), while $`H(s)`$ is defined by
$`sa=0`$ $`\text{(cocycle condition)},`$ (2.21)
$`a=sb`$ $`\text{(coboundary condition)}.`$ (2.22)
In both cases, $`a`$ and $`b`$ (and $`m,n`$) are local forms.
One may of course fix the ghost number and consider cochains with definite ghost number. At ghost number $`j`$, the corresponding groups are denoted $`H^j(s)`$ and $`H^{j,n}(s|d)`$ respectively, where in the latter instance the cochains have form-degree equal to the spacetime dimension $`n`$. The calculation of $`H^{j,n}(s|d)`$ is more complicated than that of $`H^j(s)`$ because one must allow for the possibility to integrate by parts. It is useful to consider the problem defined by (2.19), (2.20) for other values $`p`$ of the form-degree. The corresponding cohomology groups are denoted by $`H^{j,p}(s|d)`$.
Both cohomologies $`H(s)`$ and $`H(s|d)`$ capture important physical information about the system. The reason that the BRST symmetry is important at the quantum level is that the standard method for quantizing a gauge theory begins with fixing the gauge. The BRST symmetry and its cohomology become then substitutes for gauge invariance, which would be otherwise obscure. The realization that BRST theory is also useful at the classical level is more recent. We start by listing the quantum questions for which the local BRST cohomology is relevant (points 1 though 4 below). We then list the classical ones (points 5 and 6).
1. Gauge anomalies: $`H^{1,n}(s|d)`$.
Before even considering whether a quantum gauge field theory is renormalizable, one must check whether the gauge symmetry is anomaly-free. Indeed, quantum violations of the classical gauge symmetry presumably spoil unitarity and probably render the quantum theory inconsistent. As shown in , gauge anomalies $`𝒜=a`$ are ghost number one local functionals constrained by the cocycle condition $`s𝒜=0`$, i.e.,
$$sa+dm=0,$$
(2.23)
which is the BRST generalization of the Wess-Zumino consistency condition . Furthermore, trivial solutions can be removed by adding local counterterms to the action. Thus, the cohomology group $`H^{1,n}(s|d)`$ characterizes completely the form of the non trivial gauge anomalies. Once $`H^{1,n}(s|d)`$ is known, only the coefficients of the candidate anomalies need to be determined.
The computation of $`H^{1,n}(s|d)`$ was started in and completed in the antifield-independent case in . Some aspects of solutions with antifields included have been discussed in ; the general solution was worked out more recently in .
2. Renormalization of Yang-Mills gauge models: $`H^{0,n}(s|d)`$.
Yang-Mills theory in four dimensions is power-counting renormalizable . However, if one includes interactions of higher mass dimensions, or considers the Yang-Mills Lagrangian in higher spacetime dimensions, one loses power-counting renormalizability. To consistently deal with these theories, the effective theory viewpoint is necessary (for recent reviews, see ). The question arises then as to whether these theories are renormalizable in the “modern sense”, i.e., whether the gauge symmetry constrains the divergences sufficiently, so that these can be absorbed by gauge-invariant counterterms at each order of perturbation theory . This is necessary for dealing meaningfully with loops – and not just tree diagrams. This question can again be formulated in terms of BRST cohomology. Indeed, the divergences and counterterms are constrained by the cocycle condition (2.23) but this time at ghost number zero; and trivial solutions can be absorbed through field redefinitions .
Thus it is $`H^{0,n}(s|d)`$ that controls the counterterms. The question raised above can be translated, in cohomological terms, as to whether the most general solution of the consistency condition $`sa+dm=0`$ at ghost number zero can be written as $`a=hd^nx+sb+dn`$ where $`h`$ is a local function which is off-shell gauge invariant up to a total derivative. The complete answer to this question, which is affirmative when the gauge-group is semi-simple, has been worked out recently .
3. Renormalization of composite, gauge-invariant operators: $`H^0(s)`$ and $`H^{0,n}(s|d)`$.
The renormalization of composite gauge-invariant operators arises in the analysis of the operator product expansion, relevant, in particular, to deep inelastic scattering. In the mid-seventies, it was conjectured that gauge-invariant operators can only mix, upon renormalization, with gauge-invariant operators or with gauge-variant operators which vanish in physical matrix elements (see also ).
As established also there, the conjecture is equivalent to proving that in each BRST cohomological class at ghost number zero, one can choose a representative that is strictly gauge-invariant. For local operators involving the variables and their derivatives (at a given, unspecified, point) up to some finite order, the relevant cohomology is $`H^0(s)`$. The problem is to show that the most general solution of $`sa=0`$ at ghost number zero is of the form $`a=I+sb`$, where $`I`$ is an invariant function of the curvatures $`F_{\mu \nu }^I`$, the fields $`\psi ^i`$ and a finite number of their covariant derivatives. For operators of low mass dimension, the problem involves a small number of possible composite fields and can be analyzed rather directly. However, power counting becomes a less constraining tool for operators of high mass dimension, since the number of composite fields with the appropriate dimensions proliferates as the dimension increases. One must use cohomological techniques that do not rely on power counting. The conjecture turns out to be correct and was proved first (in four spacetime dimensions) in . The proof was streamlined in using the above crucial decomposition of $`s`$ into $`\delta +\gamma `$. Further information on this topic may be found in . Recent applications are given in .
In the same way, the cohomological group $`H^{0,n}(s|d)`$ controls the renormalization of integrated gauge-invariant operators $`d^nxO(x)`$, or, as one also says, operators at zero momentum. As mentioned above, its complete resolution has only been given recently .
4. Anomalies for gauge invariant operators: $`H^1(s)`$ and $`H^{1,n}(s|d)`$.
The question of mixing of gauge invariant operators as discussed in the previous paragraph may be obstructed by anomalies if a non gauge invariant regularization and renormalization scheme is or has to be used . To lowest order, these anomalies are ghost number $`1`$ local functions $`a`$ and have to satisfy the consistency condition $`sa=0`$. BRST-exact anomalies are trivial because they can be absorbed through the addition of non BRST invariant operators, which compensate for the non invariance of the scheme. This means that the cohomology constraining these obstructions is $`H^1(s)`$. Similarly, the group $`H^{1,n}(s|d)`$ controls the anomalies in the renormalization of integrated gauge invariant composite operators .
5. Generalized conservation laws - “Characteristic cohomology” and BRST cohomology in negative ghost number
The previous considerations were quantum. The BRST cohomology captures also important classical information about the system. For instance, it has been proved in that the BRST cohomology at negative ghost number is isomorphic to the so-called “characteristic cohomology” , which generalizes the familiar notion of (non-trivial) conserved currents. A conserved current can be defined (in Hodge dual terms) as an $`(n1)`$-form that is $`d`$-closed modulo the equations of motion,
$$dj0$$
(2.24)
where $``$ means “equal when the equations of motion hold”. One says that a conserved current is (mathematically) trivial when it is on-shell equal to an exact form (see e.g. ),
$$jdm,$$
(2.25)
where $`m`$ is a local form. The characteristic cohomology in form degree $`n1`$ is by definition the quotient space of conserved currents by trivial ones. The characteristic cohomology in arbitrary form degree is defined by the same equations, taken at the relevant form degrees. The characteristic cohomology in form-degree $`n2`$ plays a rôle in the concept of “charge without charge” .
The exact correspondence between the BRST cohomology and the characteristic cohomology is as follows : the characteristic cohomology $`H_{char}^{nk}`$ in form-degree $`nk`$ is isomorphic to the BRST cohomology $`H^{k,n}(s|d)`$ in ghost number $`k`$ and maximum form-degree $`n`$.<sup>1</sup><sup>1</sup>1The isomorphism assumes the De Rham cohomology of the spacetime manifold to be trivial. Otherwise, the statement needs to be refined along the lines of . A proof of the isomorphism is given in sections 6 and 7. Cocycles of the cohomology $`H^{k,n}(s|d)`$ involve necessarily the antifields since these are the only variables with negative ghost number.
The reformulation of the characteristic cohomology in terms of BRST cohomology is particularly useful in form degree $`<n1`$, where it has yielded new results leading to a complete calculation of $`H_{char}^{nk}`$ for $`k>1`$. It is also useful in form-degree $`n1`$, where it enables one to work out the explicit form of all the conserved currents that are not gauge-invariant (and cannot be invariantized by adding trivial terms) . The calculation of the gauge-invariant currents is more complicated and depends on the specific choice of the Lagrangian, contrary to the characteristic cohomology in lower form-degree.
6. Consistent interactions: $`H^{0,n}(s_0|d)`$, $`H^{1,n}(s_0|d)`$ \- Uniqueness of Yang-Mills cubic vertex
Is is generally believed that the only way to make a set of massless vector fields consistently interact is through the Yang-Mills construction - apart from interactions that do not deform the abelian gauge transformations and involve only the abelian curvatures or abelian Chern-Simons terms. Partial proofs of this result exist but these always make implicitly some restrictive assumptions on the number of derivatives involved in the coupling or the polynomial degree of the interaction. In fact, a counterexample exists in three spacetime dimensions, which generalizes the Freedman-Townsend model for two-forms in four dimensions .
The problem of constructing consistent (local) interactions for a gauge field theory has been formulated in general terms in . It turns out that this formulation has in fact a natural interpretation in terms of deformation theory and involves the computation of the free BRST cohomologies $`H^{0,n}(s_0|d)`$ and $`H^{1,n}(s_0|d)`$ (see also ), where $`s_0`$ is the free BRST differential. The BRST point of view systematizes the search for consistent interactions and the demonstration of their uniqueness up to field redefinitions.
We present in this report complete results on $`H^k(s)`$. We also provide, for a very general class of Lagrangians, a complete description of the cohomological groups $`H(s|d)`$ in terms of the non trivial conserved currents that the model may have. So, all these cohomology groups are known once all the global symmetries of the theory have been determined – a problem that depends on the Lagrangian. Furthermore, we show that the cocycles in the cohomological groups $`H^{0,n}(s|d)`$ (counterterms) and $`H^{1,n}(s|d)`$ (anomalies) may be chosen not to involve the conserved currents when the Yang-Mills gauge group has no abelian factor (in contrast to the groups $`H^{g,n}(s|d)`$ for other values of $`g`$). So, in this case, we can also work out completely $`H^{0,n}(s|d)`$ and $`H^{1,n}(s|d)`$, without specifying $`L`$. We also present complete results for $`H^{k,n}(s|d)`$ (with $`k>1`$) as well as partial results for $`H^{1,n}(s|d)`$. Finally, we establish the uniqueness of the Yang-Mills cubic vertex in four spacetime dimensions, using a result of Torre .
### 2.4 Cohomology and antifields
The cohomological investigation of the BRST symmetry was initiated as early as in the seminal papers , which gave birth to the modern algebraic approach to the renormalization of gauge theories (for a recent monograph on the subject, see ; see also ). Many results on the antifield-independent cohomology were established in the following fifteen years. However, the antifield-dependent case remained largely unsolved and almost not treated at all although a complete answer to the physical questions listed above requires one to tackle the BRST cohomology without a priori restrictions on the antifield dependence.
In order to deal efficiently with the antifields in the BRST cohomology, a new qualitative ingredient is necessary. This new ingredient is the understanding that the antifields are algebraically associated with the equations of motion in a well-defined fashion, which is in fact quite standard in cohomology theory. With this novel interpretation of the antifields, new progress could be made and previous open conjectures could be proved.
Thus, while the original point of view on the antifields (sources coupled to the BRST variations of the fields ) is useful for the purposes of renormalization theory, the complementary interpretation in terms of equations of motion is quite crucial for cohomological calculations. Because this interpretation of the antifields, related to the so-called “Koszul-Tate complex”, plays a central rôle in our approach, we shall devote two entire sections to explaining it (sections 5 and 6).
The relevant interpretation of the antifields originates from work on the Hamiltonian formulation of the BRST symmetry, developed by the Fradkin school , where a similar interpretation can be given for the momenta conjugate to the ghosts . The reference deals in particular with the case of reducible constraints, which is the closest to the Lagrangian case from the algebraic point of view. The extension of the work on the Hamiltonian Koszul-Tate complex to the antifield formalism was carried out in . Locality was analyzed in .
What enabled one to identify the Koszul-Tate differential as a key building block was the attempt to generalize the BRST construction to more general settings, in which the algebra of the gauge transformations closes only “on-shell”. In that case, it is crucial to introduce the Koszul-Tate differential from the outset; the BRST differential is then given by $`s=\delta +\gamma +\mathrm{`}\mathrm{`}\text{more}\mathrm{"}`$, where “more” involves derivations of higher antifield number. Such a generalization was first uncovered in the case of supergravity . The construction was then systematized in and given its present form in . However, even when the algebra closes off-shell, the Koszul-Tate differential is a key ingredient for cohomological purposes.
### 2.5 Further comments
The antifield formalism is extremely rich and we shall exclusively be concerned here with its cohomological aspects in the context of local gauge theories. So, many of its properties (meaning of antibracket and geometric interpretation of the master equation , anti-BRST symmetry with antifields , quantum master equation and its regularization ) will not be addressed. A review with the emphasis on the algebraic interpretation of the antifields used here is . Other reviews are .
It would be impossible - and out of place - to list here all references dealing with one aspect or the other of the antifield formalism; we shall thus quote only some papers from the last years which appear to be representative of the general trends. These are (geometric aspects of the antifield formalism), ($`Sp(2)`$-formalism), (quantum antibrackets), (higher antibrackets). We are fully aware that this list is incomplete but we hope that the interested reader can work her/his way through the literature from these references.
Finally, a monograph dealing with aspects of anomalies complementary to those discussed here is .
### 2.6 Appendix 2.A: Gauge-fixing and antighosts
The BRST differential as we have introduced it is manifestly gauge-independent since nowhere in the definitions did we ever fix the gauge. That this is the relevant differential for the classical questions described above (classical deformations of the action, conservation laws) has been established in . Perhaps less obvious is the fact that this is also the relevant differential for the quantum questions. Indeed, the BRST differential is usually introduced in the quantum context only after the gauge has been fixed and one may wonder what is the connection between the above definitions and the more usual ones.
A related aspect is that we have not included the antighosts. There is in fact a good reason for this, because these do not enter the cohomology: they only occur through “trivial” terms because, as one says, they are in the contractible part of the algebra .
To explain both issues, we first recall the usual derivation of the BRST symmetry. First, one fixes the gauge through gauge conditions
$$^I=0,$$
(2.26)
where the $`^I`$ involve the gauge potential, the matter fields and their derivatives. For instance, one may take $`^I=^\mu A_\mu ^I`$ (Lorentz gauge). Next, one writes down the gauge-fixed action
$$S_F=S^{\mathrm{𝑖𝑛𝑣}}+S^{\mathrm{𝑔𝑓}}+S^{\mathrm{𝑔ℎ𝑜𝑠𝑡}}$$
(2.27)
where $`S^{\mathrm{𝑖𝑛𝑣}}=d^nxL`$ is the original gauge-invariant action, $`S^{gf}`$ is the gauge-fixing term
$$S^{\mathrm{𝑔𝑓}}=d^nxb_I(^I+\frac{1}{2}b^I)$$
(2.28)
and $`S^{\mathrm{𝑔ℎ𝑜𝑠𝑡}}`$ is the ghost action,
$$S^{\mathrm{𝑔ℎ𝑜𝑠𝑡}}=\mathrm{i}d^nx\left[D_\mu C^J\frac{\delta (\overline{C}_I^I)}{\delta A_\mu ^J}eC^JT_{Jj}^i\psi ^j\frac{\delta (\overline{C}_I^I)}{\delta \psi ^i}\right].$$
(2.29)
The $`b^I`$’s are known as the auxiliary fields, while the $`\overline{C}_I`$ are the antighosts. We take both the ghosts and the antighosts to be real.
The gauge-fixed action is invariant under the BRST symmetry $`\sigma A_\mu ^I=D_\mu C^I`$, $`\sigma \psi ^i=eC^IT_{Ij}^i\psi ^j`$, $`\sigma C^I=(1/2)ef_{KJ}^{}{}_{}{}^{I}C^JC^K`$, $`\sigma \overline{C}_I=\mathrm{i}b_I`$ and $`\sigma b_I=0`$. This follows from the gauge-invariance of the original action as well as from $`\sigma ^2=0`$. $`\sigma `$ coincides with $`s`$ on $`A_\mu ^I`$, $`\psi ^i`$ and $`C^I`$ but we use a different letter to avoid confusion. One can view the auxiliary fields $`b^I`$ as the ghosts for the (abelian) gauge shift symmetry $`\overline{C}_I\overline{C}_I+ϵ_I`$ under which the original Lagrangian is of course invariant since it does not depend on the antighosts.
To derive the Ward-Slavnov-Taylor identities associated with the original gauge symmetry and this additional shift symmetry, one introduces sources for the BRST variations of all the variables, including the antighosts and the auxiliary $`b^I`$-fields. This yields
$$S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}=S^{\mathrm{𝑖𝑛𝑣}}+S^{\mathrm{𝑔𝑓}}+S^{\mathrm{𝑔ℎ𝑜𝑠𝑡}}+S^{\mathrm{𝑠𝑜𝑢𝑟𝑐𝑒𝑠}}$$
(2.30)
with
$$S^{\mathrm{𝑠𝑜𝑢𝑟𝑐𝑒𝑠}}=d^nx(\sigma A_\mu ^IK_I^\mu +\sigma \psi ^iK_i+\sigma C^IL_I+\mathrm{i}b_IM^I)$$
(2.31)
where $`K_I^\mu `$, $`K_i`$, $`L_I`$ and $`M_I`$ are respectively the sources for the BRST variations of $`A_\mu ^I`$, $`\psi _i`$, $`C^I`$ and $`\overline{C}_I`$. We shall also denote by $`N^I`$ the sources associated with $`b_I`$. The antighosts $`\overline{C}_I`$, the auxiliary fields $`b_I`$ and their sources define the “non-minimal sector”.
The action $`S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}`$ fulfills the identity
$$(S^{total},S^{total})=0$$
(2.32)
where the “antibracket” $`(,)`$ is defined by declaring the sources to be conjugate to the corresponding fields, i.e.,
$$(A_\mu ^I(x),K_J^\nu (y))=\delta _J^I\delta _\mu ^\nu \delta ^n(xy),(C^I(x),L_J(y))=\delta _J^I\delta ^n(xy)\text{etc.}$$
(2.33)
The generating functional $`\mathrm{\Gamma }`$ of the one-particle irreducible proper vertices coincides with $`S^{total}`$ to zeroth order in $`\mathrm{}`$,
$$\mathrm{\Gamma }=S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}+\mathrm{}\mathrm{\Gamma }^{(1)}+O(\mathrm{}^2).$$
(2.34)
The perturbative quantum problem is to prove that the renormalized finite $`\mathrm{\Gamma }`$, obtained through the addition of counterterms of higher orders in $`\mathrm{}`$ to $`S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}`$, obeys the same identity,
$$(\mathrm{\Gamma },\mathrm{\Gamma })=0$$
(2.35)
These are the Ward-Slavnov-Taylor identities (written in Zinn-Justin form) associated with the original gauge symmetry and the antighost shift symmetry. The problem involves two aspects: anomalies and stability.
General theorems guarantee that to lowest order in $`\mathrm{}`$, the breaking $`\mathrm{\Delta }_k`$ of the Ward identity,
$$(\mathrm{\Gamma },\mathrm{\Gamma })=\mathrm{}^k\mathrm{\Delta }_k+O(\mathrm{}^{k+1}),$$
(2.36)
is a local functional. The identity $`(\mathrm{\Gamma },(\mathrm{\Gamma },\mathrm{\Gamma }))=0`$, then gives the lowest order consistency condition
$$𝒮\mathrm{\Delta }_k=0,$$
(2.37)
where $`𝒮`$ is the so-called (linearized) Slavnov operator, $`𝒮=(S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}},)`$ which fulfills $`𝒮^2=0`$ because of (2.32). Trivial anomalies of the form $`𝒮\mathrm{\Sigma }_k`$ can be absorbed through the addition of finite counterterms, so that, in the absence of non trivial anomalies, (2.35) can be fulfilled to that order. Hence, non trivial anomalies are constrained by the cohomology of $`𝒮`$ in ghost number $`1`$.
The remaining counterterms $`S_k`$ of that order must satisfy
$$𝒮S_k=0,$$
(2.38)
in order to preserve the Ward identity to that order. Solutions of the form $`𝒮`$(something) can be removed through field redefinitions or a change of the gauge conditions. The question of stability in the minimal, physical, sector is the question whether any non trivial solution of this equation can be brought back to the form $`S^{\mathrm{𝑖𝑛𝑣}}`$ by redefinitions of the coupling constants and field redefinitions. Thus, it is the cohomology of $`𝒮`$ in ghost number $`0`$ which is relevant for the analysis of stability.
As it has been defined, $`𝒮`$ acts also on the sources and depends on the gauge-fixing. However, the gauge conditions can be completely absorbed through a redefinition of the antifields
$`A_I^\mu `$ $`=`$ $`K_I^\mu +{\displaystyle \frac{\delta \mathrm{\Psi }}{\delta A_\mu ^I}},`$ (2.39)
$`\psi _i^{}`$ $`=`$ $`K_i+{\displaystyle \frac{\delta \mathrm{\Psi }}{\delta \psi ^i}},`$ (2.40)
$`\overline{C}^I`$ $`=`$ $`M^I+{\displaystyle \frac{\delta \mathrm{\Psi }}{\delta \overline{C}_I}},`$ (2.41)
$`C_I^{}`$ $`=`$ $`L_I+{\displaystyle \frac{\delta \mathrm{\Psi }}{\delta C^I}},`$ (2.42)
$`b^I`$ $`=`$ $`N^I+{\displaystyle \frac{\delta \mathrm{\Psi }}{\delta b_I}},`$ (2.43)
where, in our case, $`\mathrm{\Psi }`$ is given by
$$\mathrm{\Psi }=\mathrm{i}\overline{C}_I(^I+\frac{1}{2}b^I).$$
(2.44)
This change of variables does not affect the antibracket (”canonical transformation generated by $`\mathrm{\Psi }`$”). In terms of the new variables, $`S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}`$ becomes
$$S^{\mathrm{𝑡𝑜𝑡𝑎𝑙}}=S^{\mathrm{𝑖𝑛𝑣}}d^nx\left[(sA_\mu ^I)A_I^\mu +(s\psi ^i)\psi _i^{}+(sC^I)C_I^{}\right]+S^{\mathrm{𝑁𝑜𝑛𝑀𝑖𝑛}}$$
(2.45)
where the non-minimal part is just
$$S^{\mathrm{𝑁𝑜𝑛𝑀𝑖𝑛}}=\mathrm{i}d^nxb_I\overline{C}^I$$
(2.46)
The Slavnov operator becomes then
$$𝒮=s+\mathrm{i}d^nx\left[b_I(x)\frac{\delta }{\delta \overline{C}_I(x)}\overline{C}^I(x)\frac{\delta }{\delta b^I(x)}\right]$$
(2.47)
where here, $`s`$ acts on the variables $`A_\mu ^I`$, $`\psi ^i`$, $`C^I`$ of the “minimal sector” and on their antifields $`A_I^\mu `$, $`\psi _i^{}`$, $`C_I^{}`$ and takes exactly the form given in (2.8), while the remaining piece is contractible and does not contribute to the cohomology (see appendix B). Thus, $`H(𝒮)`$ and $`H(s)`$ are isomorphic, as are $`H(𝒮|d)`$ and $`H(s|d)`$: any cocycle of $`𝒮`$ may be assumed not to depend on the variables $`\overline{C}_I`$, $`\overline{C}^I`$, $`b^I`$ and $`b_I`$ of the “non-minimal sector” and is then a cocycle of $`s`$. Furthermore, for chains depending on the variables of the minimal sector only, $`s`$-coboundaries and $`𝒮`$-coboundaries coincide. Hence, the cohomological problems indeed reduce to computing $`H(s)`$ and $`H(s|d)`$.
Remarks:
(i) Whereas the choice made above to introduce sources also for the antighosts $`\overline{C}_I`$ and the auxiliary $`b^I`$ fields is motivated by the desire to have a symmetrical description of fields and sources with respect to the antibracket (2.33) and a BRST transformation that is canonically generated on all the variables, other authors prefer to introduce sources for the BRST variations of the variables $`A_\mu ^I`$, $`\psi _i`$ and $`C^I`$ only. In that approach, the final BRST differential in the physical sector is the same as above, but the non minimal sector is smaller and consists only of $`\overline{C}_I`$ and $`b^I`$.
(ii) In the case of standard Yang-Mills theories in four dimensions, one may wonder whether one has stability of the complete action (2.30) not only in the relevant, physical sector but also in the gauge-fixing sector, i.e., whether linear gauges are stable. Stability of linear gauges can be established by imposing legitimate auxiliary conditions. In the formalism where the antighosts and auxiliary fields have no antifields, these auxiliary conditions are the gauge condition and the ghost equation, fixing the dependence of $`\mathrm{\Gamma }`$ on $`b^I`$ and $`\overline{C}_I`$ respectively. The same result can be recovered in the approach with antifields for the antighosts and the auxiliary fields. The dependence of $`\mathrm{\Gamma }`$ on the variables of the non minimal sector is now fixed by the same ghost equation as before, while the gauge condition is modified through the additional term $`\mathrm{i}\overline{C}_I^{}b^I`$. In addition, one imposes: $`\delta \mathrm{\Gamma }/\delta \overline{C}_I^{}=\mathrm{i}b^I`$ and $`\delta \mathrm{\Gamma }/\delta b_I^{}=0`$.
(iii) One considers sometimes a different cohomology, the so-called gauge-fixed BRST cohomology, in which there is no antifield and the equations of motion of the gauge-fixed theory are freely used. This cohomology is particularly relevant to the ”quantum Noether method” for gauge-theories . The connection between the BRST cohomology discussed here and the gauge-fixed cohomology is studied in , where it is shown that they are isomorphic under appropriate conditions which are explicitly stated.
### 2.7 Appendix 2.B: Contractible pairs
We show here that the antighosts and the auxiliary fields do not contribute to the cohomology of $`𝒮`$. This is because $`\overline{C}_I`$ and $`b_I`$ form “contractible pairs”,
$$𝒮\overline{C}_I=\mathrm{i}b_I,𝒮b_I=0,$$
(2.48)
and furthermore, the $`𝒮`$-transformations of the other variables do not involve $`\overline{C}_I`$ or $`b_I`$. We shall repeatedly meet the concept of “contractible pairs” in this work.
Let $`N`$ be the operator counting $`\overline{C}_I`$ and $`b_I`$ and their derivatives,
$$N=\overline{C}_I\frac{}{\overline{C}_I}+b_I\frac{}{b_I}+\underset{t>0}{}\left[_{\mu _1\mathrm{}\mu _t}\overline{C}_I\frac{}{(_{\mu _1\mathrm{}\mu _t}\overline{C}_I)}+_{\mu _1\mathrm{}\mu _t}b_I\frac{}{(_{\mu _1\mathrm{}\mu _t}b_I)}\right].$$
(2.49)
One has $`[N,𝒮]=0`$ and in fact $`N=𝒮\varrho +\varrho 𝒮`$ with
$$\varrho =\mathrm{i}\overline{C}_I\frac{}{b_I}\mathrm{i}\underset{t>0}{}_{\mu _1\mathrm{}\mu _t}\overline{C}_I\frac{}{(_{\mu _1\mathrm{}\mu _t}b_I)}.$$
(2.50)
$`\varrho `$ is called a contracting homotopy for $`N`$ with respect to $`𝒮`$. Now, let $`a`$ be $`𝒮`$-closed. One can expand $`a`$ according to the $`N`$-degree, $`a=_{k0}a_k`$ with $`Na_k=ka_k`$. One has $`𝒮a_k=0`$ since $`[N,𝒮]=0`$. It is easy to show that the components of $`a`$ with $`k>0`$ are all $`𝒮`$-exact. Indeed, for $`k>0`$ one has $`a_k=(1/k)Na_k=(1/k)(𝒮\varrho +\varrho 𝒮)a_k`$ and thus $`a_k=𝒮b_k`$ with $`b_k=(1/k)\varrho a_k`$. Accordingly, $`a=a_0+𝒮(_{k>0}b_k)`$. Hence, the ”non-minimal part” of an $`s`$-cocycle is always trivial. Furthermore, by analogous arguments, an $`s`$-cocycle $`a`$ is trivial if and only if its ”minimal part” $`a_0`$ is trivial in the minimal sector.
Similarly, if $`a`$ is a solution of the consistency condition, $`𝒮a+dm=0`$, one has $`𝒮a_k+dm_k=0`$ and one gets, for $`k0`$, $`a_k=𝒮b_k+dn_k`$ with $`b_k=(1/k)\varrho a_k`$ and $`n_k=(1/k)\varrho m_k`$ (thanks to $`[N,d]=0`$, $`d\varrho +\varrho d=0`$). Thus, the cohomology of $`𝒮`$ can be non-trivial only in the space of function(al)s not involving the antighosts and the auxiliary fields $`b_I`$.
The same reasoning applies to the antifields $`\overline{C}^I`$ and $`b^I`$, which form also contractible pairs since
$$𝒮\overline{C}^I=0,𝒮b^I=\mathrm{i}\overline{C}^I.$$
(2.51)
The argument is actually quite general and constitutes one of our primary tools for computing cohomologies. We shall make a frequent use of it in the report, whenever we have a pair of independent variables $`(x,y)`$ and a differential $`\mathrm{\Delta }`$ such that $`\mathrm{\Delta }x=y`$ and $`\mathrm{\Delta }y=0`$, and the action of $`\mathrm{\Delta }`$ on the remaining variables does not involve $`x`$ or $`y`$.
## 3 Outline of report
The calculation of the BRST cohomology is based on the decomposition of $`s`$ into $`\delta +\gamma `$, on the computation of the individual cohomologies of $`\delta `$ and $`\gamma `$ and on the descent equations. Our approach follows , but we somewhat streamline and systematize the developments of these papers by starting the calculation ab initio. This enables one to make some shortcuts in the derivation of the results.
We start by recalling some useful properties of the exterior derivative $`d`$ in the algebra of local forms (section 4). In particular, we establish the important “algebraic Poincaré lemma” (theorem 4.2), which is also a tool repeatedly used in the whole report.
We compute then $`H(\delta )`$ and $`H(\delta |d)`$ (sections 5 and 6, respectively). 0ne can establish general properties of $`H(\delta )`$ and $`H(\delta |d)`$, independently of the model and valid for other gauge theories such as gravity or supergravity. In particular, the relationship between $`H(\delta |d)`$ and the (generalized) conservation laws is quite general, although the detailed form of the conservation laws does depend of course on the model. We have written section 6 with the desire to make these general features explicit. We specialize then the analysis to gauge theories of the Yang-Mills type. Within this set of theories (and on natural regularity and normality conditions on the Lagrangian), the groups $`H(\delta |d)`$ are completely calculated, except $`H_1^n(\delta |d)`$, which is related to the global symmetries of the model and can be fully determined only when the Lagrangian is specified.
In section 7, we establish the general link between $`H(s)`$, $`H(\delta )`$ and $`H(\gamma )`$ (respectively, $`H(s|d)`$, $`H(\delta |d)`$ and $`H(\gamma |d))`$. The connection follows the line of “homological perturbation theory” and applies also to generic field theories with a gauge freedom.
We compute next $`H(\gamma )`$ (section 8). The calculation is tied to theories of the Yang-Mills type but within this class of models, it does not depend on the form of the Lagrangian since $`\gamma `$ involves only the gauge transformations and not the detailed dynamics. We first show that the calculation of $`H(\gamma )`$ reduces to the problem of computing the Lie algebra cohomology of the gauge group formulated in terms of the curvature components, the matter fields, the antifields, and their covariant derivatives, and the undifferentiated ghosts. This is a well-known mathematical problem whose general solution has been worked out long ago (for reductive Lie algebras). Knowing the connection between $`H(\delta )`$, $`H(\gamma )`$ and $`H(s)`$ makes it easy to compute $`H(s)`$ from $`H(\delta )`$ and $`H(\gamma )`$.
We then turn to the computation of $`H(s|d)`$. The relevant mathematical tool is that of the descent equations, which we first review (section 9). Equipped with this tool, we calculate $`H(s|d)`$ in all form and ghost degrees in a smaller algebra involving only the forms $`C^I`$, $`dC^I`$, $`A^I`$, $`dA^I`$ and their exterior products (section 10). Although this problem is a sub-problem of the general calculation of $`H(s|d)`$, it turns out to be crucial for investigating solutions of the consistency condition $`sa+dm=0`$ that “descend non trivially”. The general case (in the algebra of all local forms not necessarily expressible as exterior products of $`C^I`$, $`dC^I`$, $`A^I`$ and $`dA^I`$) is treated next (section 11), paying due attention to the antifield dependence.
Because the developments in section 11 are rather involved, we discuss their physical implications in a separate section (section 12). The reader who is not interested in the proofs but only in the results may skip section 11 and go directly to section 12. We explain in particular there why the antifields can be removed from the general solution of the consistency condition at ghost numbers zero (counterterms) and one (anomalies) when the gauge group is semisimple. We also explain why the anomalies can be expressed solely in terms of $`C^I`$, $`dC^I`$, $`A^I`$ and $`dA^I`$ in the semisimple case . These features do not hold, however, when there are abelian factors or for different values of the ghost number.
The case of a system of free abelian gauge fields, relevant to the construction of consistent couplings among massless vector particles, has special features and is therefore treated separately in section 13. We also illustrate the general results in the case of pure Chern-Simons theory in three dimensions, where many solutions disappear because the Yang-Mills curvature vanishes when the equations of motion hold (section 14).
Finally, the last section reviews the literature on the calculation of the local BRST cohomology $`H(s|d)`$ for other local field theories with a gauge freedom.
## 4 Locality - Algebraic Poincaré lemma: $`H(d)`$
Since locality is a fundamental ingredient in our approach, we introduce in this subsection the basic algebraic tools that allow one to deal with locality. The central idea is to consider the “fields” $`\varphi ^i`$ and their partial derivatives of first and higher order as independent coordinates of so-called jet-spaces. The approach is familiar from the variational calculus where the fields and their partial derivatives are indeed treated as independent variables when computing the derivatives $`L/\varphi ^i`$ or $`L/(_\mu \varphi ^i)`$ of the Lagrangian $`L`$. Fields, in the usual sense of “functions of the space-time coordinates”, emerge in this approach as sections of the corresponding jet-bundles.
What are the “fields” will depend on the context. In the case of gauge theories of the Yang-Mills type, the “fields” $`\varphi ^i`$ may be the original classical fields $`A_\mu ^I`$ and $`\psi ^i`$, or may be these fields plus the ghosts. In some other instances, they could be the original fields plus the antifields, or the complete set of all variables introduced in the introduction. The considerations of this section are quite general and do not depend on any specific model or field content. So, we shall develop the argument without committing ourselves to a definite set of variabes.
### 4.1 Local functions and jet-spaces
A local function $`f`$ is a smooth function of the spacetime coordinates, the field variables $`\varphi ^i`$ and a finite number of their derivatives, $`f=f(x,[\varphi ])`$, where the notation $`[\varphi ]`$ means dependence on $`\varphi ^i,\varphi _\mu ^i,\mathrm{},\varphi _{(\mu _1\mathrm{}\mu _k)}^i`$ for some finite $`k`$ (with $`\varphi _\mu ^i_\mu \varphi ^i`$, $`\varphi _{(\mu _1\mathrm{}\mu _k)}^i_{(\mu _1}\mathrm{}_{\mu _k)}\varphi ^i`$). A local function is thus a function on the “jet space of order $`k`$$`J^k(E)=M\times V^k`$ (for some $`k`$), where $`M`$ is Minkowski (or Euclidean) spacetime and where $`V^k`$ is the space with coordinates given by $`\varphi ^i,\varphi _\mu ^i,\mathrm{},\varphi _{(\mu _1\mathrm{}\mu _k)}^i`$ – some of which may be Grassmannian. The fields and their various derivatives are considered as completely independent in $`J^k(E)`$ except that the various derivatives commute, so that only the completely symmetric combinations are independent coordinates.
In particular, the jet space of order zero $`J^0(E)E`$ is coordinatized by $`x^\mu `$ and $`\varphi ^i`$. A field history is a section $`s:ME`$, given in coordinates by $`x(x,\varphi (x))`$. A section of $`E`$ induces a section of $`J^k(E)`$, with $`\varphi _{(\mu _1\mathrm{}\mu _k)}^i|_s=^k\varphi ^i(x)/x^{\mu _1}\mathrm{}x^{\mu _k}`$. Evaluation of a local function at a section yields a spacetime function. The independence of the derivatives reflects the fact that the only local function $`f(x,[\varphi ])`$ which is zero on all sections is the function $`f0`$.
Because the order in the derivatives of the relevant functions is not always known a priori, it is useful to introduce the infinite jet-bundle $`\pi ^{\mathrm{}}:J^{\mathrm{}}(E)=M\times V^{\mathrm{}}M`$, where coordinates on $`V^{\mathrm{}}`$ are given by $`\varphi ^i,\varphi _\mu ^i,\varphi _{(\mu _1\mu _2)}^i,\mathrm{}`$.
In our case where spacetime is Minkowskian (or Euclidean) and the field manifold is homeomorphic to $`^m`$ (with $`m`$ the number of independent real fields $`\varphi ^i`$, $`i=1,\mathrm{},m`$), the jet-bundles are trivial and the use of bundle terminology may appear a bit pedantic. However, this approach is crucial for dealing with more complicated situations in which the spacetime manifold and/or the field manifold is topologically non-trivial. Global aspects related to these features will not be discussed here. They come over and above the local cohomologies analyzed in this report, which must in any case be understood<sup>2</sup><sup>2</sup>2Our considerations are furthermore sufficient for the purposes of perturbative quantum field theory.. We refer to for an analysis of BRST cohomology taking into account some of these extra global features.
In the case of field theory, the local functions are usually polynomial in the derivatives. This restriction will turn out to be important in some of the next sections. However, for the present purposes, it is not necessary. The theorems of this section and the next are valid both in the space of polynomial local functions and in the space of arbitrary smooth local functions. For this reason, we shall not restrict the functional space of local functions at this stage.
### 4.2 Local functionals - Local $`p`$-forms
An important class of objects are local functionals. They are given by the integrals over space-time of local $`n`$-forms evaluated at a section. An example is of course the classical action.
Local $`p`$-forms are by definition exterior forms with coefficients that are local functions,
$$\omega =\frac{1}{p!}dx^{\mu _1}\mathrm{}dx^{\mu _p}\omega _{\mu _1\mathrm{}\mu _p}(x,[\varphi ]).$$
(4.1)
We shall drop the exterior product symbol $``$ in the sequel since no confusion can arise.
Thus, local functionals evaluated at the section $`s`$ take the form $`(f,s)=_M\omega |_s`$ with $`\omega `$ the $`n`$-form $`\omega =fd^nx`$, where $`d^nx=dx^0\mathrm{}dx^{n1}`$. In the case of the action, $`f`$ is the Lagrangian density.
It is customary to identify local functionals with the formal expression $`_M\omega `$ prior to evaluation.
### 4.3 Total and Euler-Lagrange derivatives
The total derivative $`_\mu `$ is the vector field defined on local functions by
$`_\mu ={\displaystyle \frac{}{x^\mu }}+\varphi _\mu ^i{\displaystyle \frac{}{\varphi }}+\varphi _{(\mu \nu )}^i{\displaystyle \frac{}{\varphi _\nu ^i}}+\mathrm{}={\displaystyle \frac{}{x^\mu }}+{\displaystyle \underset{k0}{}}\varphi _{(\mu \nu _1\mathrm{}\nu _k)}^i{\displaystyle \frac{}{\varphi _{(\nu _1\mathrm{}\nu _k)}^i}},`$ (4.2)
where for convenience, we define the index $`(\nu _1\mathrm{}\nu _k)`$ to be absent for $`k=0`$. These vector fields commute, $`[_\mu ,_\nu ]=0`$. Note that $`_\mu \varphi =\varphi _\mu `$, $`_\mu _\nu \varphi =\varphi _{\mu \nu }`$ etc… Furthermore, evaluation at a section and differentiation commute as well:
$`(_\mu f)|_s={\displaystyle \frac{}{x^\mu }}(f|_s).`$ (4.3)
The Euler-Lagrange derivative $`\frac{\delta }{\delta \varphi ^i}`$ is defined on a local function $`f`$ by
$`{\displaystyle \frac{\delta f}{\delta \varphi ^i}}={\displaystyle \frac{f}{\varphi ^i}}_\mu {\displaystyle \frac{f}{\varphi _\mu ^i}}+\mathrm{}={\displaystyle \underset{k0}{}}()^k_{(\mu _1\mathrm{}\mu _k)}{\displaystyle \frac{f}{\varphi _{(\mu _1\mathrm{}\mu _k)}^i}},`$ (4.4)
where $`_{(\mu _1\mathrm{}\mu _k)}=_{\mu _1}\mathrm{}_{\mu _k}`$.
### 4.4 Relation between local functionals and local functions
A familiar property of total derivatives $`_\mu j^\mu `$ is that they have vanishing Euler-Lagrange derivatives. The converse is also true. In fact, one has
###### Theorem 4.1
(i) A local function is a total derivative iff it has vanishing Euler-Lagrange derivatives with respect to all fields,
$$f=_\mu j^\mu \frac{\delta f}{\delta \varphi ^i}=0\varphi ^i.$$
(4.5)
(ii) Two local functionals $`,𝒢`$ agree for all sections $`s`$, $`(f,s)=𝒢(g,s)`$ iff their integrands differ by a total derivative, $`f=g+_\mu j^\mu `$, for some local functions $`j^\mu `$, whose boundary integral vanishes, $`_Mj=0`$.
##### Proof:
(i) Let
$$N=\underset{k0}{}\varphi _{(\mu _1\mathrm{}\mu _k)}^i\frac{}{\varphi _{(\mu _1\mathrm{}\mu _k)}^i}.$$
We start from the identity
$`f(x,[\varphi ])f(x,0)={\displaystyle _0^1}𝑑\lambda {\displaystyle \frac{d}{d\lambda }}f(x,[\lambda \varphi ])={\displaystyle _0^1}{\displaystyle \frac{d\lambda }{\lambda }}[Nf](x,[\lambda \varphi ]).`$ (4.6)
Using integrations by parts and $`f(x,0)=_\mu k^\mu (x)`$ (which holds as a consequence of the standard Poincaré lemma for $`^n`$), one gets
$$f(x,[\varphi ])=_\mu j^\mu +_0^1\frac{d\lambda }{\lambda }[\varphi ^i\frac{\delta f}{\delta \varphi ^i}](x,[\lambda \varphi ]).$$
(4.7)
for some local functions $`j^\mu `$. Thus, $`\delta f/\delta \varphi ^i=0`$ $`i`$ implies $`f=_\mu j^\mu `$. Evidently, $`j^\mu `$ is polynomial whenever $`f`$ is.
Conversely, we have
$$[\frac{}{\varphi _{(\mu _1\mathrm{}\mu _k)}^i},_\nu ]=\delta _{(\nu }^{\mu _1}\mathrm{}\delta _{\lambda _{k1})}^{\mu _k}\frac{}{\varphi _{(\lambda _1\mathrm{}\lambda _{k1})}^i}.$$
(4.8)
\[For $`k=0`$: $`[/\varphi ^i,_\nu ]=0`$.\] This gives
$`{\displaystyle \underset{k0}{}}()^k_{(\mu _1\mathrm{}\mu _k)}{\displaystyle \frac{(_\nu j^\nu )}{\varphi _{(\mu _1\mathrm{}\mu _k)}^i}}={\displaystyle \underset{k0}{}}()^k_{(\mu _1\mathrm{}\mu _k\nu )}{\displaystyle \frac{j^\nu }{\varphi _{(\mu _1\mathrm{}\mu _k)}^i}}`$
$`+{\displaystyle \underset{k1}{}}()^k_{(\nu \lambda _1\mathrm{}\lambda _{k1})}{\displaystyle \frac{j^\nu }{\varphi _{(\lambda _1\mathrm{}\lambda _{k1})}^i}}=0.`$ (4.9)
(ii) That two local functionals whose integrands differ by a total derivative with vanishing boundary integral agree for all sections $`s`$ follows from (4.3) and Stokes theorem. Conversely, $`(f,s)=𝒢(g,s)`$ for all $`s`$, implies $`I=_Md^nx(fg)|_{\varphi ^i(x)+\epsilon \eta ^i(x)}=0`$ for all $`\epsilon `$. Thus, for sections $`\eta ^i(x)`$ which vanish with a sufficient number of their derivatives at $`M`$, using integrations by parts and (4.3), one gets $`0=\frac{d}{d\epsilon }|_{ϵ=0}I=_Md^nx\eta ^i(x)\frac{\delta (fg)}{\delta \varphi ^i}|_{\varphi ^i(x)}`$, which implies $`\frac{\delta (fg)}{\delta \varphi ^i}=0`$ for all $`\varphi ^i`$ and concludes the proof by using (i).
Remarks:
(i) The first part of this theorem can be reformulated as the statement that “terms in a classical Lagrangian give no contributions to the classical equations of motion iff they are total derivatives”. It is crucial to work in jet space for this statement to be true since the Poincaré lemma for the standard De Rham cohomology in $`^n`$ implies that any function of $`x`$ can be written as a total derivative. Thus, if we were to consider naïvely the Lagrangian $`L`$ as a function of the spacetime coordinates ($`L=L(x)`$), we would get an apparent contradiction, since on the one hand the Euler-Lagrange equations are in general not empty while on the other hand $`L`$ can be written as a total derivative in $`x`$-space. The point is of course that $`L`$ would not be given in general by the total derivative of a local vector density.
(ii) The theorem leads to the following view on local functionals, put forward in explicit terms in : local functionals can be identified with equivalence classes of local functions modulo total derivatives. This is legitimate whenever the surface terms can be neglected or are not under focus. This is the case in the physical situations described in the introduction (e.g. in renormalization theory, the classical fields in the effective action $`\mathrm{\Gamma }`$ are in fact sources yielding 1-particle irreducible Green functions. They can be assumed to be of compact support in that context). In the remainder of the report, we always use this identification.
### 4.5 Algebraic Poincaré lemma - $`H(d)`$
Let $`\mathrm{\Omega }`$ be the algebra of local forms. The exterior (also called horizontal) differential in $`\mathrm{\Omega }`$ is defined by $`d=dx^\mu _\mu `$ with $`_\mu `$ given by formula (4.2).
The derivative $`d`$ satisfies $`d^2=0`$ because the $`dx^\mu `$ anticommute while the $`_\mu `$ commute. The complex $`(\mathrm{\Omega },d)`$ is called the horizontal complex. Its elements, the local forms, are also called the horizontal forms, or just the “forms”. The $`p`$-forms $`\omega ^p`$ satisfying $`d\omega ^p=0`$ are called closed, or cocycles (of $`d`$), the ones given by $`\omega ^p=d\omega ^{p1}`$, which are necessarily closed, are called exact or coboundaries (of $`d`$). The cohomology group $`H(d,\mathrm{\Omega })`$ is defined to be the space of equivalence classes of cocycles modulo coboundaries, $`[\omega ^p]H^p(d,\mathrm{\Omega })`$ if $`d\omega ^p=0`$ with $`\omega ^p\omega ^p+d\omega ^{p1}`$.
Integrands of local functionals are local $`n`$-forms. A local $`n`$-form $`\omega ^n=d^nxf`$ is exact, $`\omega ^n=d\omega ^{n1}`$ (with $`\omega ^{n1}`$ a local $`(n1)`$-form), if and only if $`f`$ is a total derivative; one has
$$d^nxf=d\omega ^{n1}f=_\mu j^\mu \frac{\delta f}{\delta \varphi ^i}=0\varphi ^i$$
(4.10)
where the first equivalence holds with $`\omega ^{n1}=\frac{1}{(n1)!}ϵ_{\mu \nu _1\mathrm{}\nu _{n1}}dx^{\nu _1}\mathrm{}dx^{\nu _{n1}}j^\mu `$ and the second one holds by theorem 4.1. Due to $`d\omega ^n=0`$, theorem 4.1 can thus be reformulated as the statement that local functionals are described by $`H^n(d,\mathrm{\Omega })`$, and that $`H^n(d,\mathrm{\Omega })`$ is given by the equivalence classes $`[\omega ^n]`$ having identical Euler-Lagrange derivatives, $`\omega ^n=fd^nx\omega _{}^{}{}_{}{}^{n}=f^{}d^nx`$ if $`\frac{\delta }{\delta \varphi ^i}(ff^{})=0`$. An important result is the following on the cohomology of $`d`$ in lower form degrees.
###### Theorem 4.2
The cohomology of $`d`$ in form degree strictly smaller than $`n`$ is exhausted by the constants in form degree $`0`$,
$`0<p<n:`$ $`d\omega ^p=0\omega ^p=d\omega ^{p1};`$
$`p=0:`$ $`d\omega ^0=0\omega ^0=\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}.`$ (4.11)
Theorem 4.1 (part (i)) and theorem 4.2, which give the cohomology of $`d`$ in the algebra of local forms, are usually collectively referred to as the “algebraic Poincaré lemma”. The proof below applies to the case of a polynomial dependence on derivatives. Proofs covering the smooth case can be found in the literature cited at the end of this section.
##### Proof:
As in (4.6), we have the decomposition $`\omega (x,dx,[\varphi ])=\omega _0+\stackrel{~}{\omega }`$, where $`\omega _0=\omega (x,dx,0)`$ and $`\stackrel{~}{\omega }(x,dx,[\varphi ])=_0^1\frac{d\lambda }{\lambda }[N\omega ](x,dx,[\lambda \varphi ])`$.
The condition $`d\omega =0`$ then implies separately $`dx^\mu \frac{}{x^\mu }\omega _0=0`$ and $`d\stackrel{~}{\omega }(x,dx,[\varphi ])=0`$ because $`d`$ is homogeneous of degree zero in $`\lambda `$ and commutes with $`N`$.
Defining $`\rho ^{}=x^\nu \frac{}{dx^\nu }`$, we have $`\{dx^\mu \frac{}{x^\mu },\rho ^{}\}=x^\mu \frac{}{x^\mu }+dx^\nu \frac{}{dx^\nu }`$. Using a homotopy formula analogous to (4.6) for the variables $`x^\mu ,dx^\mu `$, we get $`\omega _0(x,dx,0)=\omega (0,0,0)+d_0^1\frac{d\lambda }{\lambda }[\rho ^{}\omega _0](\lambda x,\lambda dx,0)`$, which is the standard Poincaré lemma.
Let $`t^\nu =_{k0}k\delta _{(\mu _1}^\nu \delta _{\mu _2}^{\lambda _1}\mathrm{}\delta _{\mu _k)}^{\lambda _{k1}}\varphi _{(\lambda _1\mathrm{}\lambda _{k1})}^i\frac{}{\varphi _{(\mu _1\mathrm{}\mu _k)}^i}`$. Then
$$[t^\nu ,_\mu ]=\delta _\mu ^\nu N.$$
(4.12)
If one defines $`D^{+\nu }\stackrel{~}{\omega }=_0^1\frac{d\lambda }{\lambda }[t^\nu \stackrel{~}{\omega }](x,dx,[\lambda \varphi ])`$, one gets
$$[D^{+\nu },_\mu ]\stackrel{~}{\omega }=\delta _\mu ^\nu \stackrel{~}{\omega },$$
(4.13)
because $`_\mu `$ is homogeneous of degree $`0`$ in $`\lambda `$. With $`\rho =D^{+\nu }\frac{}{dx^\nu }`$, one has
$$\{d,\rho \}\stackrel{~}{\omega }=[D^{+\nu }_\nu dx^\mu \frac{}{dx^\mu }]\stackrel{~}{\omega }=[_\nu D^{+\nu }+(ndx^\mu \frac{}{dx^\mu })]\stackrel{~}{\omega }.$$
(4.14)
Let $`\alpha =np`$, for $`p<n`$. Apply the previous relation to a $`d`$-closed $`p`$-form $`\stackrel{~}{\omega }^p`$ to get
$$d\stackrel{~}{\omega }^p=0\stackrel{~}{\omega }^p=d\frac{\rho }{\alpha }\stackrel{~}{\omega }^p\frac{1}{\alpha }_\nu D^{+\nu }\stackrel{~}{\omega }^p.$$
(4.15)
We want to use this formula recursively. In order to do so, we need some relations for the operators $`P_m=_{\nu _1}\mathrm{}_{\nu _m}D^{+\nu _1}\mathrm{}D^{+\nu _m}`$ where, by definition $`P_0=1`$. (4.13) implies $`[P_1,d]\stackrel{~}{\omega }=d\stackrel{~}{\omega }`$ and $`P_1P_m\stackrel{~}{\omega }=[P_{m+1}+mP_m]\stackrel{~}{\omega }`$. The latter allows one to express $`P_m`$ in terms of $`P_1`$: $`P_m\stackrel{~}{\omega }=\mathrm{\Pi }_{l=0}^{m1}(P_1l)\stackrel{~}{\omega }`$. If $`\stackrel{~}{\omega }`$ is closed, it follows that $`dP_m\stackrel{~}{\omega }=0`$. Together with (4.14) (applied to a $`p`$-form $`P_m\stackrel{~}{\omega }^p`$), this yields
$$d\stackrel{~}{\omega }^p=0(\alpha +m)P_m\stackrel{~}{\omega }^p=d\rho P_m\stackrel{~}{\omega }^pP_{m+1}\stackrel{~}{\omega }^p.$$
(4.16)
Injecting this relation for $`m=1`$ into (4.15), we find
$`\stackrel{~}{\omega }^p=d[{\displaystyle \frac{\rho }{\alpha }}{\displaystyle \frac{\rho }{\alpha (\alpha +1)}}P_1]\stackrel{~}{\omega }^p+{\displaystyle \frac{1}{\alpha (\alpha +1)}}P_2\stackrel{~}{\omega }^p.`$
Going on recursively, the procedure will stop because $`\stackrel{~}{\omega }^p`$ is polynomial in derivatives: the total number of derivatives contained $`\stackrel{~}{\omega }^p`$ is bounded by some $`\overline{m}`$, so that $`P_{\overline{m}+1}\stackrel{~}{\omega }^p=0`$. Hence, the final result is $`d\stackrel{~}{\omega }^p=0`$ $``$ $`\stackrel{~}{\omega }^p=d\eta ^{p1}`$, with
$`\eta ^{p1}={\displaystyle \underset{m=0}{\overset{\overline{m}}{}}}{\displaystyle \frac{()^m\rho P_m\stackrel{~}{\omega }^p}{\alpha (\alpha +1)\mathrm{}(\alpha +m)}},\alpha =np.`$
This proves the theorem, by noting that $`\omega (0,0,0)`$ can never be $`d`$ exact.
Note that the above construction preserves polynomiality: $`\eta ^{p1}`$ is polynomial in the fields and their derivatives if $`\omega ^p`$ is.
### 4.6 Cohomology of $`d`$ in the complex of $`x`$-independent local forms
The previous theorem holds in the space of forms that are allowed to have an explicit $`x`$-dependence. It is sometimes necessary to restrict the analysis to translation-invariant local forms, which have no explicit $`x`$-dependence. In that case, the cohomology is bigger, because the constant forms (polynomials in the $`dx^\mu `$’s with constant coefficients) are closed but not exact in the algebra of forms without explicit $`x`$-dependence ($`dx^\mu =d(x^\mu )`$, but $`x^\mu `$ is not in the algebra). In fact, as an adaptation of the proof of the previous theorem easily shows, this is the only additional cohomology.
###### Theorem 4.3
(algebraic Poincaré lemma in the $`x`$-independent case): In the algebra of local forms without explicit $`x`$-dependence, the cohomology of $`d`$ in form degree strictly less than $`n`$ is exhausted by the constant forms:
$$p<n:d\omega ^p=0\omega ^p=d\omega ^{p1}+c_{\mu _1\mathrm{}\mu _p}dx^{\mu _1}\mathrm{}dx^{\mu _p}$$
(4.17)
where the $`c_{\mu _1\mathrm{}\mu _p}`$ are constants.
If one imposes Lorentz invariance, the cohomology in form-degree $`0<p<n`$ disappears since there is no Lorentz-invariant constant form except for $`p=0`$ or $`p=n`$ (see Sections 11.1 and 11.2 for a discussion of Lorentz invariance).
### 4.7 Effective field theories
The Lagrangian of effective Yang-Mills theory contains all terms compatible with gauge invariance . Consequently, it is not a local function since it contains an infinite number of derivatives.
However, the above considerations are relevant to the study of effective theories. Indeed, in that case the Lagrangian $`L`$ is in fact a formal power series in some free parameters (including the gauge coupling constants). The coefficients of the independent powers of the parameters in this expansion are local functions in the above sense and are thus defined in a jet-space of finite order. Equality of two formal power series in the parameters always means equality of all the coefficients. For this reason, formal power series in parameters with coefficients that are local function(al)s – in fact polynomials in derivatives by dimensional analysis – can be investigated by means of the tools introduced for local function(al)s. These are the objects that we shall manipulate in the context of effective field theories.
### 4.8 A guide to the literature
Useful references on jet-spaces are . The algebraic Poincaré lemma has been proved by many authors and repeatedly rediscovered. Besides the references just quoted, one can list (without pretendence of being exhaustive!) . We have followed the proof given in .
The very suggestive terminology “algebraic Poincaré lemma” appears to be due to Stora .
## 5 Equations of motion and Koszul-Tate differential: $`H(\delta )`$
The purpose of this section is to compute the homology<sup>3</sup><sup>3</sup>3One speaks of “homology”, rather than cohomology, because $`\delta `$ acts like a boundary (rather than coboundary) operator: it decreases the degree (antifield number) of the objects on which it acts. of the Koszul-Tate differential $`\delta `$ in the algebra of local forms $`\mathrm{\Omega }_F`$ on the jet-space $`J^{\mathrm{}}(F)`$ of the original fields $`A_\mu ^I`$ and $`\psi `$, the ghosts $`C^I`$, the antifields, and their derivatives. To that end, it is necessary to make precise a few properties of the equations of motion.
The action of the Koszul-Tate differential $`\delta `$ for gauge theories of the Yang-Mills type has been defined in the introduction on the basic variables $`A_\mu ^I`$, $`\psi `$, $`C^I`$, $`A_\mu ^I`$, $`\psi _i^{}`$ and $`C_I^{}`$, through formula (2.8). The differential $`\delta `$ is then extended to the jet space $`J^{\mathrm{}}(F)`$ by requiring that $`\delta `$ be a derivation (of odd degree) that commutes with $`_\mu `$. This yields explicitly
$`\delta `$ $`=`$ $`{\displaystyle \underset{l0}{}}_{(\mu _1\mathrm{}\mu _l)}L_I^\mu {\displaystyle \frac{}{A_{I|(\mu _1\mathrm{}\mu _l)}^\mu }}+{\displaystyle \underset{l0}{}}_{(\mu _1\mathrm{}\mu _l)}L_i{\displaystyle \frac{}{\psi _{i|(\mu _1\mathrm{}\mu _l)}^{}}}`$
$`+`$ $`{\displaystyle \underset{m0}{}}_{(\nu _1\mathrm{}\nu _m)}\left[D_\mu A_I^\mu e\psi _i^{}T_{Ij}^i\psi ^j\right]{\displaystyle \frac{}{C_{I|(\nu _1\mathrm{}\nu _m)}^{}}},`$
an expression that makes it obvious that $`\delta `$ is a derivation. By setting $`\delta (dx^\mu )=0`$, one extends $`\delta `$ trivially to the algebra $`\mathrm{\Omega }_F`$ of local forms.
### 5.1 Regularity conditions
#### 5.1.1 Stationary surface
The Euler-Lagrange equations of motion and their derivatives define surfaces in the jet-spaces $`J^r(E)`$, $`J^{\mathrm{}}(E)`$ of the original fields $`A_\mu ^I`$, $`\psi ^i`$, the ghosts $`C^I`$ and their derivatives.
Consider the collection $`R^{\mathrm{}}`$ of equations $`L_I^\mu =0`$, $`L_i=0`$, $`_\nu L_I^\mu =0`$ $`_\nu L_i=0`$, $`_{(\nu _1\nu _2)}L_I^\mu =0`$, $`_{(\nu _1\nu _2)}L_i=0,\mathrm{}`$ defined on $`J^{\mathrm{}}(E)`$. They define the so-called stationary surface $`\mathrm{\Sigma }^{\mathrm{}}`$ on $`J^{\mathrm{}}(E)`$. In a given jet-space $`J^r(E)`$ of finite order $`r`$, the stationary surface $`\mathrm{\Sigma }^r`$ is the surface defined by the subset $`R^rR^{\mathrm{}}`$ of the above collection of equations which is relevant in $`J^r(E)`$.
Note that the equations of motion involve only the original classical fields $`A_\mu ^I`$ and $`\psi ^i`$ and their derivatives. They do not constrain the ghosts because one is dealing with the original gauge-covariant equations and not those of the gauge-fixed theory. This fact will turn out to be quite important later on.
#### 5.1.2 Noether identities
Because of gauge invariance, the left hand sides of the equations of motion are not all independent functions on $`J^{\mathrm{}}(E)`$, but they satisfy some relations, called Noether identities. These read
$$_{(\nu _1\mathrm{}\nu _k)}[D_\mu L_I^\mu +eL_iT_{Ij}^i\psi ^j]=0,$$
(5.1)
for all $`k=0,1,\mathrm{}`$ and $`I=1,\mathrm{},\mathrm{𝑑𝑖𝑚}(𝒢)`$.
#### 5.1.3 Statement of regularity conditions
The Yang-Mills Lagrangian $`L=(1/4)\delta _{IJ}F_{\mu \nu }^IF^{J\mu \nu }`$ fulfills important regularity conditions which we now spell out in detail. For all $`r`$, the collection $`R^r`$ of equations of motion can be split into two groups, the “independent equations” $`L_a`$ and the “dependent equations” $`L_\mathrm{\Delta }`$.<sup>4</sup><sup>4</sup>4By an abuse of terminology, we use “equations of motions” both for the actual equations and for their left hand sides. The independent equations are by definition such that they can be taken to be some of the coordinates of a new coordinate system on $`V^r`$, while the dependent equations hold as consequences of the independent ones: $`L_a=0`$ implies $`L_\mathrm{\Delta }=0`$. Furthermore, there is one and only one dependent equation for each Noether identity: these identities are the only relations among the equations and they are not redundant. To be precise, when viewed as equations for the $`L_a`$ and $`L_\mathrm{\Delta }`$, the Noether identities $`_{(\nu _1\mathrm{}\nu _k)}D_\mu L_I^\mu =0`$ are strictly equivalent to
$$L_\mathrm{\Delta }L_ak_\mathrm{\Delta }^a=0,$$
(5.2)
for some local functions $`k_\mathrm{\Delta }^a`$ of the fields and their derivatives (a complete set $`\{L_\mathrm{\Delta },L_a\}`$ is given explicitly below). Thus, the left hand sides of the Noether identities $`_{(\nu _1\mathrm{}\nu _k)}D_\mu L_I^\mu =0`$, viewed as equations for the $`L_a`$ and $`L_\mathrm{\Delta }`$, are of the form
$$_{(\nu _1\mathrm{}\nu _k)}D_\mu L_I^\mu =(L_\mathrm{\Delta }L_ak_\mathrm{\Delta }^a)_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta },$$
(5.3)
for some invertible matrix $`_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta }`$ that may depend on the dynamical variables (the range of $`(\nu _1\mathrm{}\nu _k)I`$ is equal to the range of $`\mathrm{\Delta }`$).
Similarly, one can split the gauge fields and their derivatives into “independent coordinates” $`y_A`$, which are not constrained by the equations of motion in the jet-spaces, and “dependent coordinates” $`z_a`$, which can be expressed in terms of the $`y_A`$ on the stationary surface(s). For fixed $`y_A`$, the change of variables from $`L_a`$ to $`z_a`$ is smooth and invertible (a complete set $`\{y_A,z_a\}`$ is given explicitly below).
The same regularity properties hold for the Chern-Simons action, or if one minimally couples scalar or spinor fields to the Yang-Mills potential as in the standard model.
The importance of the regularity conditions is that they will enable us to compute completely the homology of $`\delta `$ by identifying appropriate contractible pairs. We shall thus verify them explicitly.
We successively list the $`L_a,L_\mathrm{\Delta },y_A`$ and $`z_a`$ for the massless free Dirac field, for the massless free Klein-Gordon field, for pure Yang-Mills theory and for the Chern-Simons theory in three dimensions.
##### Dirac field:
We start with the simplest case, that of the free (real) Dirac field, with equations of motion $`\gamma ^\mu _\mu \mathrm{\Psi }=0`$. These equations are clearly independent (no Noether identity) and imply no restriction on the undifferentiated field components. So, the stationary surface in $`J^0(E)`$ is empty. The equations of motion start “being felt” in $`J^1(E)`$ since they are of the first order. One may rewrite them as $`_0\mathrm{\Psi }=\gamma ^0\gamma ^k_k\mathrm{\Psi }`$, so the derivatives $`_k\mathrm{\Psi }`$ may be regarded as independent, while $`_0\mathrm{\Psi }`$ is dependent. Similarly, the successive derivatives of $`_0\mathrm{\Psi }`$ may be expressed in terms of the spatial derivatives of $`\mathrm{\Psi }`$ by differentiating the equations of motion, so one gets the following decompositions:
$`\{L_a\}`$ $``$ $`\{,_\mu ,_{(\mu _1\mu _2)},\mathrm{}\},\{L_\mathrm{\Delta }\}\mathrm{is}\mathrm{empty},`$ (5.4)
$`\{y_A\}`$ $``$ $`\{\mathrm{\Psi },\mathrm{\Psi }_s,\mathrm{\Psi }_{(s_1s_2)},\mathrm{},\mathrm{\Psi }_{(s_1\mathrm{}s_m)},\mathrm{}\},`$ (5.5)
$`\{z_a\}`$ $``$ $`\{\mathrm{\Psi }_0,\mathrm{\Psi }_{(\rho _10)},\mathrm{},\mathrm{\Psi }_{(\rho _1\mathrm{}\rho _m0)},\mathrm{}\}.`$ (5.6)
The subset of equations $`R^r`$ relevant in $`J^r(E)`$ is given by the Dirac equations and their derivatives up to order $`r1`$.
##### Klein-Gordon field:
Again, the equation of motion $`_\mu ^\mu \varphi =0`$ and all its differential consequences are independent. Furthermore, they can clearly be used to express any derivative of $`\varphi `$ involving at least two temporal derivatives in terms of the other derivatives.
$`\{L_a\}`$ $``$ $`\{,_\mu ,_{(\mu _1\mu _2)},\mathrm{}\},\{L_\mathrm{\Delta }\}\mathrm{is}\mathrm{empty},`$ (5.7)
$`\{y_A\}`$ $``$ $`\{\varphi ,\varphi _\rho ,\varphi _{(s_1\rho )},\mathrm{},\varphi _{(s_1\mathrm{}s_m\rho )},\mathrm{}\},`$ (5.8)
$`\{z_a\}`$ $``$ $`\{\varphi _{(00)},\varphi _{(\rho _100)},\mathrm{},\varphi _{(\rho _1\mathrm{}\rho _m00)},\mathrm{}\}.`$ (5.9)
##### Pure Yang-Mills field:
The equations of motion $`L_I^\mu D_\nu F_I^{\nu \mu }=0`$ are defined in $`J^2(E)`$, where they are independent. The Noether identities involve the spacetime derivatives of the Euler-Lagrange derivatives of the pure Yang-Mills Lagrangian and so start playing a rôle in $`J^3(E)`$. They can be used to express $`_0`$ of the field equation for $`A_0^I`$ in terms of the other equations. Thus, in $`J^3(E)`$, $`L_I^\mu =0`$, $`_\rho L_I^m=0`$ and $`_kL_I^0=0`$ are independent equations, while $`_0L_I^0=0`$ are dependent equations following from the others. One can solve the equations $`L_I^m=0`$ and $`L_I^0=0`$ for $`A_{m|(00)}^I`$ and $`A_{0|(11)}^I`$ in terms of the other second order derivatives of the fields. Similarly, the derivatives of the equations $`L_I^m=0`$ can be solved for $`A_{m|(\rho _1\mathrm{}\rho _s00)}^I`$ while the independent derivatives of $`L_I^0=0`$ can be solved for $`A_{0|(s_1\mathrm{}s_n11)}^I`$. A possible split of the equations and the variables fulfilling the above requirements is therefore given by
$`\{L_a\}`$ $``$ $`\{L_I^\mu ,_\rho L_I^m,\mathrm{},_{(\rho _1\mathrm{}\rho _s)}L_I^m,\mathrm{},`$ (5.10)
$`_kL_I^0,\mathrm{},_{(k_1\mathrm{}k_2)}L_I^0,\mathrm{}\},`$
$`\{L_\mathrm{\Delta }\}`$ $``$ $`\{_0L_I^0,_\rho _0L_I^0,\mathrm{},_{(\rho _1\mathrm{}\rho _s)}_0L_I^0,\mathrm{}\},`$ (5.11)
$`\{y_A\}`$ $``$ $`\{A_\mu ^I,A_{\mu |\rho }^I,A_{m|(s_1\rho )}^I,\mathrm{},A_{m|(s_1\mathrm{}s_k\rho )}^I,\mathrm{},`$ (5.12)
$`A_{0|(\lambda 0)}^I,\mathrm{},A_{0|(\lambda _1\mathrm{}\lambda _k0)}^I,\mathrm{},`$
$`A_{0|(\overline{l}m)}^I,\mathrm{},A_{0|(\overline{l}_1\mathrm{}\overline{l}_km)}^I,\mathrm{}\}(\overline{l},\overline{l}_i>1),`$
$`\{z_a\}`$ $``$ $`\{A_{m|(00)}^I,A_{m|(\rho 00)}^I,\mathrm{},A_{m|(\rho _1\mathrm{}\rho _s00)}^I,\mathrm{},`$ (5.13)
$`A_{0|(11)}^I,\mathrm{},A_{0|(s11)}^I,\mathrm{},A_{0|(s_1\mathrm{}s_n11)}^I,\mathrm{}\}.`$
Finally, the matrix $`_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta }`$ in Eq. (5.3) associated with this split of the equations of motion is easily constructed: it is a triangular matrix with entries 1 on the diagonal and thus it is manifestly invertible. Indeed, one has, for the undifferentiated Noether identities, $`D_\mu L_I^\mu =\delta _I^J_0L_J^0+\text{ “more”}`$, where “more” denotes terms involving only the independent equuations. By differentiating these relations, one gets $`_{(\nu _1\mathrm{}\nu _k)}D_\mu L_I^\mu =\delta _I^J_{(\nu _1\mathrm{}\nu _k)}_0L_J^0+`$ “lower” $`+`$ “more”, where “lower” denotes terms involving the previous dependent equations.
##### Chern-Simons theory in three dimensions:
The equations are this time $`L_I^\mu \epsilon ^{\mu \rho \sigma }F_{I\rho \sigma }=0`$. They can be split as above since the Noether identities take the same form. But there are less independent field components since the equations of motion are of the first order and thus start being relevant already in $`J^1`$. From the equations $`L_I^i=0`$ and their derivatives, one can express the derivatives of the spatial components $`A_i^I`$ of the vector potential with at least one $`_0`$ in terms of the derivatives of $`A_0^I`$, which are unconstrained. Similarly, from the equations $`L_I^0=0`$ and their spatial derivatives, which are independent, one can express all spatial derivatives of $`A_2^I`$ with at least one $`_1`$ in terms of the spatial derivatives of $`A_1^I`$. Thus, we have $`L_a`$ and $`L_\mathrm{\Delta }`$ as in (5.10) and (5.11), but the $`y_A`$ and $`z_a`$ are now given by
$`\{y_A\}`$ $``$ $`\{A_\mu ^I,A_{0|\rho }^I,\mathrm{},A_{0|(\rho _1\mathrm{}\rho _k)}^I,\mathrm{},`$ (5.14)
$`A_{1|s}^I,\mathrm{},A_{1|(s_1\mathrm{}s_k)}^I,\mathrm{},`$
$`A_{2|2}^I,\mathrm{},A_{2|2\mathrm{}2}^I,\mathrm{}\},`$
$`\{z_a\}`$ $``$ $`\{A_{m|0}^I,A_{m|(0\rho )}^I,\mathrm{},A_{m|(0\rho _1\mathrm{}\rho _k)}^I,\mathrm{},`$ (5.15)
$`A_{2|1}^I,A_{2|(1s)}^I,\mathrm{},A_{2|(1s_1\mathrm{}s_k)}^I,\mathrm{}\}.`$
We have systematically used $`\mathrm{\Psi }_{(s\rho )}=_{(s\rho )}\mathrm{\Psi }`$ etc and $`A_{0|(s11)}^I=_{(s11)}A_0^I`$ etc (with $`s=1,\mathrm{},n1`$ and $`\rho =0,\mathrm{},n1`$). Note that the above splits are not unique. Furthermore, they are not covariant. We will in practice not use any of these splits. The only thing that is needed is the fact that such splits exist.
It is clear that the regularity conditions continue to hold if one minimally couples the Klein-Gordon or Dirac fields to the Yang-Mills potential since the coupling terms involve terms with fewer derivatives. Therefore, the regularity conditions hold in particular for the Lagrangian of the standard model.
For more general local Lagrangians of the Yang-Mills type, the regularity conditions are not automatic. For instance, they are not fulfilled in pure Chern-Simons theory in five dimensions because the equations of motion of that theory have no part linear in the fields and can therefore not be used as new admissible jet coordinates. The results on $`H(\delta )`$ derived in this section are valid only for Lagrangians fulfilling the regularity conditions.
For non-local Lagrangians of the type appearing in the discussion of effective field theories, the question of whether the regularity conditions are fulfilled does not arise since the equations of motion imply no restriction in the jet-spaces $`J^r(E)`$ of finite order. In some definite sense to be made precise below, one can say, however, that these theories also fulfill the regularity conditions.
#### 5.1.4 Weakly vanishing forms
An antifield independent local form vanishing when the equations of motion hold is said to be weakly vanishing. This is denoted by $`\omega 0`$. An immediate consequence of the regularity conditions is
###### Lemma 5.1
If an antifield independent local form $`\omega \mathrm{\Omega }_E`$ is weakly vanishing, $`\omega 0`$, it can be written as a linear combination of equations of motion with coefficients which are local forms, and is thus $`\delta `$-exact in the space $`\mathrm{\Omega }_F`$ of local forms,
$$\omega 0,\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(\omega )=0\omega =\delta \eta ,\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(\eta )=1.$$
##### Proof
In the coordinate system $`(x,dx,L_a,y_A)`$, $`\omega `$ satisfies $`\omega (x,dx,0,y_A)=0`$. Using a homotopy formula like in (4.6), one gets $`\omega =L_a_0^1𝑑\lambda [\frac{}{L_a}\omega ](x,dx,\lambda L_b,y_A)`$. Since the $`L_a`$ are equations of motion, there are antifield variables $`\varphi _a^{}`$ such that $`\delta \varphi _a^{}=L_a`$. This gives $`\omega =\delta \eta `$ where $`\eta =\varphi _a^{}_0^1𝑑\lambda [\frac{}{L_a}\omega ](x,dx,\lambda L_b,y_A)`$. Going back to the original coordinate system proves the lemma.
If both the equations of motion and the form $`\omega `$ are polynomial, $`\eta `$ is also polynomial.
### 5.2 Koszul-Tate resolution
Forms defined on the stationary surface can be viewed as equivalence classes of forms defined on the whole of jet-space modulo forms that vanish when the equations of motion hold. It turns out that the homology of $`\delta `$ is precisely given, in degree zero, by this quotient space. Furthermore, its homology in all other degrees is trivial. This is why one says that the Koszul-Tate differential implements the equations of motion.
More precisely, one has
###### Theorem 5.1
(Homology of $`\delta `$ in the algebra $`\mathrm{\Omega }_F`$ of local forms involving the original fields, the ghosts and the antifields)
The homology of $`\delta `$ in antifield number $`0`$ is given by the equivalence classes of local forms ($`\mathrm{\Omega }_E`$) modulo weakly vanishing ones, $`H_0(\delta ,\mathrm{\Omega }_F)=\{[\omega _0]\}`$, with $`\omega _0\omega _0^{}`$ if $`\omega _0\omega _0^{}0`$.
The homology of $`\delta `$ in strictly positive antifield number is trivial, $`H_m(\delta ,\mathrm{\Omega }_F)=0`$ for $`m>0`$.
In mathematical terminology, one says that the Koszul-Tate complex provides a “resolution” of the algebra of local forms defined on the stationary surface.
##### Proof
The idea is to exhibit appropriate contractible pairs using the regularity conditions. First, one can replace the jet-space coordinates $`A_\mu ^I,\psi ^i`$ and their derivatives by $`y_A`$ and $`L_a`$. As we have seen, this change of variables is smooth and invertible.
In the notation $`(a,\mathrm{\Delta })`$, the antifields $`A_\mu ^I,\psi _i^{}`$ and their derivatives are $`(\varphi _a^{},\varphi _\mathrm{\Delta }^{})`$ with $`\delta \varphi _a^{}=L_a`$ and $`\delta \varphi _\mathrm{\Delta }^{}=L_\mathrm{\Delta }`$. The second step is to redefine the antifields $`\varphi _\mathrm{\Delta }^{}`$ using the matrix $`_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta }`$ for the Noether identities (5.1) analogous to the matrix in (5.3) for the pure Yang-Mills case. One defines $`\stackrel{~}{\varphi }_{(\nu _1\mathrm{}\nu _k)I}^{}:=(\varphi _\mathrm{\Delta }^{}\varphi _a^{}k_\mathrm{\Delta }^a)_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta }`$. This definition makes the action of the Koszul-Tate differential particularly simple since one has $`\delta \stackrel{~}{\varphi }_{(\nu _1\mathrm{}\nu _k)I}^{}=0`$. Indeed $`(L_\mathrm{\Delta }L_ak_\mathrm{\Delta }^a)_{(\nu _1\mathrm{}\nu _k)I}^\mathrm{\Delta }`$ identically vanishes by the Noether identity. In fact, one has $`\stackrel{~}{\varphi }_{(\nu _1\mathrm{}\nu _k)I}^{}=\delta C_{I|(\nu _1\mathrm{}\nu _k)}^{}`$.
In terms of the new variables, the Koszul-Tate differential reads
$$\delta =L_a\frac{}{\varphi _a^{}}+\underset{k0}{}\stackrel{~}{\varphi }_{(\nu _1\mathrm{}\nu _k)I}^{}\frac{}{C_{I|(\nu _1\mathrm{}\nu _k)}^{}},$$
which makes it clear that $`L_a,\varphi _a^{},\stackrel{~}{\varphi }_{(\nu _1\mathrm{}\nu _k)I}^{}`$ and $`C_{I|(\nu _1\mathrm{}\nu _k)}^{}`$ form contractible pairs dropping from the homology. This leaves only the variables $`y_A`$, as well as the ghosts $`C^I`$ and their derivatives, as generators of the homology of $`\delta `$. In particular, the antifields disappear from the homology and there is thus no homology in strictly positive antifield number.
A crucial ingredient of the proof is the fact that the Noether identities are independent and exhaust all the independent Noether identities. This is what guaranteed the change of variables used in the proof of the theorem to be invertible. It allows one to generalize the theorem to theories fulfilling regularity conditions analogous to those of the Yang-Mills case. For theories with “dependent” Noether identities (“reducible case”), one must add further antifields at higher antifield number. With these additional variables, the theorem still holds. The homological rationale for the antifield spectrum is explained in .
##### Remarks:
(i) We stress that the equations of motion that appear in the theorem are the gauge-covariant equations of motion derived from the gauge-invariant Lagrangian $`L`$ (and not any gauge-fixed form of these equations).
(ii) When the Lagrangian is Lorentz-invariant, it is natural to regard the antifields $`A_I^\mu `$, $`\psi _i^{}`$ and $`C_I^{}`$ as transforming in the representation of the Lorentz group contragredient to the representation of $`A_\mu ^I`$, $`\psi ^i`$ and $`C^I`$, respectively. Thus, the $`A_I^\mu `$ are Lorentz vectors while the $`C_I^{}`$ are Lorentz scalars. Because $`\delta A_I^\mu `$, $`\delta \psi _i^{}`$ and $`\delta C_I^{}`$ have the same transformation properties as $`A_I^\mu `$, $`\psi _i^{}`$ and $`C_I^{}`$, $`\delta `$ commutes with the action of the Lorentz group. One can consider the homology of $`\delta `$ in the algebra of Lorentz-invariant local forms. Using that the Lorentz group is semi-simple, one checks that this homology is trivial in strictly positive antifield number, and given by the equivalence classes of Lorentz-invariant local forms modulo weakly vanishing ones in antifield number zero (alternatively one may verify this directly be means of the properties of the contracting homotopy used in the proof). Similar considerations apply to other linearly realized global symmetries of the Lagrangian.
(iii) Again, if the equations of motion are polynomial, theorem 5.1 holds in the algebra of local, polynomial forms.
### 5.3 Effective field theories
The results for the homology of $`\delta `$ in the Yang-Mills case extends to the analysis of effective field theories. The problem is to compute the homology of $`\delta `$ in the space of formal power series in the free parameters, generically denoted by $`g_\alpha `$, with coefficients that are local forms. We normalize the fields so that they have canonical dimensions. This means, in particular, that the Lagrangian takes the form
$$L=L_0+O(g_\alpha )$$
(5.16)
where the zeroth order Lagrangian $`L_0`$ is the free Lagrangian and is the sum of the standard kinetic term for free massless vector fields and of the free Klein-Gordon or Dirac Lagrangians.
Corresponding to this decomposition of $`L`$, there is a decomposition of $`\delta `$,
$$\delta =\delta _0+O(g_\alpha ),$$
(5.17)
where $`\delta _0`$ is the Koszul-Tate differential of the free theory. Now, $`\delta _0`$ is acyclic (no homology) in positive antifield number. The point is that this property passes on to the complete $`\delta `$. Indeed, let $`a`$ be a form which is $`\delta `$-closed, $`\delta a=0`$, and has positive antifield number. Expand $`a`$ according to the degree in the parameters, $`a=a_i+a_{i+1}+\mathrm{}`$. Since there are many parameters, $`a_j`$ is in fact the sum of independent monomials of degree $`j`$ in the $`g_\alpha `$’s. The terms $`a_i`$ of lowest order in the parameters must be $`\delta _0`$-closed, $`\delta _0a_i=0`$. But then, they are $`\delta _0`$-exact, $`a_i=\delta _0b_i`$, where $`b_i`$ is a local form (since $`a_i`$ has positive antifield number by assumption). This implies that $`a\delta b_i`$ starts at some higher order $`i^{}`$ ($`i^{}>i`$) if it does not vanish. By repeating the reasoning at order $`i^{}`$, and then successively at the higher orders, one sees that $`a`$ is indeed equal to the $`\delta `$ of a (in general infinite, formal) power series in the parameters, where each coefficient is a local form.
Similarly, at antifield number zero, a formal power series is $`\delta `$-exact if and only if it is a combination of the Euler-Lagrange derivatives of $`L`$ (such forms may be called “weakly vanishing formal power series”). Thus Theorem 5.1 holds also in the algebra of formal power series in the parameters with coefficients that are local forms, relevant to effective field theories. In that sense, effective theories fulfill the regularity conditions because the leading term $`L_0`$ does.
## 6 Conservation laws and symmetries: $`H(\delta |d)`$
In this chapter, we relate $`H(\delta |d)`$ to the characteristic cohomology of the theory. The argument is quite general and not restricted to gauge theories of the Yang-Mills type. It only relies on the fact that the Koszul-Tate complex provides a resolution of the algebra of local $`p`$-forms defined on the stationary surface, so that lemma 5.1 and theorem 5.1 hold for the gauge theory under study.
We then compute this cohomology for irreducible gauge theories in antifield number higher than $`2`$ on various assumptions and specialize the results to the Yang-Mills case.
### 6.1 Cohomological version of Noether’s first theorem
In this section, the fields $`\varphi ^i`$ are the original classical fields and $`_i`$ the Euler-Lagrange derivatives of $`L`$ with respect to $`\varphi ^i`$. The corresponding jet-spaces are denoted by $`J^r(D)`$, $`J^{\mathrm{}}(D)`$. In order to avoid cluttered formulas, we shall assume for simplicity that the fields are all bosonic. The inclusion of fermionic fields leads only to extra sign factors in the formulas below.
An infinitesimal field transformation is characterized by local functions $`\delta _Q\varphi ^i=Q^i(x,[\varphi ])`$, to which one associates the vector field
$$\stackrel{}{Q}=\underset{l0}{}_{(\mu _1\mathrm{}\mu _l)}Q^i\frac{}{\varphi _{(\mu _1\mathrm{}\mu _l)}^i}$$
on the jet-space $`J^{\mathrm{}}(D)`$. It commutes with the total derivative,
$$[_\mu ,\stackrel{}{Q}]=0.$$
(6.1)
The $`Q^i`$ are called the “characteristics” of the field transformation.
A symmetry of the theory is an infinitesimal field transformation leaving the Lagrangian invariant up to a total derivative:
$$\stackrel{}{Q}L\delta _QL=_\mu k^\mu ,$$
(6.2)
for some local functions $`k^\mu `$. One can rewrite this equation using integrations by parts as
$$Q^i_i+_\mu j^\mu =0,$$
(6.3)
for some local vector density $`j^\mu `$. This equation can also be read as
$$_\mu j^\mu 0,$$
(6.4)
which means that $`j^\mu `$ is a conserved current. This is just Noether’s result that to every symmetry there corresponds a conserved current. Note that this current could be zero in the case where $`Q^i=M^{[ij]}_j`$ (such $`Q^i`$ are examples of trivial symmetries, see below), which means that the correspondence is not one to one. On the other hand, one can associate to a given symmetry the family of currents $`j^\mu +_\nu k^{[\nu \mu ]}`$, which means that the correspondence is not onto either. As we now show, one obtains bijectivity by passing to appropriate quotient spaces.
Defining $`\omega _0^{n1}=\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}j^{\mu _n}`$, we can rewrite Eq. (6.4) in terms of the antifields as
$$d\omega _0^{n1}+\delta \omega _1^n=0.$$
(6.5)
This follows from lemma 5.1 and theorem 5.1, the superscript denoting the form degree and the subscript the antifield number (we do not write the pure ghost number because the ghosts do not enter at this stage; the pure ghost number is always zero in this and the next subsection). Because $`\delta `$ and $`d`$ anticommute, a whole class of solutions to this equation is provided by
$$\omega _0^{n1}=d\eta _0^{n2}+\delta \eta _1^{n1},$$
(6.6)
with $`\omega _1^n=d\eta _1^{n1}`$. This suggests to define equivalent conserved currents $`\omega _0^{n1}\omega _0^{n1}`$ as conserved currents that differ by terms of the form given in the right-hand side of (6.6). In other words, equivalence classes of conserved currents are just the elements of the cohomology group $`H_0^{n1}(d|\delta )`$ (defined through the cocycle condition (6.5) and the coboundary condition (6.6)). Expliciting the coboundary condition in dual notation, one thus identifies conserved currents which differ by identically conserved currents of the form $`_\nu k^{[\nu \mu ]}`$ modulo weakly vanishing currents,
$$j^\mu j^\mu +_\nu k^{[\nu \mu ]}+t^\mu ,t^\mu 0.$$
(6.7)
Equations (6.4) and (6.7) define the characteristic cohomology $`H_{\mathrm{char}}^{n1}`$ in form degree $`n1`$, which can thus be identified with $`H_0^{n1}(d|\delta )`$.
Let us now turn to symmetries of the theory. In a gauge theory, gauge transformations do not change the physics. It is therefore natural to identify two symmetries that differ by a gauge transformation. A general gauge transformation involves not only standard gauge transformations, but also, “trivial gauge transformations” that vanish on-shell (for a recent discussion, see e.g. ).
A trivial, local, gauge symmetry reads
$$\delta _M\varphi ^i=\underset{m,k0}{}()^k_{(\mu _1\mathrm{}\mu _k)}[M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}_{(\nu _1\mathrm{}\nu _m)}_j],$$
(6.8)
where the functions $`M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}`$ are arbitrary local functions antisymmetric for the exchange of the indices
$$M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}=M^{i(\mu _1\mathrm{}\mu _k)j(\nu _1\mathrm{}\nu _m)}.$$
(6.9)
It is direct to verify that trivial gauge symmetries leave the Lagrangian invariant up to a total derivative.
If
$$\delta _f\varphi ^i=\underset{l=0}{\overset{\overline{l}_\alpha }{}}R_\alpha ^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}f^\alpha ,$$
(6.10)
where the $`f^\alpha `$ are arbitrary local functions, provides a complete set of nontrivial gauge symmetries in the sense of <sup>5</sup><sup>5</sup>5For more information on complete sets of gauge transformations, see appendix A., then, the most general gauge symmetry is given by the sum of a transformation (6.10) and a trivial gauge transformation (6.8)
$$\delta _{f,M}\varphi ^i=\delta _f\varphi ^i+\delta _M\varphi ^i.$$
(6.11)
We thus define equivalent global symmetries as symmetries of the theory that differ by a gauge transformation of the form (6.11) with definite choices of the local functions $`f^\alpha `$ and $`M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}`$. The resulting quotient space is called the space of non trivial global symmetries.
The Koszul-Tate differential is defined through
$`\delta \varphi _i^{}`$ $`=`$ $`_i,`$ (6.12)
$`\delta C_\alpha ^{}`$ $`=`$ $`{\displaystyle \underset{l=0}{\overset{\overline{l}_\alpha }{}}}R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}\varphi _i^{},`$ (6.13)
where the $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}`$ are defined in terms of the $`R_\alpha ^{i(\mu _1\mathrm{}\mu _l)}`$ through
$$\underset{l=0}{\overset{\overline{l}_\alpha }{}}()^l_{(\mu _1\mathrm{}\mu _l)}[R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}f^\alpha ]=\underset{l=0}{\overset{\overline{l}_\alpha }{}}R_\alpha ^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}f^\alpha $$
(6.14)
and where the $`C_\alpha ^{}`$ are the antifields conjugate to the ghosts. For instance, for pure Yang-Mills theory, one has $`R_{\mu J}^I=ef_{KJ}^IA_\mu ^K=R_{\mu J}^{+I}`$, $`R_{\mu J}^{I\nu }=\delta _\mu ^\nu \delta _J^I=R_{\mu J}^{+I\nu }`$ and one recovers from (6.13) the formula (2.8) for $`\delta C_I^{}`$.
To any infinitesimal field transformation $`Q^i`$, one can associate a local $`n`$-form linear in the antifields $`\varphi _i^{}`$ through the formula $`\omega _1^n=d^nxa_1=d^nxQ^i\varphi _i^{}`$. Conversely, given an arbitrary local $`n`$-form of antifield number one, $`\omega _1^n=d^nxa_1=d^nx_{l0}a^{i(\mu _1\mathrm{}\mu _l)}\varphi _{i|(\mu _1\mathrm{}\mu _l)}^{}`$, one can add to it a $`d`$-exact term in order to remove the derivatives of $`\varphi _i^{}`$. The coefficient of $`\varphi _i^{}`$ in the resulting expression defines an infinitesimal transformation. Explicitly, $`Q^i=\frac{\delta a_1}{\delta \varphi _i^{}}=_{l0}()^l_{(\mu _1\mathrm{}\mu _l)}a^{i(\mu _1\mathrm{}\mu _l)}`$. There is thus a bijective correspondence between infinitesimal field transformations and equivalence classes of local $`n`$-forms of antifield number one (not involving the ghosts), where one identifies two such local $`n`$-forms that differ by a $`d`$-exact term.
Now, it is clear that $`Q^i`$ defines a symmetry of the theory if and only if the corresponding $`\omega _1^n`$’s are $`\delta `$-cocycles modulo $`d`$. In fact, one has
###### Lemma 6.1
Equivalence classes of global symmetries are in bijective correspondence with the elements of $`H_1^n(\delta |d)`$.
##### Proof:
The proof simply follows by expanding the most general local $`n`$-form $`a_2`$ of antifield number $`2`$, computing $`\delta a_2`$ and making integrations by parts. It is left to the reader. We only remark that the antisymmetry in (6.9) follows from the fact that the antifields are Grassmann odd.
This cohomological set-up allows to prove Noether’s first theorem in the case of (irreducible) gauge theories in a straightforward way, using theorems 4.2 and 5.1.
###### Theorem 6.1
The cohomology groups $`H_1^n(\delta |d)`$ and $`H_0^{n1}(d|\delta )`$ are isomorphic in spacetime dimensions $`n>1`$. In classical mechanics (n=1), they are isomorphic up to the constants, $`H_1^1(\delta |d)H_0^0(d|\delta )/`$.<sup>6</sup><sup>6</sup>6We derive and write cohomological results systematically for the case that the cohomology under study is computed over $``$. They hold analogously over $``$. In other words, there is an isomorphism between equivalence classes of global symmetries and equivalence classes of conserved currents (modulo constant currents when $`n=1`$).
##### Proof:
The proof relies on the triviality of the (co)homologies of $`\delta `$ and $`d`$ in appropriate degrees and follows a standard pattern. We define a mapping from $`H_1^n(\delta |d)`$ to $`H_0^{n1}(d|\delta )/\delta _0^{n1}`$ as follows. Let $`\omega _1^n`$ be a $`\delta `$-cocycle modulo $`d`$ in form-degree $`n`$ and antifield number $`1`$,
$$\delta \omega _1^n+d\omega _0^{n1}=0$$
(6.15)
for some $`\omega _0^{n1}`$ of form-degree $`(n1)`$ and antifield number $`0`$. Note that $`\omega _0^{n1}`$ is a $`d`$-cocycle modulo $`\delta `$. Furthermore, given $`\omega _1^n`$, $`\omega _0^{n1}`$ is defined up to a $`d`$-closed term, i.e., up to a $`d`$-exact term ($`n>1`$) or a constant ($`n=1`$) (algebraic Poincaré lemma). If one changes $`\omega _1^n`$ by a term which is $`\delta `$-exact modulo $`d`$, $`\omega _0^{n1}`$ is changed by a term which is $`d`$-closed modulo $`\delta `$. Formula (6.15) defines therefore a mapping from $`H_1^n(\delta |d)`$ to $`H_0^{n1}(d|\delta )/\delta _0^{n1}`$. This mapping is surjective because (6.15) is the cocycle condition both for $`H_1^n(\delta |d)`$ and for $`H_0^{n1}(d|\delta )`$. It is also injective because $`H_1^n(\delta )=0`$.
### 6.2 Characteristic cohomology and $`H(\delta |d)`$
We now consider the cohomology groups $`H_0^{np}(d|\delta )`$ for all values $`p=1,\mathrm{},n`$, and not just for $`p=1`$. Using again lemma 5.1 and theorem 5.1, these groups can be described independently of the antifields. In that context, they are known as the “characteristic cohomology groups” $`H_{\mathrm{char}}^{np}`$ of the stationary surface $`\mathrm{\Sigma }^{\mathrm{}}`$ and define the “higher order conservation laws”. For instance, for $`p=2`$, they are given, in dual notation, by “supercurrents” $`k^{[\mu \nu ]}`$ such that $`_\mu k^{[\mu \nu ]}0`$, where two such supercurrents are identified if they differ on shell by an identically conserved supercurrent: $`k^{[\mu \nu ]}k^{[\mu \nu ]}+_\lambda l^{[\lambda \mu \nu ]}+t^{[\mu \nu ]}`$, where $`t^{[\mu \nu ]}0`$.
In the same way, one generalizes non-trivial global symmetries by considering the cohomology groups $`H_k^n(\delta |d)`$, for $`k=1,2,\mathrm{}`$. These groups are referred to as “higher order (non-trivial) global symmetries”.
The definitions are such that the isomorphism between higher order conserved currents and higher order symmetries still holds:
###### Theorem 6.2
One has the following isomorphisms
$`k<n:`$ $`H_k^n(\delta |d)H_{k1}^{n1}(\delta |d)\mathrm{}H_1^{nk+1}(\delta |d)H_0^{nk}(d|\delta );`$ (6.17)
$`H_n^n(\delta |d)H_{n1}^{n1}(\delta |d)\mathrm{}H_1^1(\delta |d)H_0^0(d|\delta )/;`$
$`k>n:`$ $`H_k^n(\delta |d)H_{k1}^{n1}(\delta |d)\mathrm{}H_{kn+1}^1(\delta |d)H_{kn}^0(\delta )=0.`$ (6.18)
and
$$H_k^p(\delta |d)H_{k1}^{p1}(d|\delta )\mathrm{for}k>1\mathrm{and}0<pn.$$
(6.19)
In particular, the cohomology groups $`H_k^n(\delta |d)`$ $`(1kn)`$ are isomorphic to the characteristic cohomology groups $`H_0^{nk}(d|\delta )`$ (modulo the constants for $`k=n`$).
##### Proof:
One proves equation (6.19) and the last isomorphisms in (6.17), (6.17) as Theorem 6.1. The last equality in (6.18) holds because of the acyclicity of $`\delta `$ in all positive antifield numbers (see Section 5). The proof of the remaining isomorphisms illustrates the general technique of the descent equations of which we shall make ample use in the sequel. Let $`a_j^i`$ be a $`\delta `$-cocycle modulo $`d`$, $`\delta a_j^i+da_{j1}^{i1}=0`$, $`i>1`$, $`j>1`$. Then, $`d\delta a_{j1}^{i1}=0`$, from which one infers, using the triviality of $`d`$ in form-degree $`i1`$ ($`0<i1<n`$) that $`\delta a_{j1}^{i1}+da_{j2}^{i2}=0`$ for some $`a_{j2}^{i2}`$. Thus, $`a_{j1}^{i1}`$ is also a $`\delta `$-cocycle modulo $`d`$. If $`a_j^i`$ is modified by trivial terms ($`a_j^ia_j^i+\delta b_{j+1}^i+db_j^{i1}`$), then, $`a_{j1}^{i1}`$ is also modified by trivial terms. This follows again from $`H^{i1}(d)=0`$. Thus the “descent” $`[a_j^i][a_{j1}^{i1}]`$ from the class of $`a_j^i`$ in $`H_j^i(\delta |d)`$ to the class of $`a_{j1}^{i1}`$ in $`H_{j1}^{i1}(\delta |d)`$ defines a well-defined application from $`H_j^i(\delta |d)`$ to $`H_{j1}^{i1}(\delta |d)`$. This application is both injective (because $`H_j(\delta )=0`$) and surjective (because $`H_{j1}(\delta )=0`$). Hence, the groups $`H_j^i(\delta |d)`$ and $`H_{j1}^{i1}(\delta |d)`$ are isomorphic.
##### Remark:
The isomorphism $`H_1^{nk+1}(\delta |d)H_0^{nk}(d|\delta )`$ ($`n>k`$) uses $`H^{nk}(d)=0`$, which is true only in the space of forms with an explicit $`x`$-dependence. If one does not allow for an explicit $`x`$-dependence, $`H^{nk}(d)`$ is isomorphic to the space $`^{nk}`$ of constant forms. The last equality in (6.17) reads then $`H_1^{nk+1}(\delta |d)H_0^{nk}(d|\delta )/^{nk}`$.
The results on the groups $`H_k^p(\delta |d)`$ are summarized in the table below. The first row contains the characteristic cohomology groups $`H_{\mathrm{char}}^{np}`$ while $`(\mathrm{\Sigma })`$ corresponds to the local functionals defined on the stationary surface, i.e., the equivalence classes of $`n`$-forms depending on the original fields alone, where two such forms are identified if they differ, on the stationary surface, by a $`d`$-exact $`n`$-form, $`\omega _0^n\omega _0^n+d\eta _0^{n1}+\delta \eta _1^n`$. The characteristic cohomology group $`H_{\mathrm{char}}^0`$, in particular, contains the functions that are constant when the equations of motion hold. All the cohomology groups $`H_i^i(\delta |d)`$ along the principal diagonal are isomorphic to $`H_{\mathrm{char}}^0/`$; those along the parallel diagonals are isomorphic among themselves. The unwritten groups $`H_k^p(\delta |d)`$ with $`k>n`$ all vanish.
$`\begin{array}{ccccccccc}& & & & & & & & \\ k\backslash p& 0& 1& 2& \mathrm{}& \mathrm{}& n2& n1& n\\ & & & & & & & & \\ 0& H_{\mathrm{char}}^0/& H_{\mathrm{char}}^1& H_{\mathrm{char}}^2& & & H_{\mathrm{char}}^{n2}& H_{\mathrm{char}}^{n1}& (\mathrm{\Sigma })\\ & & & & & & & & \\ 1& 0& H_{\mathrm{char}}^0/& H_1^2& & & H_1^{n2}& H_1^{n1}& H_1^n\\ & & & & & & & & \\ 2& 0& 0& H_{\mathrm{char}}^0/& & & H_2^{n2}& H_2^{n1}& H_2^n\\ & & & & & & & & \\ \mathrm{}& & & & & & & & \\ & & & & & & & & \\ \mathrm{}& & & & & & & & \\ & & & & & & & & \\ n2& 0& 0& 0& & & H_{\mathrm{char}}^0/& H_{n2}^{n1}& H_{n2}^n\\ & & & & & & & & \\ n1& 0& 0& 0& & & 0& H_{\mathrm{char}}^0/& H_{n1}^n\\ & & & & & & & & \\ n& 0& 0& 0& & & 0& 0& H_{\mathrm{char}}^0/\end{array}`$
### 6.3 Ghosts and $`H(\delta |d)`$
So far, we have not taken the ghosts into account in the calculation of the homology of $`\delta `$ modulo $`d`$ (the ghosts are denoted by $`C^\alpha `$; in Yang-Mills theories one has $`C^\alpha C^I`$). These are easy to treat since they are not constrained by the equations of motion. As we have seen, $`\delta `$ acts trivially on them, $`\delta C^\alpha =0`$, and they do not occur in the $`\delta `$-transformation of any antifield. Let $`N_C`$ be the counting operator for the ghosts and their derivatives, $`N_C=_{n0}C_{(\mu _1\mathrm{}\mu _n)}^\alpha /C_{(\mu _1\mathrm{}\mu _n)}^\alpha `$. This counting operator is of course the pure ghost number. The vector space of local forms is the direct sum of the vector space of forms with zero pure ghost number ($`=`$ forms which do not depend on the $`C_{(\mu _1\mathrm{}\mu _n)}^\alpha `$) and the vector space of forms that vanish when one puts the ghosts and their derivatives equal to zero, $`\mathrm{\Omega }^{}=\mathrm{\Omega }_{N_C=0}^{}\mathrm{\Omega }_{N_C>0}^{}`$. In fact, $`\mathrm{\Omega }_{N_C>0}^{}`$ is itself the direct sum of the vector spaces of forms with pure ghost number one, two, etc. Since $`[N_C,\delta ]=0`$, $`\delta (\mathrm{\Omega }_{N_C=0}^{})`$ is included in $`\mathrm{\Omega }_{N_C=0}^{}`$ and $`\delta (\mathrm{\Omega }_{N_C>0}^{})`$ is included in $`\mathrm{\Omega }_{N_C>0}^{}`$.
###### Theorem 6.3
The cohomology of $`\delta `$ modulo $`d`$ in form degree $`n`$ and positive antifield number vanishes for forms in $`\mathrm{\Omega }_{N_C>0}^{}`$, $`H_k^n(\delta |d,\mathrm{\Omega }_{N_C>0}^{})=0`$ for $`k1`$.
##### Proof:
Let $`\omega _k=d^nxa_k`$, with $`k1`$, be a cycle of $`H_k^n(\delta |d,\mathrm{\Omega }_{N_C>0}^{})`$, $`\delta a_k+_\mu k^\mu =0`$. Because of theorem 4.1, $`\frac{\delta }{\delta C^\alpha }\delta a_k`$ = 0. Since $`\delta `$ does not involve the ghosts, this implies $`\delta (\frac{\delta }{\delta C^\alpha }a_k)=0`$. Theorem 5.1 then yields
$$\frac{\delta a_k}{\delta C^\alpha }=\delta b_{\alpha k+1}.$$
(6.21)
Now, by an argument similar to the one that leads to the homotopy formula (4.7), one finds that $`a_k`$ satisfies
$$a_k(\xi ,[C])a_k(\xi ,0)=_\mu k^\mu +_0^1\frac{d\lambda }{\lambda }[C^\alpha \frac{\delta a_k}{\delta C^\alpha }](\xi ,[\lambda C^\beta ])$$
(6.22)
where $`\xi `$ stands for $`x`$, the antifields, and all the fields but the ghosts. The second term in the left-hand side of (6.22) is zero when $`a_k`$ belongs to $`\mathrm{\Omega }_{N_C>0}^{}`$. Using this information and (6.21) in (6.22), together with $`[\delta ,C^\alpha ]=0`$, one finally gets that $`a_k=\delta b_{k+1}+_\mu k^\mu `$ for some $`b_{k+1},k^\mu `$.
Using that the isomorphisms of theorem 6.2 remain valid when the ghosts are included in $`\mathrm{\Omega }_{N_C>0}^{}`$ (because they are only based on the algebraic Poincaré lemma and on the vanishing homology of $`\delta `$ in all positive antifield numbers), we have
###### Corollary 6.1
The cohomology groups $`H_k^p(\delta |d,\mathrm{\Omega }_{N_C>0}^{})`$ and $`H_0^{nk}(d|\delta ,\mathrm{\Omega }_{N_C>0}^{})`$ vanish for all $`k1`$.
Note that the constant forms do not appear in $`H_0^{nk}(d|\delta ,\mathrm{\Omega }_{N_C>0}^{})`$, even if one considers forms with no explicit $`x`$-dependence. This is because the constant forms do not belong to $`\mathrm{\Omega }_{N_C>0}^{}`$.
To summarize, any mod-$`d`$ $`\delta `$-closed form can be decomposed as a sum of terms of definite pure ghost number, $`\omega =_l\omega ^l`$, where $`\omega ^l`$ has pure ghost number $`l`$. Each component $`\omega ^l`$ is $`\delta `$-closed modulo $`d`$. According to the above discussion, it is then necessarily $`\delta `$-exact modulo $`d`$, unless $`l=0`$.
### 6.4 General results on $`H(\delta |d)`$
#### 6.4.1 Cauchy order
In order to get additional vanishing theorems on $`H_k^n(\delta |d)`$, we need more information on the detailed structure of the theory.
An inspection of the split of the field variables in Eqs. (5.4) through (5.15) shows that for the Dirac and Klein-Gordon field, the set of independent variables $`\{y_A\}`$ is closed under spatial differentiation: $`_\mu y_A\{y_B\}`$ for $`\mu =1,\mathrm{}n1`$, while there are $`y_A`$ such that $`_0y_A`$ involves $`z_a`$. For the standard Yang-Mills and Chern-Simons theories, we have $`_\mu y_A\{y_B\}`$ for $`\mu =2,\mathrm{}n1`$, while there are $`y_A`$ such that $`_0y_A`$ or $`_1y_A`$ involves $`z_a`$.
We define the Cauchy order of a theory to be the minimum value of $`q`$ such that the space of local functions $`f(y)`$ is stable under $`_\mu `$ for $`\mu =q,q+1,\mathrm{},n1`$ (or, equivalently, $`_\mu y_A=f_{\mu A}(y)`$ for all $`A`$ and all $`\mu =q,\mathrm{},n1`$ where $`f_{\mu A}(y)`$ are local functions which can be expressed solely in terms of the $`y_A`$). The minimum is taken over all sets of space-time coordinates and all choices of $`\{y_A\}`$. The Dirac and Klein-Gordon theories are of Cauchy order one, while Chern-Simons and Yang-Mills theories are of Cauchy order two. The Lagrangian of the standard model defines therefore a theory of Cauchy order two.
The usefulness of the concept of Cauchy order lies in the following theorem.
###### Theorem 6.4
For theories of Cauchy order $`q`$, the characteristic cohomology is trivial for all form-degrees $`p=1,\mathrm{},nq1`$:
$$H_0^p(d|\delta )=\delta _0^p\text{for}p<nq.$$
Equivalently, among all cohomological groups $`H_k^n(\delta |d)`$ only those with $`kq`$ may possibly be nontrivial.
The proof of the theorem is given in the appendix 6.B.
In particular, for Klein-Gordon or Dirac theory, only $`H_0^{n1}(d|\delta )H_1^n(\delta |d)`$ may be nonvanishing (standard conserved currents), while for Yang-Mills or Chern-Simons theory, there can be in addition a nonvanishing $`H_0^{n2}(d|\delta )H_2^n(\delta |d)`$. We shall strengthen this result by showing that this latter group is in fact zero unless there are free abelian factors.
##### Remark:
The results on $`H_k^n(\delta |d)`$ in theorem 6.4 hold in the space of forms with or without an explicit coordinate dependence. By contrast, the results on the characteristic cohomology hold only in the space of $`x`$-dependent forms. If one restricts the forms to have no explicit $`x`$-dependence, there is additional cohomology: the constant forms encountered above are nontrivial even if one uses the field equations.
#### 6.4.2 Linearizable theories
Let $`N=N_\varphi +N_\varphi ^{}+N_C^{}`$, where $`N_\varphi =_{l0}\varphi _{(\lambda _1\mathrm{}\lambda _l)}^i\frac{}{\varphi _{(\lambda _1\mathrm{}\lambda _l)}^i}`$ (and similarly for the other fields), i.e., $`N`$ is the counting operator for the fields, the antifields and their derivatives. Decompose the Lagrangian $`L`$ and the reducibility functions $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}`$ according to the $`N`$-degree, $`L=_{n2}L^{(n)}`$, $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}=_{n0}(R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)})^{(n)}`$, $`\delta =_{n0}\delta ^{(n)}`$. So, $`L^{(2)}`$ is quadratic in the fields and their derivatives, $`L^{(3)}`$ is cubic etc, while $`\delta ^{(0)}`$ preserves the polynomial degree, $`\delta ^{(1)}`$ increases it by one unit, etc. We say that a gauge theory can be linearized if the cohomology of $`\delta ^{(0)}`$ (i) is trivial for all positive antifield numbers and (ii) is in antifield number $`0`$ given by the equivalence classes of local forms modulo forms vanishing on the surface in the jet space defined by the linearized equations of motion $`_{\mu _1\mathrm{}\mu _k}\frac{\delta L^{(2)}}{\delta \varphi ^j}=0`$. This just means that the field independent $`(R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)})^{(0)}`$ provide an irreducible generating set of Noether identities for the linearized theory.
The Lagrangian of the standard model is clearly linearizable since its quadratic piece is the sum of the Lagrangians for free Klein-Gordon, Dirac and $`U(1)`$ gauge fields. Pure Chern-Simons theory in three dimensions is linearizable too, and so are effective field theories sketched in Section 5.3.
One may view the condition of linearizability as a regularity condition on the Lagrangian, which is not necessarily fulfilled by all conceivable Lagrangians of the Yang-Mills type, although it is fulfilled in the cases met in practice in the usual physical models. An example of a non-linearizable theory is pure Chern-Simons theory in $`(2k+1)`$ dimensions with $`k>1`$. The lowest order piece of the Lagrangian is of order $`(k+1)`$ and so $`L^{(2)}=0`$ when $`k>1`$. The zero Lagrangian has a much bigger gauge symmetry than the Yang-Mills gauge symmetry. The non-linearizability of pure Chern-Simons theory in $`(2k+1)`$ dimensions ($`k>1`$) explains some of its pathologies. \[By changing the “background” from zero to a non-vanishing one, one may try to improve on this, but the issue will not be addressed here\].
###### Theorem 6.5
For irreducible linear gauge theories, (i) $`H_k^n(\delta |d)=0`$ for $`k3`$, (ii) if $`N_C^{}\omega _2^n=0`$, then $`\delta \omega _2^n+d\omega _1^{n1}=0`$ implies $`\omega _2^n=\delta \eta _3^n+d\eta _2^{n1}`$ and (iii) if $`\omega _1^n0`$, then $`\delta \omega _1^n+d\omega _0^{n1}=0`$ implies $`\omega _1^n=\delta \eta _2^n+d\eta _1^{n1}`$.
For irreducible linearizable gauge theories, the above results hold in the space of forms with coefficients that are formal power series in the fields, the antifields and their derivatives.
The proof is given in the appendix 6.B.
The theorem settles the case of effective field theories where the natural setting is the space of formal power series. In order to go beyond this and to make sure that the power series stop and are thus in fact local forms in the case of theories with a local Lagrangian, an additional condition is needed.
#### 6.4.3 Control of locality. Normal theories
For theorem 6.5 to be valid in the space of local forms, we need more information on how the derivatives appear in the Lagrangian. Let $`N_{}`$ be the counting operator of the derivatives of the fields and antifields, $`N_{}=N_\varphi +N_\varphi ^{}+N_C^{}`$, where $`N_\varphi =_kk\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i\frac{}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}`$ and similarly for the antifields. The equations of motions $`_i=0`$ are partial differential equations of order $`r_i`$ and gauge transformations involve a maximum of $`\overline{l}_\alpha `$ derivatives ($`\overline{l}_\alpha =1`$ for theories of the Yang-Mills type). We define
$$A=\underset{k0}{}\left[r_i\varphi _{i(\lambda _1\mathrm{}\lambda _k)}^{}\frac{}{\varphi _{i(\lambda _1\mathrm{}\lambda _k)}^{}}+m_\alpha C_{\alpha (\lambda _1\mathrm{}\lambda _k)}^{}\frac{}{C_{\alpha (\lambda _1\mathrm{}\lambda _k)}^{}}\right]$$
where $`m_\alpha =\overline{l}_\alpha +\mathrm{max}_{i,l,(\nu _1\mathrm{}\nu _l)}\{r_i+n_\varphi (R_\alpha ^{+i(\nu _1\mathrm{}\nu _l)})\}`$ with $`n_\varphi (R_\alpha ^{+i(\nu _1\mathrm{}\nu _l)})`$ the largest eigenvalue of $`N_\varphi `$ contained in $`R_\alpha ^{+i(\nu _1\mathrm{}\nu _l)}`$. In standard, pure Yang-Mills theory, $`A`$ reads explicitly
$$A=\underset{k0}{}\left[2A_{I(\lambda _1\mathrm{}\lambda _k)}^\mu \frac{}{A_{I(\lambda _1\mathrm{}\lambda _k)}^\mu }+3C_{I(\lambda _1\mathrm{}\lambda _k)}^{}\frac{}{C_{I(\lambda _1\mathrm{}\lambda _k)}^{}}\right],$$
(6.23)
since the equations of motion are of second order. For pure Chern-Simons theory in three dimensions, $`A`$ is
$$A=\underset{k0}{}\left[A_{I(\lambda _1\mathrm{}\lambda _k)}^\mu \frac{}{A_{I(\lambda _1\mathrm{}\lambda _k)}^\mu }+2C_{I(\lambda _1\mathrm{}\lambda _k)}^{}\frac{}{C_{I(\lambda _1\mathrm{}\lambda _k)}^{}}\right],$$
(6.24)
since the equations of motion are now of first order.
The degree $`K=N_{}+A`$ is such that $`[K,_\mu ]=[N_{},_\mu ]=_\mu `$ and
$`[K,\delta ]={\displaystyle \underset{k0}{}}[_{(\mu _1\mathrm{}\mu _k)}[(N_\varphi r_i)_i]{\displaystyle \frac{}{\varphi _{i(\mu _1\mathrm{}\mu _k)}^{}}}`$
$`+_{(\mu _1\mathrm{}\mu _k)}[{\displaystyle \underset{l0}{\overset{\overline{l}_\alpha }{}}}(N_\varphi +r_i+lm_\alpha )R_\alpha ^{+i(\nu _1\mathrm{}\nu _l)}\varphi _{i(\nu _1\mathrm{}\nu _l)}^{}]{\displaystyle \frac{}{C_{\alpha (\mu _1\mathrm{}\mu _k)}^{}}}].`$ (6.25)
It follows that $`\delta =_t\delta ^t`$, $`[K,\delta ^t]=t\delta ^t`$ with $`t0`$. For a linearizable theory, we have now two degrees: the degree of homogeneity in the fields, antifields and their derivatives, for which $`\delta `$ has only nonnegative eigenvalues and the $`K`$ degree, for which $`\delta `$ has only nonpositive eigenvalues.
A linearizable theory is called a normal theory if the homology of $`\delta ^{(0),0}`$ is trivial in positive antifield number. Let us define furthermore $`\delta ^{\mathrm{int},t}:=_{n1}\delta ^{(n),t}`$. Examples of normal theories are (i) pure Chern-Simons theory in three dimensions, (ii) pure Yang-Mills theory; (iii) standard model. For instance, in the first case, $`\delta ^{(0),0}`$ reduces to the Koszul-Tate differential of the $`U(1)^{\mathrm{𝑑𝑖𝑚}(G)}`$ Chern-Simons theory, while in the second case, it reduces to the Koszul-Tate differential for a set of free Maxwell fields. For these free theories, we have seen that theorem 5.1 holds, and thus, indeed, we have “normality” of the full theory.
###### Theorem 6.6
For normal theories, the results of theorem 6.5 extend to the space of forms with coefficients that are polynomials in the differentiated fields, the antifields and their derivatives and power series in the undifferentiated fields. Furthermore, if $`\delta ^{\mathrm{int},0}=0`$, they extend to the space of polynomials in the fields, the antifields and their derivatives.
The proof is given in the appendix 6.B.
The condition in the last part of this theorem is fulfilled by the Lagrangian of pure Chern-Simons theory, pure Yang-Mills theory or the standard model, because the interaction terms in the Lagrangian of those theories contain less derivatives than the quadratic terms. Thus theorem 6.6 holds in full in these cases. The condition would not be fulfilled if the theory contained for instance the local function $`\mathrm{exp}(_\mu ^\mu \varphi /k)`$.
We thus see that normal, local theories and effective theories have the same properties from the point of view of the cohomology groups $`H(\delta |d)`$. For this reason, the terminology “normal theories” will cover both cases in the sequel.
Remark. Part (iii) of theorems 6.5, 6.6 means that global symmetries with on-shell vanishing characteristics are necessarily trivial global symmetries in the sense of lemma 6.1. In particular, in the absence of non trivial Noether identities, weakly vanishing global symmetries are necessarily related to antisymmetric combinations of the equations of motions through integrations by parts.
#### 6.4.4 Global reducibility identities and $`H_2^n(\delta |d)`$
We define a “global reducibility identity” by a collection of local functions $`f^\alpha `$ such that they give a gauge transformation $`\delta _f\varphi ^i`$ as in Eq. (6.10) which is at the same time an on-shell trivial gauge symmetry $`\delta _M\varphi ^i`$ as in Eqs. (6.8). Explicitly a global reducibility identity requires thus
$$\underset{l=0}{\overset{\overline{l}_\alpha }{}}R_\alpha ^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}f^\alpha =\underset{k,m0}{}()^k_{(\mu _1\mathrm{}\mu _k)}[M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}_{(\nu _1\mathrm{}\nu _m)}_j]$$
(6.26)
for some local functions $`M^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}`$ with the antisymmetry property (6.9). Note that this is a stronger condition than just requiring that the transformations $`\delta _f\varphi ^i`$ vanish on-shell.
A global reducibility identity is defined to be trivial if all $`f^\alpha `$ vanish on-shell, $`f^\alpha 0`$, because $`f^\alpha 0`$ implies $`\delta _f\varphi ^i=\delta _M\varphi ^i`$ for some $`\delta _M\varphi ^i`$. This is seen as follows: $`f^\alpha 0`$ means $`f^\alpha =\delta g_1^\alpha `$ for some $`g_1^\alpha `$ and implies $`\varphi _i^{}\delta _f\varphi ^i=\delta (C_\alpha ^{}\delta g_1^\alpha )+_\mu ()^\mu =\delta [(\delta C_\alpha ^{})g_1^\alpha ]+_\mu ()^\mu `$; taking now the Euler-Lagrange derivative with respect to $`\varphi _i^{}`$ yields indeed $`\delta _f\varphi ^i=\delta _M\varphi ^i`$ because $`(\delta C_\alpha ^{})g_1^\alpha `$ is quadratic in the $`\varphi _{i(\mu _1\mathrm{}\mu _m)}^{}`$.
The space of nontrivial global reducibility identities is defined to be the space of equivalence classes of global reducibility identities modulo trivial ones.
###### Theorem 6.7
In normal theories, $`H_2^n(\delta |d)`$ is isomorphic to the space of non trivial global reducibility identities.
##### Proof:
Every cycle of $`H_2^n(\delta |d)`$ can be assumed to be of the form $`d^nxa_2`$ with $`a_2=C_\alpha ^{}f^\alpha +M_2+_\mu ()^\mu `$, where $`M_2=\frac{1}{2}_{n,m0}\varphi _{j(\nu _1\mathrm{}\nu _n)}^{}\varphi _{i(\mu _1\mathrm{}\mu _m)}^{}M^{j(\nu _1\mathrm{}\nu _n)i(\mu _1\mathrm{}\mu _m)}`$ such that (6.9) holds (indeed, all derivatives can be removed from $`C_\alpha ^{}`$ by subtracting a total derivative from $`a_2`$; the antisymmetry of the $`M`$’s follows from the odd grading of the $`\varphi _i^{}`$). Taking the Euler-Lagrange derivative of the cycle condition with respect to $`\varphi _i^{}`$ gives (6.26). Conversely, multiplying (6.26) by $`\varphi _i^{}`$ and integrating by parts an appropriate number of times yield $`\delta a_2+_\mu ()^\mu =0`$. Hence, cycles of $`H_2^n(\delta |d)`$ correspond to global reducibility identities and vice versa.
We still have to show that an element of $`H_2^n(\delta |d)`$ is trivial iff the corresponding global reducibility identity is trivial. The term $`b_3`$ in the coboundary condition $`a_2=\delta b_3+_\mu ()^\mu `$ contains terms with one $`C^{}`$ and one $`\varphi ^{}`$ and terms trilinear in $`\varphi ^{}`$’s. Taking the Euler-Lagrange derivative with respect to $`C_\alpha ^{}`$ of the coboundary condition implies $`f^\alpha 0`$, or $`f^\alpha =\delta g_1^\alpha `$. Conversely, $`f^\alpha =\delta g_1^\alpha `$ implies that $`a_2\delta (C_\alpha ^{}g_1^\alpha )`$ is a $`\delta `$-cycle modulo a total derivative in antifield number $`2`$, which does not depend on $`C_\alpha ^{}`$. Part (ii) of theorems 6.5 or 6.6 then implies that $`a_2\delta (C_\alpha ^{}g_1^\alpha )`$ is a $`\delta `$-boundary modulo a total derivative, and thus that $`d^nxa_2`$ is trivial in $`H_2^n(\delta |d)`$.
The same result applies to effective field theories since one can then use theorem 6.5.
#### 6.4.5 Results for Yang-Mills gauge models
For irreducible normal gauge theories, we have entirely reduced the computation of the higher order characteristic cohomology groups to properties of the gauge transformations.
We now perform explicitly the calculation of the global reducibility identities in the case of gauge theories of the Yang-Mills type, which are irreducible. We start with free electromagnetism, which has a non vanishing $`H_2^n(\delta |d)`$. The Koszul-Tate differential is defined on the generators by $`\delta A_\mu =0`$, $`\delta A^\mu =_\nu F^{\nu \mu }`$, $`\delta C^{}=_\mu A^\mu `$.
###### Theorem 6.8
For a free abelian gauge field with Lagrangian $`L=\frac{1}{4}F_{\mu \nu }F^{\mu \nu }`$ in dimensions $`n>2`$, $`H_2^n(\delta |d)`$ is represented by $`d^nxC^{}`$. A corresponding representative of the characteristic cohomology $`H_{\mathrm{char}}^{n2}`$ is $`F=\frac{1}{(n2)!2}dx^{\mu _1}\mathrm{}dx^{\mu _{n2}}ϵ_{\mu _1\mathrm{}\mu _n}F^{\mu _{n1}\mu _n}`$.
##### Proof:
According to theorem 6.7, $`H_2^n(\delta |d)`$ is determined by the nontrivial global reducibility conditions. A necessary condition for the existence of a global reducibility condition is that a gauge transformation $`\delta _fA_\mu =_\mu f`$ vanishes weakly, i.e. $`df0`$. Hence, $`f`$ is a cocycle of $`H_0^0(d|\delta )`$. By the isomorphisms (6.17) we have $`H_0^0(d|\delta )H_n^n(\delta |d)+`$. $`H_n^n(\delta |d)`$ vanishes for $`n>2`$ according to part (i) of theorem 6.5. Hence, if $`n>2`$, we conclude $`f\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}`$, i.e. the nontrivial global reducibility conditions are exhausted by constant $`f`$. The nontrivial representatives of $`H_2^n(\delta |d)`$ can thus be taken proportional to $`d^nxC^{}`$ if $`n>2`$ which proves the first part of the theorem. The second part of the theorem follows from the chain of equations $`\delta C^{}+_\mu A^\mu =0`$, $`\delta A^\mu _\nu F^{\nu \mu }=0`$ by the isomorphisms (6.17) (see also proof of these isomorphisms).
The reason that there is a non-trivial group $`H_2^n(\delta |d)`$ for free electromagnetism is that there is in that case a global reducibility identity associated with gauge transformations with constant gauge parameter. As the proof of the theorem shows, this property remains true if one includes gauge-invariant self-couplings of the Born-Infeld or Euler-Heisenberg type. The corresponding representatives of $`H_{\mathrm{char}}^{n2}`$ are obtained through the descent equations. Furthermore the result extends straightforwardly to models with a set of abelian gauge fields $`A_\mu ^I`$, $`I=1,2,\mathrm{}`$: then $`H_2^n(\delta |d)`$ is represented by $`d^nxC_I^{}`$, $`I=1,2,\mathrm{}`$
However, if one turns on self-couplings of the Yang-Mills type, which are not invariant under the abelian gauge symmetries, or if one includes minimal couplings to charged matter fields, the situation changes: there is no non-trivial reducibility identity any more. Indeed, gauge transformations leaving the Yang-Mills field $`A_\mu ^I`$ and the matter fields $`\psi ^i`$ invariant on-shell fulfill
$$D_\mu f^I0,f^IT_{Ij}^i\psi ^j0$$
(6.27)
whose only solution $`f^I([A],[\psi ])`$ is $`f^I0`$. By theorem 6.7, $`H_2^n(\delta |d)`$ vanishes in those cases. To summarize, we get the following result.
###### Corollary 6.2
For normal theories of the Yang-Mills type in dimensions $`n>2`$, the cohomology groups $`H_k^n(\delta |d)`$ vanish for $`k>2`$. The group $`H_2^n(\delta |d)`$ also vanishes, unless there are abelian gauge symmetries under which all matter fields are uncharged, in which case $`H_2^n(\delta |d)`$ is represented by those $`d^nxC_I^{}`$ which correspond to these abelian gauge symmetries.
This theorem covers pure Chern-Simons theory in three dimensions, pure Yang-Mills theory, the standard model as well as effective theories of the Yang-Mills type (this list is not exhaustive).
Finally, the group $`H_1^n(\delta |d)`$ is related to the standard conserved currents through theorem 6.1. Its dimension depends on the specific form of the Lagrangian, which may or may not have non trivial global symmetries. The complete calculation of $`H_1^n(\delta |d)`$ is a question that must be investigated on a case by case basis. For free theories, there is an infinite number of conserved currents. At the other extreme, for effective theories, which include all possible terms compatible with gauge symmetry and a definite set of global symmetries (such as Lorentz invariance), the only global symmetries and conservation laws should be the prescribed ones.
### 6.5 Comments
The characteristic cohomology associated with a system of partial differential equations has been investigated in the mathematical literature for some time . The connection with the Koszul-Tate differential is more recent . This new point of view has even enabled one to strengthen and generalize some results on the characteristic cohomology, such as the result on $`H_0^{n2}(d|\delta )`$ (isomorphic to $`H_2^n(\delta |d)`$). The connection with the reducibility properties of the gauge transformations was also worked out in this more recent work and turns out to be quite important for $`p`$-form gauge theories, where higher order homology groups $`H_k^n(\delta |d)`$ are non-zero .
The relation to the characteristic cohomology provides a physical interpretation of the nontrivial homology groups $`H(\delta |d)`$ in terms of conservation laws. In particular it establishes a useful cohomological formulation of Noether’s first theorem and a direct interpretation of the (nontrivial) homology groups $`H_1^n(\delta |d)`$ in terms of the (nontrivial) global symmetries. Technically, the use of the antifields allows one, among other things, to deal with trivial symmetries in a very efficient way. For instance the rather cumbersome antisymmetry property (6.9) of on-shell trivial symmetries is automatically reproduced through the coboundary condition in $`H_1^n(\delta |d)`$ thanks to the odd Grassmann parity of the antifields.
### 6.6 Appendix 6.A: Noether’s second theorem
We discuss in this appendix the general relationship between Noether identities, gauge symmetries and “dependent” field equations.
In order to do so, it is convenient to extend the jet-spaces by introducing a new field $`ϵ`$. A gauge symmetry on the enlarged jet-space is defined to be an infinitesimal field transformation $`\stackrel{}{Q}(ϵ)`$ leaving the Lagrangian invariant up to a total derivative,
$$\stackrel{}{Q}(ϵ)L_0+_\mu j^\mu (ϵ)=0.$$
(6.28)
The characteristic $`Q^i(ϵ)=_{l0}Q^{i(\mu _1\mathrm{}\mu _l)}ϵ_{(\mu _1\mathrm{}\mu _l)}`$ depends linearly and homogeneously on $`ϵ`$ and its derivatives $`ϵ_{(\mu _1\mathrm{}\mu _l)}`$ up to some finite order.
A “Noether operator” is a differential operator $`N^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}`$ that yields an identity between the equations of motion,
$$\underset{l0}{}N^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}_i=0.$$
(6.29)
We consider theories described by a Lagrangian that fulfills regularity conditions as described in section 5.1.3 (“irreducible gauge theories”). Namely, the original equations of motion are equivalent to a set of independent equations $`\{L_a\}`$ (which can be taken as coordinates in a new coordinate system on the jet-space) and to a set of dependent equations $`\{L_\mathrm{\Delta }\}`$ (which hold as a consequence of the independent ones). Explicitly, $`_{(\mu _1\mathrm{}\mu _l)}_i=L_\mathrm{\Delta }𝒩_{(\mu _1\mathrm{}\mu _l)i}^\mathrm{\Delta }+L_a𝒩_{(\mu _1\mathrm{}\mu _l)i}^a`$, where the matrix $`𝒩_{(\mu _1\mathrm{}\mu _l)i}^M`$, with $`M=\{a,\mathrm{\Delta }\}`$ is invertible.
Furthermore, we assume that the dependent equations are generated by a finite set $`\{L_\alpha \}`$ of equations (living on finite dimensional jet-spaces) through repeated differentiations, $`\{L_\mathrm{\Delta }\}\{L_\alpha ,_\rho L_\alpha ,_{(\rho _1\rho _2)}L_\alpha ,\mathrm{}\}`$, and that these successive derivatives are independent among themselves. For instance, the split $`\{L_\mathrm{\Delta },L_a\}`$ made in section 5.1.3 in the pure Yang-Mills case corresponds to $`\{L_\alpha \}\{_0L_I^0\}`$.
###### Lemma 6.2
Associated to the dependent equations of motions, there exists a set of Noether operators $`\{_{l0}^{\overline{l}_\alpha }R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}\}`$, which are non trivial, in the sense that they do not vanish weakly, and which are irreducible, in the sense that if $`_{m0}Z^{+\alpha (\nu _1\mathrm{}\nu _m)}_{(\nu _1\mathrm{}\nu _m)}[_{l0}^{\overline{l}_\alpha }R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}]0`$ (as an operator identity) then $`_{m0}Z^{+\alpha (\nu _1\mathrm{}\nu _m)}_{(\nu _1\mathrm{}\nu _m)}0`$ (i.e. all $`Z`$’s vanish weakly).
##### Proof:
Applying the equivalent of lemma 5.1 to the equations $`L_\alpha `$, we get $`L_\alpha =L_{\overline{a}}k_\alpha ^{\overline{a}}`$ where $`\{L_{\overline{a}}\}`$ is a finite subset of $`\{L_a\}`$. These are Noether identities whose left hand sides can be written in terms of the original equations of motion,
$`{\displaystyle \underset{l0}{\overset{\overline{l}_\alpha }{}}}R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}_i=L_\alpha L_{\overline{a}}k_\alpha ^{\overline{a}},`$ (6.30)
for some $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}`$. Note that the expression on the right hand side takes the form
$`L_\alpha L_{\overline{a}}k_\alpha ^{\overline{a}}=R_\alpha ^{+\beta }L_\beta +R_\alpha ^{+\overline{a}}L_{\overline{a}},`$ (6.31)
where $`R_\alpha ^{+\beta }=\delta _\alpha ^\beta `$ and $`R_\alpha ^{+\overline{a}}=k_\alpha ^{\overline{a}}`$, i.e., $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}(R_\alpha ^{+\beta },R_\alpha ^{+\overline{a}})`$. The presence of $`\delta _\alpha ^\beta `$ then implies the first part of the lemma.
Taking derivatives $`_{(\rho _1\mathrm{}\rho _m)}`$, $`m=0,1,\mathrm{}`$, of (6.30) we get the identities
$$_{(\rho _1\mathrm{}\rho _m)}[R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}_i]=[L_\mathrm{\Delta }L_ak_\mathrm{\Delta }^a]_{(\rho _1\mathrm{}\rho _m)\alpha }^\mathrm{\Delta },$$
(6.32)
for some functions $`k_\mathrm{\Delta }^a`$ and for an invertible matrix $`_{(\rho _1\mathrm{}\rho _m)\alpha }^\mathrm{\Delta }`$ analogous to the one in (5.3). The Noether identities $`L_\mathrm{\Delta }L_ak_\mathrm{\Delta }^a=0`$ are equivalent to $`R_\mathrm{\Delta }^{+\mathrm{\Gamma }}L_\mathrm{\Gamma }+R_\mathrm{\Delta }^{+a}L_a=0`$, where $`R_\mathrm{\Delta }^{+\mathrm{\Gamma }}=\delta _\mathrm{\Delta }^\mathrm{\Gamma }`$ and $`R_\mathrm{\Delta }^{+a}=k_\mathrm{\Delta }^a`$.
Thus, because of $`\delta _\mathrm{\Delta }^\mathrm{\Gamma }`$, if $`Z^{+\mathrm{\Delta }}(R_\mathrm{\Delta }^{+\mathrm{\Gamma }},R_\mathrm{\Delta }^{+a})0`$ then $`Z^{+\mathrm{\Delta }}0`$. The lemma follows from the (6.32), the fact that $`Z^{+\mathrm{\Delta }}`$ is related to $`Z^{+\alpha (\rho _1\mathrm{}\rho _m)}`$, $`m=0,1,\mathrm{}`$ through the invertible matrix $`_{(\rho _1\mathrm{}\rho _m)\alpha }^\mathrm{\Delta }`$ and the fact that the $`(L_\mathrm{\Delta },L_a)`$ are related to $`_{(\mu _1\mathrm{}\mu _l)}_i`$ through the invertible matrix $`𝒩_{(\mu _1\mathrm{}\mu _l)i}^M`$.
In terms of the equations $`L_\mathrm{\Delta }=0`$ and $`L_a=0`$, the acyclicity of the Koszul-Tate operator $`\delta C_\mathrm{\Delta }^{}=\varphi _\mathrm{\Delta }^{}\varphi _a^{}k_\mathrm{\Delta }^a`$, $`\delta \varphi _\mathrm{\Delta }^{}=L_\mathrm{\Delta }`$, $`\delta \varphi _a^{}=L_a`$ follows directly by introducing new generators $`\stackrel{~}{\varphi }_\mathrm{\Delta }^{}=\varphi _\mathrm{\Delta }^{}\varphi _a^{}k_\mathrm{\Delta }^a`$. Using both matrices $`𝒩_{(\mu _1\mathrm{}\mu _l)i}^M`$ and $`_{(\rho _1\mathrm{}\rho _m)\alpha }^\mathrm{\Delta }`$, one can verify that the Koszul-Tate operator is given in terms of the original equations by (6.13).
###### Lemma 6.3
The irreducible set of Noether operators associated to the dependent equations of motion is a generating set of non trivial Noether identities in the following sense: every Noether operator $`_{m0}N^{i(\mu _1\mathrm{}\mu _m)}_{(\mu _1\mathrm{}\mu _m)}`$ can be decomposed into the direct sum of
$$\underset{n0}{}Z^{+\alpha (\rho _1\mathrm{}\rho _n)}_{(\rho _1\mathrm{}\rho _n)}[\underset{l0}{\overset{\overline{l}}{}}R_\alpha ^{+i(\lambda _1\mathrm{}\lambda _l)}_{(\lambda _1\mathrm{}\lambda _l)}]$$
(6.33)
with $`Z^{+\alpha (\rho _1\mathrm{}\rho _n)}0`$, $`n=0,1,\mathrm{}`$ and the weakly vanishing piece
$$\underset{m,n0}{}M^{j(\nu _1\mathrm{}\nu _n)i(\mu _1\mathrm{}\mu _m)}[_{(\nu _1\mathrm{}\nu _n)}_j]_{(\mu _1\mathrm{}\mu _m)},$$
(6.34)
where $`M^{j(\nu _1\mathrm{}\nu _n)i(\mu _1\mathrm{}\mu _m)}`$ is antisymmetric in the exchange of $`j(\nu _1\mathrm{}\nu _n)`$ and $`i(\mu _1\mathrm{}\mu _m)`$.
##### Proof:
Every Noether identity can be written as a $`\delta `$ cycle in antifield number $`1`$, $`\delta (_mN^{i(\mu _1\mathrm{}\mu _m)}\varphi _{i|(\mu _1\mathrm{}\mu _m)}^{})=0`$, which implies because of theorem 5.1 that $`_mN^{i(\mu _1\mathrm{}\mu _m)}\varphi _{i|(\mu _1\mathrm{}\mu _m)}^{}=\delta b_2`$ with
$$b_2=\underset{n0}{}C_{\alpha |(\rho _1\mathrm{}\rho _n)}^{}Z^{+\alpha (\rho _1\mathrm{}\rho _n)}+\frac{1}{2}\underset{n,m0}{}\varphi _{j(\nu _1\mathrm{}\nu _n)}^{}\varphi _{i(\mu _1\mathrm{}\mu _m)}^{}M^{j(\nu _1\mathrm{}\nu _n)i(\mu _1\mathrm{}\mu _m)},$$
(6.35)
or explicitly
$`{\displaystyle \underset{m0}{}}N^{i(\mu _1\mathrm{}\mu _m)}\varphi _{i|(\mu _1\mathrm{}\mu _m)}^{}={\displaystyle \underset{n0}{}}Z^{+\alpha (\rho _1\mathrm{}\rho _n)}_{(\rho _1\mathrm{}\rho _n)}[{\displaystyle \underset{l0}{\overset{\overline{l}_\alpha }{}}}R_\alpha ^{+i(\lambda _1\mathrm{}\lambda _l)}\varphi _{i(\lambda _1\mathrm{}\lambda _l)}^{}]`$
$`+{\displaystyle \underset{m,n0}{}}M^{j(\nu _1\mathrm{}\nu _n)i(\mu _1\mathrm{}\mu _m)}[_{(\nu _1\mathrm{}\nu _n)}_j]\varphi _{i(\mu _1\mathrm{}\mu _m)}^{}.`$ (6.36)
Identification of the coefficients of the independent $`\varphi _{i|(\mu _1\mathrm{}\mu _m)}^{}`$ gives the result that every Noether operator can be written as the sum of (6.33) and (6.34). In order to prove that the decomposition is direct for non weakly vanishing $`Z^{+\alpha (\rho _1\mathrm{}\rho _n)}`$, we have to show that every weakly vanishing Noether identity can be written as in (6.34) (and in particular we have to show this for a Noether identity of the form (6.33), without sum over $`n`$ and $`Z^{+\alpha (\rho _1\mathrm{}\rho _n)}0`$ for this $`n`$). Using the set of indices $`(a,\mathrm{\Delta })`$, a weakly vanishing Noether identity is defined by $`N^aL_a+N^\mathrm{\Delta }L_\mathrm{\Delta }=0`$ with $`N^a0`$ and $`N^\mathrm{\Delta }0`$. The last equation implies as in the proof of lemma 5.1, that $`N^\mathrm{\Delta }=l^{\mathrm{\Delta }a}L_a`$ so that the Noether identity becomes $`(N^a+L_\mathrm{\Delta }l^{\mathrm{\Delta }a})L_a=0`$. In terms of the new generators $`\stackrel{~}{\varphi }_\mathrm{\Delta }^{}=\varphi _\mathrm{\Delta }^{}k_\mathrm{\Delta }^a\varphi _a^{}`$, $`\stackrel{~}{\varphi }_a^{}=\varphi _a^{}`$ the Koszul-Tate differential $`\delta =L_a\frac{}{\stackrel{~}{\varphi }_a^{}}+\stackrel{~}{\varphi }_\mathrm{\Delta }^{}\frac{}{C_\mathrm{\Delta }^{}}`$ involves only the contractible pairs. The Noether identity $`\delta [(N^a+L_\mathrm{\Delta }l^{\mathrm{\Delta }a})\stackrel{~}{\varphi }_a^{}]=0`$ then implies $`(N^a+L_\mathrm{\Delta }l^{\mathrm{\Delta }a})\stackrel{~}{\varphi }_a^{}=\delta [\frac{1}{2}\stackrel{~}{\varphi }_b^{}\stackrel{~}{\varphi }_a^{}\mu ^{[ba]}]`$. This proves the corollary, because we get $`N^a=\mu ^{[ba]}L_bL_\mathrm{\Delta }l^{\mathrm{\Delta }a}`$ and $`N^\mathrm{\Delta }=l^{\mathrm{\Delta }a}L_a`$.
###### Theorem 6.9
(Noether’s second theorem) To every Noether operator $`_{l0}N^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}`$ there corresponds a gauge symmetry $`\stackrel{}{Q}(ϵ)`$ given by $`Q^i(ϵ)=_{l0}()^l_{(\mu _1\mathrm{}\mu _l)}[N^{i(\mu _1\mathrm{}\mu _l)}ϵ]`$ and, vice versa, to every gauge symmetry $`Q^i(ϵ)`$, there corresponds the Noether operator defined by $`_{l0}Q^{+i(\nu _1\mathrm{}\nu _l)}_{(\nu _1\mathrm{}\nu _l)}_{l0}()^l_{(\mu _1\mathrm{}\mu _l)}[Q^{i(\mu _1\mathrm{}\mu _l)}]`$.
##### Proof:
The first part follows by multiplying the Noether identity
$$\underset{l0}{}N^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}_i$$
(6.37)
by $`ϵ`$ and then removing the derivatives from the equations of motion by integrations by parts to get $`_{l0}()^l_{(\mu _1\mathrm{}\mu _l)}[ϵN^{i(\mu _1\mathrm{}\mu _l)}]_i+_\mu j^\mu =0`$, which can be transformed to (6.28).
The second part follows by starting from (6.28) and doing the reverse integrations by parts to get $`ϵ(_{l0}()^l_{(\mu _1\mathrm{}\mu _l)}[Q^{i(\mu _1\mathrm{}\mu _l)}_i])+_\mu j^{\prime \prime \mu }=0`$. Taking the Euler-Lagrange derivatives with respect to $`ϵ`$, which annihilates the total derivative according to theorem 4.1, proves the theorem.
The gauge transformations associated with a generating (or “complete”) set of Noether identities are said to form a generating (or complete) set of gauge symmetries.
Trivial gauge symmetries are defined as those that correspond to weakly vanishing Noether operators:
$`Q_T^i(ϵ)={\displaystyle \underset{m,k0}{}}()^k_{(\mu _1\mathrm{}\mu _k)}[_{(\nu _1\mathrm{}\nu _m)}_jM^{j(\nu _1\mathrm{}\nu _m)i(\mu _1\mathrm{}\mu _k)}ϵ]`$
$`={\displaystyle \underset{m,n0}{}}M^{+j(\nu _1\mathrm{}\nu _m)i(\lambda _1\mathrm{}\lambda _n)}_{(\nu _1\mathrm{}\nu _m)}_jϵ_{(\lambda _1\mathrm{}\lambda _n)},`$ (6.38)
where the last equation serves as the definition of the functions $`M^{+j(\nu _1\mathrm{}\nu _m)i(\lambda _1\mathrm{}\lambda _n)}`$.
Note that trivial gauge symmetries do not only vanish weakly, they are moreover related to antisymmetric combinations of equations of motions through integrations by parts.
Non trivial gauge transformations are defined as gauge transformations corresponding to non weakly vanishing Noether identities. In particular, the gauge transformations corresponding to the generating set constructed above are given by
$$R_\alpha ^i(ϵ)=\underset{l0}{\overset{\overline{l}_\alpha }{}}()^l_{(\mu _1\mathrm{}\mu _l)}[R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}ϵ]=\underset{l0}{\overset{\overline{l}_\alpha }{}}R_\alpha ^{i(\mu _1\mathrm{}\mu _l)}_{(\mu _1\mathrm{}\mu _l)}ϵ.$$
(6.39)
The operator $`Z^{+\alpha }_{m0}Z^{+\alpha (\rho _1\mathrm{}\rho _m)}_{(\rho _1\mathrm{}\rho _m)}0`$ iff the operator $`Z^\alpha _{m0}()^m_{(\rho _1\mathrm{}\rho _m)}Z^{+\alpha (\rho _1\mathrm{}\rho _m)}0`$. A direct consequence of theorem 6.9 and lemma 6.3 is then
###### Corollary 6.3
Every gauge symmetry $`Q^i(ϵ)`$ can be decomposed into the direct sum $`Q^i(ϵ)=R_\alpha ^i(Z^\alpha (ϵ))+Q_T^i(ϵ)`$, where the operator $`Z^\alpha `$ is not weakly vanishing, while $`Q_T^i(ϵ)`$ is weakly vanishing. Furthermore, $`Q_T^i(ϵ)`$ is related to an antisymmetric combination of equations of motion through integrations by parts.
It is in that sense that a complete set of gauge transformations generates all gauge symmetries.
### 6.7 Appendix 6.B: Proofs of theorems 6.4, 6.5 and 6.6
##### Proof of theorem 6.4:
We decompose the spacetime indices into two subsets, $`\{\mu \}=\{a,\mathrm{}\}`$ where $`a=0,\mathrm{},q1`$ and $`\mathrm{}=q,\mathrm{},n1`$. The cocycle condition $`da+\delta b=0`$ decomposes into
$$d^1a^M+\delta b^{M+1}=0,d^0a^M+d^1a^{M1}+\delta b^M=0,\mathrm{}$$
(6.40)
where the superscript is the degree in the $`dx^{\mathrm{}}`$ and $`M`$ is the highest degree in the decomposition of $`a`$,
$$a=\underset{mM}{}a^m,d^1=dx^{\mathrm{}}_{\mathrm{}},d^0=dx^a_a.$$
Note that $`M`$ cannot exceed $`nq`$ because the $`dx^{\mathrm{}}`$ anticommute. Without loss of generality we can assume that $`a`$ depends only on the $`y_A`$, $`dx^\mu `$, $`x^\mu `$ because this can be always achieved by adding a $`\delta `$-exact piece to $`a`$ if necessary. In particular, $`a^M`$ can thus be assumed to be of the form
$$a^M=dx^\mathrm{}_1\mathrm{}dx^\mathrm{}_Mf_{\mathrm{}_1\mathrm{}\mathrm{}_M}(dx^a,x^\mu ,y_A).$$
Since we assume that the theory has Cauchy order $`q`$, $`d^1a^M`$ depends also only on the $`y_A`$, $`dx^\mu `$, $`x^\mu `$ and therefore vanishes on-shell only if it vanishes even off-shell. The first equation in (6.40) implies thus
$$d^1a^M=0.$$
(6.41)
To exploit this equation, we need the cohomology of $`d^1`$. It is given by a variant of the algebraic Poincaré lemma in section 4.5 and can be derived by adapting the derivation of that lemma as follows. Since $`d^1`$ contains only the subset $`\{_{\mathrm{}}\}`$ of $`\{_\mu \}`$, the jet coordinates $`_{(a_1\mathrm{}a_k)}\varphi ^i`$ play now the same rôle as the $`\varphi ^i`$ in the derivation of the algebraic Poincaré lemma (it does not matter that the set of all $`_{(a_1\mathrm{}a_k)}\varphi ^i`$ is infinite because a local form contains only finitely many jet coordinates). The $`dx^a`$ and $`x^a`$ are inert to $`d^1`$ and play the rôle of constants. Forms of degree $`nq`$ in the $`dx^{\mathrm{}}`$ take the rôle of the volume forms. One concludes that the $`d^1`$-cohomology is trivial in all $`dx^{\mathrm{}}`$-degrees $`1,\mathrm{},nq1`$ and in degree 0 represented by functions $`f(dx^a,x^a)`$. Eq. (6.41) implies thus:
$`0<M<nq:`$ $`a^M=d^1\eta ^{M1};`$
$`M=0:`$ $`a^0=f(dx^a,x^a).`$ (6.42)
In the case $`0<M<nq`$ we introduce $`a^{}:=ad\eta ^{M1}`$ which is equivalent to $`a`$. Since $`a^{}`$ contains only pieces with $`dx^{\mathrm{}}`$-degrees strictly smaller than $`M`$, one can repeat the arguments until the $`dx^{\mathrm{}}`$-degree drops to zero. In the case $`M=0`$, one has $`a=a^0=f(dx^a,x^a)`$ and the cocycle condition imposes $`d^0f(dx^a,x^a)0`$. This requires $`d^0f(dx^a,x^a)=0`$ which implies $`f(dx^a,x^a)=\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}+dg(dx^a,x^a)`$ by the ordinary Poincaré lemma in $`^q`$. Hence, up to a constant, $`a`$ is trivial in $`H(d|\delta )`$ whenever $`M<nq`$. This proves the theorem because one has $`M<nq`$ whenever the form-degree of $`a`$ is smaller than $`nq`$ since $`M`$ cannot exceed the form-degree.
##### Proof of theorem 6.5:
Let us first establish two additional properties satisfied by Euler-Lagrange derivatives. These are
$$\underset{k0}{}()^k_{(\lambda _1\mathrm{}\lambda _k)}\left[\frac{(_\mu f)}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}}g\right]=\underset{k0}{}()^k_{(\lambda _1\mathrm{}\lambda _k)}\left[\frac{f}{\varphi _{(\lambda _1\mathrm{}\lambda _k})}_\mu g\right],$$
(6.43)
for any local functions $`f,g`$ and
$`\stackrel{}{Q}{\displaystyle \frac{\delta f}{\delta \varphi ^j}}=()^{ϵ_jϵ_Q}{\displaystyle \frac{\delta }{\delta \varphi ^j}}\left[Q^i{\displaystyle \frac{\delta f}{\delta \varphi ^i}}\right]()^{ϵ_jϵ_Q}{\displaystyle \underset{k0}{}}()^k_{(\lambda _1\mathrm{}\lambda _k)}\left[{\displaystyle \frac{Q^i}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^j}}{\displaystyle \frac{\delta f}{\delta \varphi ^i}}\right]`$ (6.44)
for local functions $`f`$ and $`Q^i`$ ($`ϵ_i`$ and $`ϵ_Q`$ denote the Grassmann parities of $`\varphi ^i`$ and $`\stackrel{}{Q}`$ respectively). Indeed, using (4.8), the left hand side of (6.43) is
$$\underset{k0}{}()^k_{(\lambda _1\mathrm{}\lambda _k)}\left[_\mu (\frac{f}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}})g\right]+\underset{k0}{}()^k_{(\nu _1\mathrm{}\nu _{k1}\mu )}\left[\frac{f}{\varphi _{(\nu _1\mathrm{}\nu _{k1})}}g\right].$$
Integrating by parts the $`_\mu `$ in the first term and using the same cancellation as before in (4.9) gives (6.43). Similarly, because of (6.1), the left hand side of (6.44) is $`_{k0}()^k_{(\lambda _1\mathrm{}\lambda _k)}[\stackrel{}{Q}(\frac{f}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^j})]`$. Commuting $`\stackrel{}{Q}`$ with $`\frac{}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^j}`$ gives (6.44): the terms with $`\stackrel{}{Q}`$ and $`\frac{}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^j}`$ in reverse order reproduce the first term on the right hand side of (6.44) due to theorem 4.1, while the commutator terms yield the second term upon repreated use of Eq. (6.43).
Let us now turn to the proof of the theorem. Using $`\omega _k=d^nxa_k`$, the cocycle condition reads $`\delta a_k+_\mu j^\mu =0`$. Using theorem 4.1, the Euler-Lagrange derivatives of this condition with respect to a field and or antifield $`Z`$ gives
$$\frac{\delta }{\delta Z}\underset{\mathrm{\Phi }^{}=C_\alpha ^{},\varphi _i^{}}{}(\delta \mathrm{\Phi }^{})\frac{\delta a_k}{\delta \mathrm{\Phi }^{}}=0.$$
(6.45)
Using Eq. (6.44) (for $`\stackrel{}{Q}\delta `$), the previous formula is now exploited for $`ZC_\alpha ^{}`$, $`Z\varphi _i^{}`$ and $`Z\varphi ^i`$. For $`ZC_\alpha ^{}`$ it gives
$$\delta \frac{\delta a_k}{\delta C_\alpha ^{}}=0.$$
When $`k3`$, $`\delta a_k/\delta C_\alpha ^{}`$ has positive antifield number. Due to the acyclicity of $`\delta `$ in positive antifield number, the previous equation gives
$$\frac{\delta a_k}{\delta C_\alpha ^{}}=\delta \sigma _{k1}^\alpha .$$
(6.46)
For the proof of part (ii) of the theorem, we note that this relation holds trivially with $`\sigma _{k1}^\alpha =0`$. For $`Z\varphi _i^{}`$, Eq. (6.45) gives
$$\delta \frac{\delta a_k}{\delta \varphi _i^{}}=R_\alpha ^i(\frac{\delta a_k}{\delta C_\alpha ^{}})$$
where we used the same notation as in Eq. (6.39). Using (6.46) in the previous equation, and once again the acyclicity of $`\delta `$ in positive antifield number gives
$$\frac{\delta a_k}{\delta \varphi _i^{}}=R_\alpha ^i(\sigma _{k1}^\alpha )+\delta \sigma _k^i.$$
(6.47)
For the proof of part (iii) of the theorem, we note that this relation holds with $`\sigma _{k1}^\alpha =0`$ because by assumption $`\delta a_1/\delta \varphi _i^{}`$ is weakly vanishing, and so it is $`\delta `$-exact. Finally Eq. (6.45) gives for $`Z\varphi ^i`$, using (6.44) and (6.43) repeatedly,
$`\delta {\displaystyle \frac{\delta a_k}{\delta \varphi ^i}}`$ $`=`$ $`{\displaystyle \underset{k0}{}}()^k_{(\lambda _1\mathrm{}\lambda _k)}[{\displaystyle \frac{_j}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}}{\displaystyle \frac{\delta a_k}{\delta \varphi _j^{}}}`$
$`+\varphi _j^{}{\displaystyle \underset{l0}{}}{\displaystyle \frac{R_\alpha ^{j(\mu _1\mathrm{}\mu _l)}}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}}_{(\mu _1\mathrm{}\mu _l)}{\displaystyle \frac{\delta a_k}{\delta C_\alpha ^{}}}].`$
One now inserts (6.46), (6.47) in the previous equation and uses
$$\underset{k0}{}\underset{l0}{}()^k_{(\lambda _1\mathrm{}\lambda _k)}\left[\frac{(R_\alpha ^{j(\mu _1\mathrm{}\mu _l)}_j)}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}_{(\mu _1\mathrm{}\mu _l)}\sigma _{k1}^\alpha \right]=0,$$
which follows from repeated application of (6.43) and the fact that $`R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)}`$ defines a Noether identity. This gives an expression $`\delta (\mathrm{})=0`$. Using then acyclicity of $`\delta `$ in positive antifield number, one gets $`(\mathrm{})=\delta \sigma _{ik+1}`$:
$`{\displaystyle \frac{\delta a_k}{\delta \varphi ^i}}`$ $`=`$ $`{\displaystyle \underset{k0}{}}()^k_{(\lambda _1\mathrm{}\lambda _k)}[{\displaystyle \frac{_j}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}}\sigma _k^j`$ (6.48)
$`{\displaystyle \underset{l0}{}}{\displaystyle \frac{(R_\alpha ^{+j(\mu _1\mathrm{}\mu _l)}\varphi _{j(\mu _1\mathrm{}\mu _l)}^{})}{\varphi _{(\lambda _1\mathrm{}\lambda _k)}^i}}\sigma _{k1}^\alpha ]+\delta \sigma _{ik+1}.`$
On the other hand, we have
$$Na_k=\varphi ^i\frac{\delta a_k}{\delta \varphi ^i}+\varphi _i^{}\frac{\delta a_k}{\delta \varphi _i^{}}+C_\alpha ^{}\frac{\delta a_k}{\delta C_\alpha ^{}}+_\mu j^\mu .$$
(6.49)
Using (6.46)-(6.48), integrations by parts and (6.44), we get
$`Na_k`$ $`=`$ $`\delta (\varphi ^i\sigma _{ik+1}\varphi _i^{}\sigma _k^i+C_\alpha ^{}\sigma _{k1}^\alpha )+_\mu j_{}^{}{}_{}{}^{\mu }`$ (6.50)
$`+\left[2_j{\displaystyle \frac{\delta (N_\varphi L)}{\delta \varphi ^j}}\right]\sigma _k^j+{\displaystyle \underset{l0}{}}(N_\varphi R_\alpha ^{+i(\mu _1\mathrm{}\mu _l)})\varphi _{i(\mu _1\mathrm{}\mu _l)}^{}\sigma _{k1}^\alpha .`$
If the theory is linear, the two terms in the last line vanish. We can then use this result in the homotopy formula $`a_k=_0^1\frac{d\lambda }{\lambda }[Na_k](x,\lambda \varphi ,\lambda \varphi ^{},\lambda C^{})`$ and the fact that $`\delta =\delta ^{(0)}`$ and $`_\mu `$ are homogeneous of degree $`0`$ in $`\lambda `$ to conclude that $`a_k=\delta ()+_\mu ()^\mu `$. This ends the proof in the case of irreducible linear gauge theories.
If a theory is linearizable, we decompose $`a_k`$, with $`k1`$ into pieces of definite homogeneity $`n`$ in all the fields, antifields and their derivatives, $`a_k=_{nl}a_k^{(n)}`$ where $`l2`$ due to the assumptions of the theorem. We then use the acyclicity of $`\delta ^{(0)}`$ to show that if $`c_k=\delta e_{k+1}`$ with the expansion of $`c`$ starting at homogeneity $`l1`$, then the expansion of $`e`$ can be taken to start also at homogeneity $`l`$. Indeed, we have $`\delta ^{(0)}e_{k+1}^{(1)}=0`$, $`\delta ^{(0)}e_{k+1}^{(2)}+\delta ^{(1)}e_{k+1}^{(1)}=0,\mathrm{},`$ $`\delta ^{(0)}e_{k+1}^{(l1)}+\mathrm{}+\delta ^{(l2)}e_{k+1}^{(1)}=0`$. The first equation implies $`e_{k+1}^{(1)}=\delta ^{(0)}f_{k+2}^{(1)}`$, so that the redefinition $`e_{k+1}\delta f_{k+2}^{(1)}`$, which does not modify $`c_k`$ allows to absorb $`e_{k+1}^{(1)}`$. This process can be continued until $`e_{k+1}^{(l1)}`$ has been absorbed.
Hence, we can choose $`\sigma _{ik+1},\sigma _k^i,\sigma _{k1}^\alpha `$ to start at homogeneity $`l1`$. This implies that the two last terms in (6.50) are of homogeneity $`l+1`$. Due to $`a_k=\frac{1}{l}Na_k+_{n>l}a^{(n)}`$, Eq. (6.50) yields $`a_k=\delta ()+_\mu ()^\mu +a_k^{\prime \prime }`$, where $`a_k^{\prime \prime }`$ starts at homogeneity $`l+1`$ (unless it vanishes). Going on recursively proves the theorem.
##### Proof of theorem 6.6:
In the space of forms which are polynomials in the derivatives of the fields, the antifields and their derivatives with coefficients that are power series in the fields, the $`K`$ degree is bounded. It is of course also bounded in the space of forms that are polynomials in the undifferentiated fields as well. We can use the acyclicity of $`\delta ^{(0),0}`$ to prove the acyclicity of $`\delta ^0`$ in the respective spaces. Indeed, suppose that $`c`$ is of strictly positive antifield number, its polynomial expansion starts with $`l`$ and its $`K`$ bound is $`M`$. From $`\delta ^0c_M=0`$ it follows that $`\delta ^{(0),0}c_{l,M}=0`$ and then that $`c_{l,M}=\delta ^{(0),0}e_{l,M}`$. This means that $`c\delta ^0e_{l,M}`$ starts at homogeneity $`l+1`$. Going on in this way allows to absorb all of $`c_M`$. Note that if $`c`$ is a polynomial in the undifferentiated fields and $`\delta ^{\mathrm{int},0}=0`$, the procedure stops after a finite number of steps because the terms modifying the terms in $`c_M`$ of homogeneity higher than $`l`$ in the absorption of $`c_{l,M}`$ have a $`K`$ degree which is strictly smaller than $`M`$. One can then go on recursively to remove $`c_{M1},\mathrm{}`$. Because $`\delta ^0`$ is acyclic, we can assume that in the equation $`c_k=\delta e_{k+1}`$, the $`K`$ degrees of $`e_{k+1}`$ is bounded by the same $`M`$ bounding the $`K`$ degree of $`c_K`$. Indeed, if the $`K`$ degree of $`e_{k+1}`$ were bounded by $`N>M`$, we have $`\delta ^0e_{k+1,N}=0`$. Acyclicity of $`\delta ^0`$ then implies $`e_{k+1,N}=\delta ^0f_{k+1,N}`$ and, the redefinition $`e\delta f_{k+1,N}`$, which does not affect $`c_k`$ allows to absorb $`e_{k+1,N}`$.
If the $`K`$ degree of $`a_k`$ is bounded by $`M`$, the $`K`$ degree of $`\frac{\delta a_k}{\delta \varphi ^i}`$, $`\frac{\delta a_k}{\delta \varphi _i^{}}`$, $`\frac{\delta a_k}{\delta C_\alpha ^{}}`$ is bounded respectively by $`M`$, $`Mr_i`$, $`Mm_\alpha `$, because of $`[K,_\mu ]=_\mu `$. It follows from the definitions of $`r_i`$ and $`m_\alpha `$ that the $`K`$ degree of $`\sigma _{ik+1},\sigma _k^i,\sigma _{k1}^\alpha `$ is also bounded respectively by $`M,Mr_i,Mm_\alpha `$ and that the $`K`$ degree of the second line of equation (6.50), modifying the terms of higher homegeneity in the fields in the absorption of the term of order $`l`$ in the proof of theorem 6.5, is also bounded by $`M`$. This proves the theorem, by noticing as before that in the space of polynomials in all the variables, with $`\delta ^{\mathrm{int},0}=0`$, the procedure stops after a finite number of steps.
## 7 Homological perturbation theory
### 7.1 The longitudinal differential $`\gamma `$
In the introduction, we have defined the $`\gamma `$-differential for Yang-Mills gauge models in terms of generators. Contrary to the Koszul-Tate differential, $`\gamma `$ does not depend on the Lagrangian but only on the gauge symmetries. Thus, it takes the same form for all gauge theories of the Yang-Mills type. One has explicitly
$`\gamma `$ $`=`$ $`{\displaystyle \underset{m}{}}\left[_{\mu _1\mathrm{}\mu _m}(D_\mu C^I){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}A_\mu ^I)}}+_{\mu _1\mathrm{}\mu _m}(eC^IT_{Ij}^i\psi ^j){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}\psi ^i)}}\right]`$ (7.1)
$`+{\displaystyle \underset{m}{}}_{\mu _1\mathrm{}\mu _m}(\frac{1}{2}ef_{KJ}^{}{}_{}{}^{I}C^JC^K){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}C^I)}}`$
$`+{\displaystyle \underset{m}{}}\left[_{\mu _1\mathrm{}\mu _m}(ef_{JI}^{}{}_{}{}^{K}C^JA_K^\mu ){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}A_I^\mu )}}+_{\mu _1\mathrm{}\mu _m}(eC^I\psi _j^{}T_{Ii}^j){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}\psi _i^{})}}\right]`$
$`+{\displaystyle \underset{m}{}}_{\mu _1\mathrm{}\mu _m}(ef_{JI}^{}{}_{}{}^{K}C^JC_K^{}){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}C_I^{})}}.`$
Clearly, $`\gamma ^2=0`$. The differential $`\gamma `$ increases the pure ghost number by one unit, $`[N_C,\gamma ]=\gamma `$.
One may consider the restriction $`\gamma _R`$ of $`\gamma `$ to the algebra generated by the original fields and the ghosts, without the antifields,
$`\gamma _R`$ $`=`$ $`{\displaystyle \underset{m}{}}\left[_{\mu _1\mathrm{}\mu _m}(D_\mu C^I){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}A_\mu ^I)}}+_{\mu _1\mathrm{}\mu _m}(eC^IT_{Ij}^i\psi ^j){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}\psi ^i)}}\right]`$ (7.2)
$`+{\displaystyle \underset{m}{}}_{\mu _1\mathrm{}\mu _m}(\frac{1}{2}ef_{KJ}^{}{}_{}{}^{I}C^JC^K){\displaystyle \frac{}{(_{\mu _1\mathrm{}\mu _m}C^I)}}.`$
One has also $`\gamma _R^2=0`$, i.e., the antifields are not necessary for nilpotency of $`\gamma `$. It is sometimes this differential which is called the BRST differential.
However, the fact that this restricted differential – or even (7.1) – is nilpotent is an accident of gauge theories of the Yang-Mills type. For more general gauge theories with so-called “open algebras”, $`\gamma _R`$ (known as the “longitudinal exterior differential along the gauge orbits”) is nilpotent only on-shell, $`\gamma _R^20`$. Accordingly, it is a differential only on the stationary surface. Alternatively, when the antifields are included, $`\gamma `$ fulfills $`\gamma ^2=(\delta s_1+s_1\delta )`$ and is a differential only in the homology of $`\delta `$. Thus, one can define, in general, only the cohomological groups $`H(\gamma ,H(\delta ))`$. \[For Yang-Mills theories, however, $`H(\gamma )`$ makes sense even in the full algebra since $`\gamma `$ is strictly nilpotent on all fields and antifields. The cohomology $`H(\gamma )`$ turns out to be important and will be computed below.\]
In the general case the BRST differential $`s`$ is not simply given by $`s=\delta +\gamma `$, but contains higher order terms
$$s=\delta +\gamma +s_1+\text{ “higher order terms”},$$
(7.3)
where the higher order terms have higher antifield number, and $`s_1`$ and possibly higher order terms are necessary for $`s`$ to be nilpotent, $`s^2=0`$. This can even happen for a “closed algebra”. Indeed, in the case of non constant structure functions, $`\gamma ^2`$ does not necessarily vanish on the antifields and a non vanishing $`s_1`$ may be needed.
The construction of $`s`$ from $`\delta `$ and $`\gamma `$ follows a recursive pattern known as “homological perturbation theory”. We shall not explain it here since this machinery is not needed in the Yang-Mills context where $`s`$ is simply given by $`s=\delta +\gamma `$. However, even though the ideas of homological perturbation theory are not necessary for constructing $`s`$ in the Yang-Mills case, they are crucial in elucidating some aspects of the BRST cohomology and in relating it to the cohomologies $`H(\delta )`$ and $`H(\gamma ,H(\delta ))`$. In particular, they show the importance of the antifield number as auxiliary degree useful to split the BRST differential. They also put into light the importance of the Koszul-Tate differential in the BRST construction<sup>7</sup><sup>7</sup>7In fact, the explicit decomposition $`s=\delta +\gamma `$ appeared in print relatively recently, even though it is of course rather direct.. It is this step that has enabled one, for instance, to solve long-standing conjectures regarding the BRST cohomology.
### 7.2 Decomposition of BRST cohomology
The BRST cohomology groups are entirely determined by cohomology groups involving the first two terms $`\delta `$ and $`\gamma `$ in the decomposition $`s=\delta +\gamma +s_1+\mathrm{}`$ This result is quite general, so we shall state and demonstrate it without sticking to theories of the Yang-Mills type. In fact, it is based solely on the acyclicity of $`\delta `$ in positive antifield number which is crucial for the whole BRST construction,
$$H_k(\delta )=0\text{for}k>0.$$
(7.4)
###### Theorem 7.1
In the space of local forms, one has the following isomorphisms:
$`H^g(s)`$ $``$ $`H_0^g(\gamma ,H(\delta )),`$ (7.5)
$`H^{g,p}(s|d)`$ $``$ $`\{\begin{array}{ccc}H_0^{g,p}(\gamma ,H(\delta |d))& \mathrm{if}& g0,\\ H_g^p(\delta |d)& \mathrm{if}& g<0,\end{array}`$ (7.8)
where the superscripts $`g`$ and $`p`$ indicate the (total) ghost number and the form-degree respectively and the subscript indicates the antifield number.
Explanation and proof. Both isomorphisms are based upon the expansion in the antifield number and state that solutions $`a`$ to $`sa=0`$ or $`sa+dm=0`$ can be fully characterized in the cohomological sense (i.e., up to respective trivial solutions) through properties of the lowest term in their expansion. Let $`g`$ denote the ghost number of $`a`$. When $`g`$ is nonnegative, $`a`$ may contain a piece that does not involve an antifield at all; in contrast, when $`g`$ is negative, the lowest possible term in the expansion of $`a`$ has antifield number $`g`$,
$$a=a_{\underset{¯}{k}}+a_{\underset{¯}{k}+1}+a_{\underset{¯}{k}+2}+\mathrm{},\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}(a_k)=k,\underset{¯}{k}\{\begin{array}{cc}0\mathrm{if}g0,\hfill & \\ g\mathrm{if}g<0,\hfill & \end{array}$$
(7.9)
because there are no fields of negative ghost number.
Now, (7.5) expresses on the one hand that every nontrivial solution $`a`$ to $`sa=0`$ has an antifield independent piece $`a_0`$ which fulfills
$$\gamma a_0+\delta a_1=0,a_0\gamma b_0+\delta b_1$$
(7.10)
and is thus a nontrivial element of $`H_0^g(\gamma ,H(\delta ))`$ since $`\gamma a_0+\delta a_1=0`$ and $`a_0=\gamma b_0+\delta b_1`$ are the cocycle and coboundary condition in that cohomology respectively (a cocycle of $`H_0^g(\gamma ,H(\delta ))`$ is $`\gamma `$-closed up to a $`\delta `$-exact form since $`\delta `$-exact forms vanish in $`H(\delta )`$). Note that (7.10) means $`\gamma a_00`$ and $`a_0\gamma b_0`$. In particular, $`H(s)`$ thus vanishes at all negative ghost numbers because then $`a`$ has no antifield independent piece $`a_0`$. Furthermore (7.5) expresses that each solution $`a_0`$ to (7.10) can be completed to a nontrivial $`s`$-cocycle $`a=a_0+a_1+\mathrm{}`$ and that this correspondence between $`a`$ and $`a_0`$ is unique up to terms which are trivial in $`H^g(s)`$ and $`H_0^g(\gamma ,H(\delta ))`$ respectively.
To prove these statements, we show first that $`sa=0`$ implies $`a=sb`$ whenever $`\underset{¯}{k}>0`$, for some $`b`$ whose expansion starts at antifield number $`\underset{¯}{k}+1`$,
$$sa=0,\underset{¯}{k}>0a=s(b_{\underset{¯}{k}+1}+\mathrm{}).$$
(7.11)
This is seen as follows. $`sa=0`$ contains the equation $`\delta a_{\underset{¯}{k}}=0`$ which implies $`a_{\underset{¯}{k}}=\delta b_{\underset{¯}{k}+1}`$ when $`\underset{¯}{k}>0`$, thanks to (7.4). Consider now $`a^{}:=asb_{\underset{¯}{k}+1}`$. If $`a^{}`$ vanishes we get $`a=sb_{\underset{¯}{k}+1}`$ and thus that $`a`$ is trivial. If $`a^{}`$ does not vanish, its expansion in the antifield number reads $`a^{}=a_{\underset{¯}{k}^{}}^{}+\mathrm{}`$ where $`\underset{¯}{k}^{}>\underset{¯}{k}`$ because of $`a_{\underset{¯}{k}}^{}=a_{\underset{¯}{k}}\delta b_{\underset{¯}{k}+1}=0`$. Furthermore we have $`sa^{}=sas^2b_{\underset{¯}{k}+1}=0`$. Applying the same arguments to $`a^{}`$ as before to $`a`$, we conclude $`a_{\underset{¯}{k}^{}}^{}=\delta b_{\underset{¯}{k}^{}+1}^{}`$ for some $`b_{\underset{¯}{k}^{}+1}^{}`$. We now consider $`a^{\prime \prime }=a^{}sb_{\underset{¯}{k}^{}+1}^{}=as(b_{\underset{¯}{k}+1}+b_{\underset{¯}{k}^{}+1}^{})`$. If $`a^{\prime \prime }`$ vanishes we get $`a=s(b_{\underset{¯}{k}+1}+b_{\underset{¯}{k}^{}+1}^{})`$ and stop. If $`a^{\prime \prime }`$ does not vanish we continue until we finally get $`a=sb`$ for some $`b=b_{\underset{¯}{k}+1}+b_{\underset{¯}{k}^{}+1}^{}+b_{\underset{¯}{k}^{\prime \prime }+1}^{\prime \prime }+\mathrm{}`$ (possibly after infinitely many steps).
(7.11) shows that every $`s`$-cocycle with $`\underset{¯}{k}>0`$ is trivial. When $`\underset{¯}{k}=0`$, $`a_0`$ satisfies automatically $`\delta a_0=0`$ since it contains no antifield. The first nontrivial equation in the expansion of $`sa=0`$ is then $`\gamma a_0+\delta a_1=0`$, while $`a=sb`$ contains $`a_0=\gamma b_0+\delta b_1`$. Hence, every nontrivial $`s`$-cocycle contains indeed a solution to (7.10).
To show that every solution to (7.10) can be completed to a nontrivial $`s`$-cocycle, we consider the cocycle condition in $`H_0^g(\gamma ,H(\delta ))`$, $`\gamma a_0+\delta a_1=0`$, and define $`X:=s(a_0+a_1)`$. When $`X`$ vanishes we have $`sa=0`$ with $`a=a_0+a_1`$ and thus that $`a`$ is an $`s`$-cocycle. When $`X`$ does not vanish, its expansion starts at some antifield number $`1`$ due to $`X_0=\gamma a_0+\delta a_1=0`$. Furthermore we have $`sX=s^2(a_0+a_1)=0`$. Applying (7.11) to $`X`$ yields thus $`X=sY`$ for some $`Y=a_k+\mathrm{}`$ where $`k2`$. Hence we get $`X=s(a_0+a_1)=s(a_k+\mathrm{})`$ and thus $`sa=0`$ where $`a=a_0+a_1+a_k+\mathrm{}`$ with $`k2`$. So, each solution of $`\gamma a_0+\delta a_1=0`$ can indeed be completed to an $`s`$-cocycle $`a`$. Furthermore $`a`$ is trivial if $`a_0=\gamma b_0+\delta b_1`$ and nontrivial otherwise. Indeed, $`a_0=\gamma b_0+\delta b_1`$ implies that $`Z:=as(b_0+b_1)`$ fulfills $`sZ=0`$ and $`Z_0=a_0(\gamma b_0+\delta b_1)=0`$, and thus, by arguments used before, either $`Z=0`$ or $`Z=s(b_k+\mathrm{})`$, $`k2`$ which both give $`a=sb`$. $`a_0\gamma b_0+\delta b_1`$ guarantees $`asb`$ because $`a=sb`$ would imply $`a_0=\gamma b_0+\delta b_1`$. We have thus seen that (non)trivial elements of $`H(s)`$ correspond to (non)trivial elements of $`H_0^g(\gamma ,H(\delta ))`$ and vice versa which establishes (7.5).
(7.8) expresses that every nontrivial solution $`a`$ to $`sa+dm=0`$ has a piece $`a_0`$ if $`g0`$, or $`a_g`$ if $`g<0`$, fulfilling
$`g0:`$ $`\gamma a_0+\delta a_1+dm_0=0,a_0\gamma b_0+\delta b_1+dn_0,`$ (7.12)
$`g<0:`$ $`\delta a_g+dm_{g1}=0,a_g\delta b_{g+1}+dn_g.`$ (7.13)
(7.12) states that $`a_0`$ is a nontrivial cocycle of $`H_0^{g,}(\gamma ,H(\delta |d))`$ because $`\gamma a_0+\delta a_1+dm_0=0`$ and $`a_0=\gamma b_0+\delta b_1+dn_0`$ are the cocycle and coboundary condition in that cohomology respectively. (7.13) states that $`a_g`$ is a nontrivial cocycle of $`H_g^p(\delta |d)`$. Furthermore (7.8) expresses that every solution to (7.12) or (7.13) can be completed to a nontrivial solution $`a=a_0+a_1+\mathrm{}`$ or $`a=a_g+a_{g+1}+\mathrm{}`$ of $`sa+db=0`$. Note that $`a_0`$ contains no antifield, while $`a_g`$ contains no ghost due to $`\mathrm{𝑔ℎ}(a)=g`$. Hence, (7.12) means $`\gamma a_0+dm_00`$ and $`a_0\gamma b_0+dn_0`$, while (7.13) means that $`a_g`$ is related to a nontrivial element of the characteristic cohomology as explained in detail in Section 6.
These statements can be proved along lines whose logic is very similar to the derivation of (7.5) given above. Therefore we shall only sketch the proof, leaving the details to the reader. The derivation is based on corollary 6.1 which itself is a direct consequence of (7.4) as the proof of that theorem shows. The rôle of Eq. (7.11) is now taken by the following result:
$$sa+dm=0,\underset{¯}{k}>\{\begin{array}{cc}0\mathrm{if}g0\hfill & \\ g\mathrm{if}g<0\hfill & \end{array}a=s(b_{\underset{¯}{k}+1}+\mathrm{})+d(n_{\underset{¯}{k}}+\mathrm{}).$$
(7.14)
This is proved as follows. $`sa+dm=0`$ contains the equation $`\delta a_{\underset{¯}{k}}+dm_{\underset{¯}{k}1}=0`$. When $`\underset{¯}{k}>0`$ and $`g0`$, or when $`\underset{¯}{k}>g`$ and $`g<0`$, $`a_{\underset{¯}{k}}`$ has both positive antifield number and positive pureghost number (due to $`\mathrm{𝑔ℎ}=\mathrm{𝑎𝑛𝑡𝑖𝑓𝑑}+\mathrm{𝑝𝑢𝑟𝑒𝑔ℎ}`$). Using theorem 6.3, we then conclude $`a_{\underset{¯}{k}}=\delta b_{\underset{¯}{k}+1}+dn_{\underset{¯}{k}}`$ for some $`b_{\underset{¯}{k}+1}`$ and $`n_{\underset{¯}{k}}`$. One now considers $`a^{}:=asb_{\underset{¯}{k}+1}dn_{\underset{¯}{k}}`$ which fulfills $`sa^{}+dm^{}=0`$ ($`m^{}=msn_{\underset{¯}{k}}`$) and derives (7.14) using recursive arguments analogous to those in the derivation of (7.11). The only values of $`\underset{¯}{k}`$ which are not covered in (7.14) are $`\underset{¯}{k}=0`$ if $`g0`$, and $`\underset{¯}{k}=g`$ if $`g<0`$. In these cases, $`sa+dm=0`$ contains the equation $`\gamma a_0+\delta a_1+dm_0=0`$ if $`g0`$, or $`\delta a_g+dm_{g1}=0`$ if $`g<0`$. These are just the first equations in (7.12) and (7.13) respectively. To finish the proof one finally shows that $`a`$ is trivial ($`a=sb+dn`$) if and only if $`a_0=\gamma b_0+\delta b_1+dn_0`$ for $`g0`$, or $`a_g=\delta b_{g+1}+dn_g`$ for $`g<0`$ by arguments which are again analogous to those used in the derivation of (7.5).
### 7.3 Bounded antifield number
As follows from the proof, the isomorphisms in theorem 7.1 hold under the assumption that the local forms in the theory may contain terms of arbitrarily high antifield number. That is, if one expands the BRST cocycle $`a`$ associated with a given element $`a_0`$ of $`H_0^g(\gamma ,H(\delta ))`$ or $`H_0^{g,p}(\gamma ,H(\delta |d))`$ according to the antifield number as in Eq. (7.9), there is no guarantee, in the general case, that the expansion stops even if $`a_0`$ is a local form. So, although each term in the expansion would be a local form in this case, $`a`$ may contain arbitrarily high derivatives if the number of derivatives in $`a_k`$ grows with $`k`$. This is not a problem for effective field theories, but is in conflict with locality otherwise.
In the case of normal theories with a local Lagrangian, which include, as we have seen, the original Yang-Mills theory, the standard model as well as pure Chern-Simons theory in $`3`$ dimensions (among others), one can easily refine the theorems and show that the expansion (7.9) stops, so that $`a`$ is a local form. This is done by introducing a degree that appropriately controls the antifield number as well as the number of derivatives.
To convey the idea, we illustrate the procedure in the simplest case of pure electromagnetism. We leave it to the reader to extend the argument to the general case. The degree in question – call it $`D`$ – may then be taken to be the sum of the degree counting the number of derivatives plus the degree assigning weight one to the antifields $`A^\mu `$ and $`C^{}`$. Our assumption of locality and polynomiality in the derivatives for $`a_0`$ implies that it has bounded degree $`D`$. In fact, since the differentials $`\delta `$, $`\gamma `$ and $`d`$ all increase this degree by one unit, one can assume that $`a_0`$ is homogeneous of definite (finite) $`D`$-degree $`k`$. The recursive equations in $`sa+dm=0`$ determining $`a_{i+1}`$ from $`a_i`$ read in this case $`\delta a_{i+1}+\gamma a_i+dm_i=0`$ (thanks to $`s=\delta +\gamma `$), and so, one can assume that $`a_{i+1}`$ has also $`D`$-degree $`k`$. Thus, all terms in the expansion (7.9) have same $`D`$-degree equal to $`k`$. This means that as one goes from one term $`a_i`$ to the next $`a_{i+1}`$ in (7.9), the antifield number increases (by definition) while the number of derivatives decreases until one reaches $`a_m=a_{m+1}=\mathrm{}=0`$ after a finite number of (at most $`2k`$) steps.
The fact that the expansion (7.9) stops is particularly convenient because it enables one to analyse the BRST cohomology starting from the last term in (7.9) (which exists). Although this is not always necessary, this turns out to be often a convenient procedure.
### 7.4 Comments
The ideas of homological perturbation theory appeared in the mathematical literature in . They have been applied in the context of the antifield formalism in (with locality analyzed in ) and are reviewed in , chapters 8 and 17.
## 8 Lie algebra cohomology: $`H(s)`$ and $`H(\gamma )`$ in Yang-Mills type theories
### 8.1 Eliminating the derivatives of the ghosts
Our first task in the computation of $`H(s)`$ for gauge theories of the Yang-Mills type is to get rid of the derivatives of the ghosts. This can be achieved for every Lagrangian $`L`$ fulfilling the conditions of the introduction; it is performed by making a change of jet-space coordinates adapted to the problem at hand (see ).
We consider subsets $`W^k`$ ($`k=1,0,1,\mathrm{}`$) of jet coordinates where $`W^1`$ contains only the undifferentiated ghosts $`C^I`$, while $`W^k`$ for $`k0`$ contains
$`A_{\mu |(\nu _1\mathrm{}\nu _l)}^I,\psi _{(\nu _1\mathrm{}\nu _l)}^i,C_{(\nu _1\mathrm{}\nu _{l+1})}^I,`$
$`A_{I(\nu _1\mathrm{}\nu _{l2})}^\mu ,\psi _{i(\nu _1\mathrm{}\nu _{lm})}^{},C_{I(\nu _1\mathrm{}\nu _{l3})}^{}`$ (8.1)
for $`l=0,\mathrm{},k`$ and $`m=1,2`$, for matter fields with first and second order field equations respectively. \[These definitions are in fact taylored to the standard model; if the gauge fields obey equations of motion of order $`k`$, $`\nu _{l2}`$ should be replaced by $`\nu _{lk}`$ and $`\nu _{l3}`$ by $`\nu _{lk1}`$ in (8.1); similarly, $`m`$ is generally the derivative order of the matter field equations.\]
We can take as new coordinates on $`W^k`$ the following functions of the old ones:
$`_{(\nu _1\mathrm{}\nu _l}A_{\mu )}^I,_{(\nu _1\mathrm{}\nu _l}D_{\mu )}C^I,`$ (8.2)
$`C^I,`$ (8.3)
$`D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I,D_{(\nu _1}\mathrm{}D_{\nu _l)}\psi ^i,`$ (8.4)
$`D_{(\nu _1}\mathrm{}D_{\nu _{l2})}A_I^\mu ,D_{(\nu _1}\mathrm{}D_{\nu _{l3})}C_I^{},D_{(\nu _1}\mathrm{}D_{\nu _{lm})}\psi _i^{},`$ (8.5)
for $`l=0,\mathrm{},k`$ and $`m=1,2`$. This change of coordinates is invertible because $`_{\nu _1\mathrm{}\nu _l}A_\mu ^I=_{(\nu _1\mathrm{}\nu _l}A_{\mu )}^I+\frac{l}{l+1}D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I+O(l1)`$ where $`O(l1)`$ collects terms with less than $`l`$ derivatives. There are no algebraic relations between the $`_{(\nu _1\mathrm{}\nu _l}A_{\mu )}^a`$ and the $`D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^a`$, which correspond to the independent irreducible components of $`_{\nu _1\mathrm{}\nu _l}A_\mu ^I`$. Similarly, one has $`D_{(\nu _1}\mathrm{}D_{\nu _l)}\psi ^i=_{\nu _1\mathrm{}\nu _l}\psi ^i+O(l1)`$.
The new coordinates can be grouped into two sets : contractible pairs (8.2) on the one hand, and gauge covariant coordinates (8.4), (8.5) plus undifferentiated ghosts (8.3) on the other hand. These sets transform among themselves under $`s`$. Indeed we have $`s_{(\nu _1\mathrm{}\nu _l}A_{\mu )}^I=_{(\nu _1\mathrm{}\nu _l}D_{\mu )}C^I`$ and $`s_{(\nu _1\mathrm{}\nu _l}D_{\mu )}C^I=0`$. Similarly, if we collectively denote by $`\chi _\mathrm{\Delta }^u`$ the coordinates (8.4) and (8.5), we have $`\gamma \chi _\mathrm{\Delta }^u=eC^IT_{Iv}^u\chi _\mathrm{\Delta }^v`$ where the $`T_{Iv}^u`$ are the entries of representation matrices of $`𝒢`$ and $`\mathrm{\Delta }`$ labels the various multiplets of $`𝒢`$ formed by the $`\chi `$’s. For instance, for every fixed set of spacetime indices, the $`D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^K`$ ($`K=1,\mathrm{},\mathrm{𝑑𝑖𝑚}(𝒢)`$) form a multiplet $`\chi _\mathrm{\Delta }^K`$ (with fixed $`\mathrm{\Delta }`$) of the coadjoint representation with $`T_{Iv}^u`$ given by $`f_{IJ}^{}{}_{}{}^{K}`$.<sup>8</sup><sup>8</sup>8 Our notation is slightly sloppy because the index $`u`$ (and in particular its range) really depends on the given multiplet and thus should carry a subindex $`\mathrm{\Delta }`$. The substitution $`uu_\mathrm{\Delta }`$ should thus be understood in the formulas below.
Finally, $`sC^I`$ is a function of the ghosts alone, and $`\delta `$ on the antifields (8.5) only involves the coordinates (8.5) and (8.4). The latter statement about the $`\delta `$-transformations is equivalent to the gauge covariance of the equations of motion. It can be inferred from $`\gamma \delta +\delta \gamma =0`$, without referring to a particular Lagrangian. Namely $`(\gamma \delta +\delta \gamma )A_I^\mu =0`$ gives $`\gamma L_I^\mu =eC^Jf_{JI}^{}{}_{}{}^{K}L_K^\mu `$ while $`(\gamma \delta +\delta \gamma )\psi _i^{}=0`$ gives $`\gamma L_i=eC^IT_{Ii}^jL_j`$, see Eq. (2.8). The absence of derivatives of the ghosts in $`\gamma L_I^\mu `$ and $`\gamma L_i`$ implies that $`L_I^\mu `$ and $`L_i`$ can be expressed solely in terms of the variables (8.4) (and the $`x^\mu `$ when the Lagrangian involves $`x^\mu `$ explicitly) because $`\gamma `$ is stable in the subspaces of local functions with definite degree in the variables (8.2).
The coordinates (8.2) form thus indeed contractible pairs and do not contribute to the cohomology of $`s`$ according to a reasoning analogous to the one followed in section 2.7.
Note that the removal of the vector potential, its symmetrized derivatives, and the derivatives of the ghosts, works both for $`H(s)`$ and $`H(\gamma )`$ since $`s`$ and $`\gamma `$ coincide in this sector.
### 8.2 Lie algebra cohomology with coefficients in a representation
One of the interests of the elimination of the derivatives of the ghosts is that the connection between $`H(\gamma )`$, $`H(s)`$ and ordinary Lie algebra cohomology becomes now rather direct.
We start with $`H(\gamma )`$, for which matters are straightforward. We have reduced the computation of the cohomology of $`\gamma `$ in the algebra of all local forms to the calculation of the cohomology of $`\gamma `$ in the algebra $`𝒦`$ of local forms depending on the covariant objects $`\chi _\mathrm{\Delta }^u`$ and the undifferentiated ghosts $`C^I`$. More precisely, the relevant algebra is now
$$𝒦=\mathrm{\Omega }(^n)\mathrm{\Lambda }(C)$$
(8.6)
where $`\mathrm{\Omega }(^n)`$ is the algebra of exterior forms on $`^n`$, $``$ the algebra of functions of the covariant objects $`\chi _\mathrm{\Delta }^u`$, and $`\mathrm{\Lambda }(C)`$ the algebra of polynomials in the ghosts $`C^I`$ (which is just the antisymmetric algebra with $`\mathrm{𝑑𝑖𝑚}(𝒢)`$ generators).
The subalgebra $`\mathrm{\Omega }(^n)`$ provides a representation of the Lie algebra $`𝒢`$, the factor $`\mathrm{\Omega }(^n)`$ being trivial since it does not transform under $`𝒢`$. We call this representation $`\rho `$. The differential $`\gamma `$ can be written as
$$\gamma =eC^I\rho (e_I)+\frac{e}{2}C^JC^If_{IJ}^{}{}_{}{}^{K}\frac{}{C^K}$$
(8.7)
where the $`e_I`$ form a basis for the Lie algebra $`𝒢`$ and the $`\rho (e_I)`$ are the corresponding “infinitesimal generators” in the representation,
$$\rho (e_I)=T_{Iv}^u\chi _\mathrm{\Delta }^v\frac{}{\chi _\mathrm{\Delta }^u}.$$
(8.8)
The identification of the polynomials in $`C^I`$ with the cochains on $`𝒢`$ then allows to identify the differential $`\gamma `$ with the standard Chevalley-Eilenberg differential for Lie algebra cochains with values in the representation space $`\mathrm{\Omega }(^n)C^{\mathrm{}}(\chi _\mathrm{\Delta }^u)`$.
Thus, we see that the cohomology of $`\gamma `$ is just standard Lie algebra cohomology with coefficients in the representation $`\rho `$ of functions in the covariant objects $`\chi _\mathrm{\Delta }^u`$ (times the spacetime exterior forms).
The space of smooth functions in the variables $`\chi _\mathrm{\Delta }^u`$ is evidently infinite-dimensional. In order to be able to apply theorems on Lie algebra cohomology, it is necessary to make some restrictions on the allowed functions so as to effectively deal with finite dimensional representations. This condition will be met, for instance, if one considers polynomial local functions in the $`\chi _\mathrm{\Delta }^u`$ with coefficients that can possibly be smooth functions of invariants (e.g. $`\mathrm{exp}\varphi `$ can occur if $`\varphi `$ does not transform under $`𝒢`$). This space is still infinite-dimensional, but splits as the direct sum of finite-dimensional representation spaces of $`𝒢`$. Indeed, because $`\rho (e_I)`$ is homogeneous of degree 0 in the $`\chi _\mathrm{\Delta }^u`$, we can consider separately polynomials of a given homogeneity in the $`\chi _\mathrm{\Delta }^u`$, which form finite dimensional representation spaces. Thus, the problem of computing the Lie algebra cohomology of $`𝒢`$ with coefficients in the representation $`\rho `$ is effectively reduced to the problem of computing the Lie algebra cohomology of $`𝒢`$ with coefficients in a finite-dimensional representation. The same argument applies, of course, to effective field theories. From now on, it will be understood that such restrictions are made on $``$.
### 8.3 $`H(s)`$ versus $`H(\gamma )`$
The previous section shows that the computation of $`H(\gamma )`$ boils down to a standard problem of Lie algebra cohomology with coefficients in a definite representation. This is also true for $`H(s)`$, but the representation space is now different.
Indeed, we have seen that $`H(s)H(\gamma ,H(\delta ))`$. This result was established in the algebra of all local forms depending also on the differentiated ghosts and symmetrized derivatives of the vector potential, but also holds in the algebra $`𝒦`$. Moreover, the cohomology of the Koszul-Tate differential in $`𝒦`$ can be computed in the same manner as above. The antifields drop out with the variables constrained to vanish with the equations of motion. More precisely, among the field strength components, the matter field components, and their covariant derivatives, some can be viewed as constrained by the equations of motion and the others can be viewed as independent. Let $`X_A^u`$ be the independent ones and $`^R`$ be the algebra of smooth functions in $`X_A^u`$ (with restrictions analogous to those made in the previous subsection). Because the equations of motion are gauge covariant (see above), one can take the $`X_A^u`$ to transform in a linear representation of $`𝒢`$, which we denote by $`\rho ^R`$. Again, since one can work order by order in the derivatives, the representation $`\rho ^R`$ effectively splits as a direct sum of finite-dimensional representations. \[See below for an explicit construction of the $`X_A^u`$ in the standard model.\]
By our general discussion of section 7, it follows that
###### Lemma 8.1
The cohomology of $`s`$ is isomorphic to the cohomology of $`\gamma `$ in the space of local forms depending only on the undifferentiated ghosts $`C^I`$ and the $`X_A^u`$.
In the case of the standard model, the $`X_A^u`$ may be constructed as follows. First, the $`D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I`$ are split into the algebraically independent completely traceless combinations $`(D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I)_{\mathrm{tracefree}}`$ (i.e., $`\eta ^{\nu _1\nu _2}(D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I)_{\mathrm{tracefree}}=0`$) and the traces $`D_{(\nu _1}\mathrm{}D_{\nu _{l2}}D_{\lambda )}F_\mu ^{I\lambda }`$ . Second the covariant derivatives $`D_{(s_1}\mathrm{}D_{s_l)}\psi ^i`$ of the matter fields are replaced by $`D_{(s_1}\mathrm{}D_{s_l)}\widehat{\psi }^i`$ and $`D_{(\nu _1}\mathrm{}D_{\nu _{l1})}\widehat{}_i`$ for matter fields $`\widehat{\psi }^i`$ with first order equations and by $`D_{(s_1}\mathrm{}D_{s_l)}\stackrel{~}{\psi }^i`$, $`D_{(s_1}\mathrm{}D_{s_{l1}}D_{0)}\stackrel{~}{\psi }^i`$, $`D_{(\nu _1}\mathrm{}D_{\nu _{l2})}\stackrel{~}{}_i`$ for matter fields $`\stackrel{~}{\psi }^i`$ with second order field equations. The $`X_A^u`$ are then the coordinates $`(D_{(\nu _1}\mathrm{}D_{\nu _{l1}}F_{\nu _l)\mu }^I)_{\mathrm{tracefree}}`$, $`D_{(s_1}\mathrm{}D_{s_l)}\widehat{\psi }^i`$, $`D_{(s_1}\mathrm{}D_{s_l)}\stackrel{~}{\psi }^i`$, $`D_{(s_1}\mathrm{}D_{s_{l1}}D_{0)}\stackrel{~}{\psi }^i`$. Other splits are of course available.
In the algebra
$$𝒦^R=\mathrm{\Omega }(^n)^R\mathrm{\Lambda }(C)$$
(8.9)
the differential $`\gamma `$ reads
$$\gamma =eC^I\rho ^R(e_I)+\frac{e}{2}C^JC^If_{IJ}^{}{}_{}{}^{K}\frac{}{C^K}$$
(8.10)
where the $`\rho ^R(e_I)`$ are the infinitesimal generators in the representation $`\rho ^R`$,
$$\rho ^R(e_I)=T_{Iv}^uX_A^v\frac{}{X_A^u}.$$
(8.11)
The representations of $`𝒢`$ which occur in $``$ and $`^R`$ are the same ones, but with a smaller multiplicity in $`^R`$. One can thus identify the cohomology of $`s`$ with the Lie-algebra cohomology of $`𝒢`$, with coefficients in the representation $`\rho ^R`$. The difference between $`H(s)`$ and $`H(\gamma )`$ lies only in the space of coefficients.
### 8.4 Whitehead’s theorem
In order to proceed, we shall now assume that the gauge group $`G`$ is the direct product of a compact abelian group $`G_0`$ times a semi-simple group $`G_1`$. Thus, $`G=G_0\times G_1`$, with $`G_0=(U(1))^q`$. This assumption on $`G`$ was not necessary for the previous analysis, but is used for the subsequent developments since in this case one has complete results on the Lie algebra cohomology. On these conditions, it can then be shown that any finite-dimensional representation of $`𝒢`$ is completely reducible .
We can now follow the standard literature (e.g. ). For any representation space $`V`$ with representation $`\rho `$, let $`V_{\rho =0}`$ be the invariant subspace of $`V`$ carrying the trivial representation, which may occur several times ($`vV_{\rho =0}\rho (e_I)v=0I`$). Note that the space $`\mathrm{\Lambda }(C)`$ of ghost polynomials is a representation of $`𝒢`$ for
$$\rho ^C(e_I)=C^Jf_{IJ}^{}{}_{}{}^{K}\frac{}{C^K}.$$
(With the above interpretation of the $`C^I`$, it is the extension of the coadjoint representation to $`\mathrm{\Lambda }𝒢^{}`$.) The total representation on $`V\mathrm{\Lambda }(C)`$ is $`\rho ^T(e_I)=\rho (e_I)+\rho ^C(e_I)`$. It satisfies
$$\{\gamma ,\frac{}{C^I}\}=e\rho ^T(e_I),[\gamma ,\rho ^T(e_I)]=0,$$
(8.12)
the first relation following by direct computation, the second one from the first and $`\gamma ^2=0`$. Hence the cohomology $`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})`$ is well defined. We can write
$$\gamma =eC^I\rho (e_I)+\frac{e}{2}C^I\rho ^C(e_I)=eC^I\rho ^T(e_I)+\widehat{\gamma },$$
where $`\widehat{\gamma }`$ is the restriction of $`\gamma `$ to $`\mathrm{\Lambda }(C)`$, up to the sign:
$$\widehat{\gamma }=\frac{e}{2}C^IC^Jf_{IJ}^{}{}_{}{}^{K}\frac{}{C^K}.$$
It follows that
$`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})H(\widehat{\gamma },(V\mathrm{\Lambda }(C))_{\rho ^T=0}).`$ (8.13)
Note that we also have,
$$\{\widehat{\gamma },\frac{}{C^I}\}=e\rho ^C(e_I),[\widehat{\gamma },\rho ^C(e_I)]=0.$$
(8.14)
The first mathematical result that we shall need reduces the problem of computing the Lie algebra cohomology of $`𝒢`$ with coefficients in $`V`$ to that of finding the invariant subspace $`V_{\rho =0}`$ and computing the Lie algebra cohomology of $`𝒢`$ with coefficients in the trivial, one-dimensional, representation.
###### Theorem 8.1
(i) $`H(\gamma ,V\mathrm{\Lambda }(C))`$ is isomorphic to $`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})`$. In particular, $`H(\widehat{\gamma },\mathrm{\Lambda }(C))`$ is isomorphic to $`\mathrm{\Lambda }(C)_{\rho ^C=0}`$.
(ii) $`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})`$ is isomorphic to $`V_{\rho =0}\mathrm{\Lambda }(C)_{\rho ^C=0}`$.
The proof of this theorem is given in the appendix 8.A. The result $`H(\gamma ,(V\mathrm{\Lambda }(C))V_{\rho =0}H(\widehat{\gamma },\mathrm{\Lambda }(C))`$ is known as Whitehead’s theorem.
To determine the cohomology of $`s`$, we thus need to determine on the one hand the invariant monomials in the $`X_A^u`$, which depends on the precise form of the matter field representations and which will not be discussed here; and on the other hand, $`H(\widehat{\gamma },\mathrm{\Lambda }(C))\mathrm{\Lambda }(C)_{\rho ^C=0}`$. This latter cohomology is known as the Lie algebra cohomology of $`𝒢`$ and is discussed in the next section.
### 8.5 Lie algebra cohomology - Primitive elements
We shall only give the results (in ghost notations), without proof. We refer to the mathematical literature for the details .
The cohomology $`H^g(\widehat{\gamma },\mathrm{\Lambda }(C))`$ can be described in terms of particular ghost polynomials $`\theta _r(C)`$ representing the so-called primitive elements. These are in bijective correspondence with the independent Casimir operators $`𝒪_r`$,
$$𝒪_r=d^{I_1\mathrm{}I_{m(r)}}\delta _{I_1}\mathrm{}\delta _{I_{m(r)}},r=1,\mathrm{},rank(𝒢).$$
(8.15)
The $`d^{I_1\mathrm{}I_{m(r)}}`$ are symmetric invariant tensors,
$`{\displaystyle \underset{i=1}{\overset{m(r)}{}}}f_{JL}^{}{}_{}{}^{K}d^{I_1\mathrm{}I_{i1}LI_{i+1}\mathrm{}I_{m(r)}}=0,`$ (8.16)
while $`\delta _I=\rho (e_I)`$ for some representation $`\rho (e_I)`$ of $`𝒢`$. The ghost polynomial $`\theta _r(C)`$ corresponding to $`𝒪_r`$ is homogeneous of degree $`2m(r)1`$, and given by
$`\theta _r(C)=(e)^{m(r)1}{\displaystyle \frac{m(r)!(m(r)1)!}{2^{m(r)1}(2m(r)1)!}}f_{I_1\mathrm{}I_{2m(r)1}}C^{I_1}\mathrm{}C^{I_{2m(r)1}},`$
$`f_{I_1J_1\mathrm{}I_{m(r)1}J_{m(r)1}K_{m(r)}}=d_{K_1\mathrm{}K_{m(r)1}[K_{m(r)}}f_{I_1J_1}^{}{}_{}{}^{K_1}\mathrm{}f_{I_{m(r)1}J_{m(r)1}]}^{}{}_{}{}^{K_{m(r)1}}.`$ (8.17)
This definition of $`\theta _r(C)`$ involves, for later purpose, a normalization factor containing both the gauge coupling constant and the order $`m(r)`$ of $`𝒪_r`$. The $`d_{K_1\mathrm{}K_{m(r)}}`$ arise from the invariant symmetric tensors in (8.15) by lowering the indices with the invertible metric $`g_{IJ}`$ obtained by adding the Killing metrics for each simple factor, trivially extended to the whole of $`𝒢`$, and with the identity for an abelian $`\delta _I`$. Using an appropriate (possibly complex) matrix representation $`\{T_I\}`$ of $`𝒢`$ (cf. example below), $`\theta _r(C)`$ can also be written as
$$\theta _r(C)=(e)^{m(r)1}\frac{m(r)!(m(r)1)!}{(2m(r)1)!}\mathrm{Tr}(C^{2m(r)1}),C=C^IT_I.$$
(8.18)
Indeed, using that the $`C^I`$ anticommute and that $`\{T_I\}`$ represents $`𝒢`$ ($`[T_I,T_J]=f_{IJ}^{}{}_{}{}^{K}T_K`$), one easily verifies that (8.18) agrees with (8.17) for
$$d_{I_1\mathrm{}I_{m(r)}}=\mathrm{Tr}[T_{(I_1}\mathrm{}T_{I_{m(r)})}].$$
Those $`\theta _r(C)`$ with degree 1 coincide with the abelian ghost fields (if any), in accordance with the above definitions: the abelian elements of $`𝒢`$ count among the Casimir operators as they commute with all the other elements of $`𝒢`$. We thus set
$$\{\theta _r(C):m(r)=1\}=\{\text{abelian ghosts}\}.$$
(8.19)
Note that this is consistent with (8.18), as it corresponds to the choice $`\{T_I\}=\{0,\mathrm{},0,1,0,\mathrm{},0\}`$, where one of the abelian elements of $`𝒢`$ is represented by the number 1, while all the other elements of $`𝒢`$ are represented by 0.
Each $`\theta _r(C)`$ is $`\widehat{\gamma }`$ closed, as is easily verified using the matrix notation (8.18),
$$\widehat{\gamma }\mathrm{Tr}(C^{2m1})=e\mathrm{Tr}(C^{2m})=0,$$
where the first equality holds due to $`\widehat{\gamma }C=eC^2`$ and the second equality holds because the trace of any even power of a Grassmann odd matrix vanishes. The cohomology of $`\widehat{\gamma }`$ is generated precisely by the $`\theta _r(C)`$, i.e., the corresponding cohomology classes are represented by polynomials in the $`\theta _r(C)`$, and no nonvanishing polynomial of the $`\theta _r(C)`$ is cohomologically trivial,
$`\widehat{\gamma }h(C)=0h(C)=P(\theta (C))+\widehat{\gamma }g(C);`$
$`P(\theta (C))=\widehat{\gamma }g(C)P=0.`$ (8.20)
Note that the $`\theta _r(C)`$ anticommute because they are homogeneous polynomials of odd degree in the ghost fields. Therefore the dimension of the cohomology of $`\widehat{\gamma }`$ is $`2^{rank(𝒢)}`$. Note also that the highest nontrivial cohomology class (i.e., the one with highest ghost number) is represented by the product of all the $`\theta _r(C)`$. This product is always proportional to the product of all the ghost fields (see, e.g., ), and has thus ghost number equal to the dimension of $`𝒢`$,
$$\underset{r=1}{\overset{rank(𝒢)}{}}\theta _r(C)\underset{I=1}{\overset{dim(𝒢)}{}}C^I.$$
(8.21)
##### Example 1.
Let us spell out the result for the gauge group of the standard model, $`G=U(1)\times SU(2)\times SU(3)`$. The $`U(1)`$-ghost is a $`\theta `$ by itself, see (8.19). We set
$$\theta _1(C)=\text{U(1)-ghost}.$$
$`SU(2)`$ has only one Casimir operator which has order 2. The corresponding $`\theta `$ has thus degree 3 and is given by
$$\theta _2(C)=\frac{e_{\mathrm{su}(2)}}{3}\mathrm{Tr}_{\mathrm{su}(2)}(C^3),$$
with $`C=C^IT_I`$, $`\{T_I\}=\{0,\text{i}\sigma _\alpha ,0,\mathrm{},0\}`$ where the zeros represent $`u(1)`$ and $`su(3)`$, and $`\sigma _\alpha `$ are the Pauli matrices.
$`SU(3)`$ has two independent Casimir operators, with degree 2 and 3 respectively. This gives two additional $`\theta `$’s of degree 3 and 5 respectively,
$$\theta _3(C)=\frac{e_{\mathrm{su}(3)}}{3}\mathrm{Tr}_{\mathrm{su}(3)}(C^3),\theta _4(C)=\frac{e_{\mathrm{su}(3)}^2}{10}\mathrm{Tr}_{\mathrm{su}(3)}(C^5),$$
with $`\{T_I\}=\{0,0,0,0,\text{i}\lambda _a\}`$ where $`\lambda _a`$ are the Gell-Mann matrices.
A complete list of inequivalent representatives of $`H(\widehat{\gamma },\mathrm{\Lambda }(C))`$ is:
$$\begin{array}{cc}\text{ghost number}& \text{representatives of }H(\widehat{\gamma },\mathrm{\Lambda }(C))\\ & \\ \text{}0& 1\\ & \\ \text{}1& \theta _1\\ & \\ \text{}3& \theta _2,\theta _3\\ & \\ \text{}4& \theta _1\theta _2,\theta _1\theta _3\\ & \\ \text{}5& \theta _4\\ & \\ \text{}6& \theta _2\theta _3,\theta _1\theta _4\\ & \\ \text{}7& \theta _1\theta _2\theta _3\\ & \\ \text{}8& \theta _2\theta _4,\theta _3\theta _4\\ & \\ \text{}9& \theta _1\theta _2\theta _4,\theta _1\theta _3\theta _4\\ & \\ \text{}11& \theta _2\theta _3\theta _4\\ & \\ \text{}12& \theta _1\theta _2\theta _3\theta _4\end{array}$$
##### Example 2.
Let us denote by $`C^{I_a}`$ the ghosts of the abelian factors, $`I_a=1,\mathrm{},l`$. A basis of the first cohomology groups $`H^g(\widehat{\gamma },\mathrm{\Lambda }(C))`$ ($`g=0,1,2`$) is given by (i) $`1`$ for $`g=0`$; (ii) $`C^{I_a}`$, with $`I_a=1,\mathrm{},l`$, for $`g=1`$; and (iii) $`C^{I_a}C^{J_a}`$, with $`I_a<J_a`$, for $`g=2`$. In particular, $`H^1(\widehat{\gamma },\mathrm{\Lambda }(C))`$ and $`H^2(\widehat{\gamma },\mathrm{\Lambda }(C))`$ are trivial if there is no abelian factor. For a compact group, a basis for $`H^3(\widehat{\gamma },\mathrm{\Lambda }(C))`$ is given by $`C^{I_a}C^{J_a}C^{K_a}`$ ($`I_a<J_a<K_a`$) and $`f_{I_sJ_sK_s}C^{I_s}C^{J_s}C^{K_s}\mathrm{Tr}_{𝒢_\mathrm{s}}C^3`$, where $`s`$ runs over the simple factors $`G_s`$ in $`G`$.
### 8.6 Implications for the renormalization of local gauge invariant operators
Class I local operators are defined as local, non integrated, gauge invariant operators (built out of the gauge potentials and the matter fields) that are linearly independent, even when the gauge covariant equations of motions are used . In the absence of anomalies, it can be shown that these operators only mix, under renormalization, with BRST closed local operators (built out of all the fields and antifields). BRST exact operators are called class II operators. They can be shown not to contribute to the physical S matrix and to renormalize only among themselves. The question is then whether class I operators can only mix with class I operators and class II operators under renormalization.
That the answer is affirmative follows from lemma 8.1 in the case of ghost number $`0`$. Indeed, the $`\gamma `$ cohomology in the space of forms in the $`X_A^u`$, (i.e., combinations of the covariant derivatives of the field strength components and the matter field components not constrained by the equations of motions) reduces to the gauge invariants in these variables. This is precisely the required statement that class I operators give a basis of the BRST cohomology $`H^{0,}(s,\mathrm{\Omega })`$ in ghost number $`0`$.
The statement was first proved in . A different proof has been given in .
### 8.7 Appendix 8.A: Proof of theorem 8.1
Our proof of theorem 8.1 uses ghost notations.
(i) Let us first of all prove general relations for a completely reducible representation commuting with a differential, i.e., take into account only the second relation of (8.12). This relation implies that the representation $`(\rho ^T)^\mathrm{\#}(e_I)`$ induced in cohomology by $`(\rho ^T)^\mathrm{\#}(e_I)[a]=[\rho ^T(e_I)a]`$ is well defined. The induced representation is completely reducible: since the space of $`\gamma `$ cocycles $`Z`$ is stable under $`\rho ^T(e_I)`$, there exists a stable subspace $`EV\mathrm{\Lambda }(C)`$, such that $`V\mathrm{\Lambda }(C)=ZE`$. Similarly, because the space of $`\gamma `$ coboundaries $`B`$ is stable under $`\rho ^T(e_I)`$, there exists a stable subspace $`FZ`$ such that $`Z=FB`$. It follows that $`H(\gamma ,V\mathrm{\Lambda }(C))`$ is isomorphic to $`F`$. Since $`F`$ is completely reducible for $`\rho ^T(e_I)`$, so is $`H(\gamma ,V\mathrm{\Lambda }(C))`$ for $`(\rho ^T)^\mathrm{\#}(e_I)`$. Complete reducibility also implies that $`Z(\rho ^T(V\mathrm{\Lambda }(C)))=Z\rho ^T(V\mathrm{\Lambda }(C))=\rho ^TZ`$. This means that $`H(\gamma ,\rho ^T(V\mathrm{\Lambda }(C)))(\rho ^T)^\mathrm{\#}H(\gamma ,V\mathrm{\Lambda }(C))`$. In the same way, $`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})H(\gamma ,V\mathrm{\Lambda }(C))_{(\rho ^T)^\mathrm{\#}=0}.`$ Complete reducibility of $`\rho ^T`$ then implies $`H(\gamma ,V\mathrm{\Lambda }(C))=H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})H(\gamma ,\rho ^T(V\mathrm{\Lambda }(C)))`$, while complete reducibility of $`(\rho ^T)^\mathrm{\#}`$ implies $`H(\gamma ,V\mathrm{\Lambda }(C))=H(\gamma ,V\mathrm{\Lambda }(C))_{(\rho ^T)^\mathrm{\#}=0}(\rho ^T)^\mathrm{\#}H(\gamma ,V\mathrm{\Lambda }(C))`$. It follows that $`H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^T=0})=H(\gamma ,V\mathrm{\Lambda }(C))_{(\rho ^T)^\mathrm{\#}=0}`$ and $`H(\gamma ,\rho ^T(V\mathrm{\Lambda }(C)))=(\rho ^T)^\mathrm{\#}H(\gamma ,V\mathrm{\Lambda }(C))`$. Using now the first relation of (8.12), it follows that $`(\rho ^T)^\mathrm{\#}=0`$ so that $`H(\gamma ,\rho ^T(V\mathrm{\Lambda }(C)))=0`$ and $`H(\gamma ,V\mathrm{\Lambda }(C))=H(\gamma ,(V\mathrm{\Lambda }(C))_{\rho ^C=0})`$. This proves the first part of (i).
For the second part, we note that in $`\mathrm{\Lambda }(C)`$, the representation $`\rho ^C(e_I)`$ reduces to the representation of the semi-simple factor. Let us denote the generators of this factor by $`e_a`$ and its representation on $`\mathrm{\Lambda }(C)`$ by $`\rho ^C(e_a)`$. This representation is completely reducible, because defining properties of semi-simple Lia algebras are (a) the Killing metric $`g_{ab}=f_{ac}^{}{}_{}{}^{d}f_{bd}^{}{}_{}{}^{c}`$ is invertible; (b) it is the direct sum of simple ideals, (a Lie algebra being simple if it is non abelian and contains no proper non trivial ideals); (c) all its representations in finite dimensional vector spaces are completely reducible . This proves then the second part by the same reasoning as above with $`V=\rho =0`$, respectively (8.14) instead of (8.12).
(ii) $`\mathrm{\Lambda }(C)=\mathrm{\Lambda }(C)_{\rho ^C=0}\rho ^C\mathrm{\Lambda }(C)`$ implies $`(V\mathrm{\Lambda }(C))_{\rho ^T=0}=V_{\rho =0}\mathrm{\Lambda }(C)_{\rho ^C=0}(V\rho ^C\mathrm{\Lambda }(C))_{\rho ^T=0}`$. Because it follows from $`[\widehat{\gamma },\rho ^T]=0=[\widehat{\gamma },\rho ^C]=[\widehat{\gamma },\rho ]`$ that all the spaces are stable under $`\widehat{\gamma }`$, the Künneth formula gives $`H(\widehat{\gamma },(V\mathrm{\Lambda }(C))_{\rho ^T=0})=V_{\rho =0}H(\widehat{\gamma },\mathrm{\Lambda }(C)_{\rho ^C=0})H(\widehat{\gamma },(V\rho ^C\mathrm{\Lambda }(C))_{\rho ^T=0})`$. The result (ii) then follows from (8.13) and $`\widehat{\gamma }=\frac{e}{2}C^I\rho ^C(e_I)`$, if we can show that $`H(\widehat{\gamma },(V\rho ^C\mathrm{\Lambda }(C))_{\rho ^T=0})=0`$.
The contracting homotopy that allows to prove $`H(\widehat{\gamma },\rho ^C\mathrm{\Lambda }(C))=0`$ can be constructed explicitely as follows. Define the Casimir operator $`\mathrm{\Gamma }=\frac{1}{2}g^{ab}\rho ^C(e_a)\rho ^C(e_b)`$, where $`g^{ab}`$ is the inverse of the Killing metric associated to the semi-simple Lie subalgebra of $`𝒢`$. From the complete skew symmetry of the structure constants lowered or raised through the Killing metric or its inverse (this being a consequence of the Jacobi identity), it follows that this operator commutes with all the operators of the representation, $`[\mathrm{\Gamma },\rho ^C(e_a)]=0`$, while the first relation of (8.14) implies that $`[\widehat{\gamma },\mathrm{\Gamma }]=0`$.
A property of semi-simple Lie algebras is that the Casimir operator $`\mathrm{\Gamma }`$ is invertible on $`\rho ^C\mathrm{\Lambda }(C)`$. Obviously, in this case, $`[\mathrm{\Gamma }^1,\rho ^C(e_a)]=0`$ and $`[\widehat{\gamma },\mathrm{\Gamma }^1]=0`$. Take $`a\rho ^C\mathrm{\Lambda }(C)`$, with $`\widehat{\gamma }a=0`$. We have $`a=\mathrm{\Gamma }\mathrm{\Gamma }^1a=\frac{1}{2}g^{ab}\{\widehat{\gamma },\frac{}{C^a}\}\rho ^C(e_b)\mathrm{\Gamma }^1a=\widehat{\gamma }\rho ^C(e_b)\frac{1}{2}g^{ab}\frac{}{C^a}\mathrm{\Gamma }^1a`$, where we have used in addition $`g^{ab}[\frac{}{C^a},\rho ^C(e_b)]=0`$, as follows from the first relation of (8.14) and the graded Jacobi identity for the graded commutator of operators. Hence, $`H(\widehat{\gamma },\rho ^C\mathrm{\Lambda }(C))=0`$.
Similarly, let $`a=vb`$, where $`b\rho ^C\mathrm{\Lambda }(C)`$, with $`\gamma a=0=eC^I\rho (e_I)vbe()^vv\widehat{\gamma }b`$ and $`\rho ^T(e_I)a=0=\rho (e_I)vb+v\rho (e_I)b`$. We have $`a=v(\widehat{\gamma }\rho ^C(e_b)\frac{1}{2}g^{ab}\frac{}{C^a}\mathrm{\Gamma }^1b+\rho ^C(e_b)\frac{1}{2}g^{ab}\frac{}{C^a}\mathrm{\Gamma }^1\widehat{\gamma }b)=\gamma \rho ^C(e_b)\frac{1}{2}g^{ab}\frac{}{C^a}\mathrm{\Gamma }^1(vb)`$. Furthermore, direct computation using skew symmetry of structure constants with lifted indices shows that $`[\rho ^T(e_I),\rho ^C(e_b)\frac{1}{2}g^{ab}\frac{}{C^a}\mathrm{\Gamma }^1]=0`$, which proves that $`H(\widehat{\gamma },(V\rho ^C\mathrm{\Lambda }(C))_{\rho ^T=0}=0`$ and thus (ii).
## 9 Descent equations: $`H(s|d)`$
### 9.1 Introduction
The descent equation technique is a powerful tool to calculate $`H(s|d)`$ which we shall use below. Its usefulness rests on the fact that it relates $`H(s|d)`$ to $`H(s)`$ which is often much simpler than $`H(s|d)`$ \- and which we have determined.
In subsections 9.2, 9.3 and 9.4, we shall review general properties of the descent equations and work out the relation between $`H(s|d)`$ and $`H(s)`$ in detail. Our only assumption for doing so will be that in the space of local forms under study, the cohomology of $`d`$ is trivial at all form-degrees $`p=1,\mathrm{},n1`$ and is represented at $`p=0`$ by pure numbers,
$$H^p(d)=\delta _0^p\text{for}p<n.$$
(9.1)
Since this is the only assumption being made at this stage, the considerations in subsections 9.2 through 9.4 are not restricted to gauge theories of the Yang-Mills type but apply whenever (9.1) is fulfilled.
In the case of theories of the Yang-Mills type (in $`^n`$), the considerations apply in the space of local smooth forms or in the space of local polynomial forms, provided one allows for an explicit $`x`$-dependence. Indeed, the algebraic Poincaré lemma guarantees then that 9.1 holds (theorem 4.2). Although our ultimate goal is to cover the polynomial (or formal power series) case, such a restriction is not necessary in this section. The considerations are also valid in subalgebras of the algebra of local forms for which 9.1 remains true.
However the considerations of this section do not immediately apply, for instance, if no explicit spacetime coordinate dependence is allowed. In this case, the cohomology of $`d`$ is non-trivial in degrees $`0`$ and contains the constant forms (see theorem 4.3). It turns out, however, that the constant forms cannot come in the way, so that the same descent equation techniques in fact apply. This is explained in subsection 9.5.
Finally, we carry out the explicit derivation of the descent equations in the case of the differential $`s`$, but a similar discussion applies to $`\gamma `$ or $`\delta `$ (or, for that matter, any differential $`D`$ such that $`Dd+dD=0`$). In fact, this tool has already been used in section 6 for the mod $`d`$ cohomology of $`\delta `$, to prove the isomorphism $`H_k^p(\delta |d)H_{k1}^{p1}(\delta |d)`$ for $`p>1,k>1`$. The same techniques can be followed for $`H(s|d)`$ (or $`H(\gamma |d)`$), with, however, one complication. While one had $`H_k(\delta )=0`$ ($`k>0`$) for $`\delta `$, the cohomology of $`s`$ is non trivial. As a result, while it is easy to “go down” the descent (because this uses the triviality of $`d`$ – see below), it is more intricate to “go up”.
### 9.2 General properties of the descent equations
We shall now review the derivation and some basic properties of the descent equations, assuming that (9.1) holds.
##### Derivation of the descent equations.
Let $`\omega ^m`$ be a cocycle of $`H^{,m}(s|d)`$,
$$s\omega ^m+d\omega ^{m1}=0.$$
(9.2)
Two cocycles are equivalent in $`H^{,m}(s|d)`$ when they differ by a trivial solution of the consistency condition,
$$\omega ^m\omega ^{}{}_{}{}^{m}\omega ^m\omega ^{}{}_{}{}^{m}=s\eta ^m+d\eta ^{m1}.$$
(9.3)
Applying $`s`$ to Eq. (9.2) yields $`d(s\omega ^{m1})=0`$ (due to $`s^2=0`$ and $`sd+ds=0`$), i.e., $`s\omega ^{m1}`$ is a $`d`$-closed $`(m1)`$-form. Let us assume that $`m1>0`$ (the case $`m1=0`$ is treated below). Using (9.1), we conclude that $`s\omega ^{m1}`$ is $`d`$-exact, i.e., there is an $`(m2)`$-form such that $`s\omega ^{m1}+d\omega ^{m2}=0`$. Hence, $`\omega ^{m1}`$ is a cocycle of $`H^{,m1}(s|d)`$. Moreover, due to the ambiguity (9.3) in $`\omega ^m`$, $`\omega ^{m1}`$ is also determined only up to a coboundary of $`H^{,m1}(s|d)`$. Indeed, when $`\omega ^m`$ solves Eq. (9.2), then $`\omega ^{}{}_{}{}^{m}=\omega ^m+s\eta ^m+d\eta ^{m1}`$ fulfills $`s\omega ^{}{}_{}{}^{m}+d\omega ^{}{}_{}{}^{m1}=0`$ with $`\omega ^{}{}_{}{}^{m1}=\omega ^{m1}+s\eta ^{m1}+d\eta ^{m2}`$. Now, two things can happen:
(a) either $`\omega ^{m1}`$ is trivial in $`H^{,m1}(s|d)`$, $`\omega ^{m1}=s\eta ^{m1}+d\eta ^{m2}`$; then we can substitute $`\omega ^{}{}_{}{}^{m1}=\omega ^{m1}s\eta ^{m1}d\eta ^{m2}=0`$ and $`\omega ^{}{}_{}{}^{m}=\omega ^md\eta ^{m1}`$ for $`\omega ^{m1}`$ and $`\omega ^m`$ respectively and obtain $`s\omega ^{}{}_{}{}^{m}=0`$; we say that $`\omega ^m`$ has a trivial descent;
(b) or $`\omega ^{m1}`$ is nontrivial in $`H^{,m1}(s|d)`$; then there is no way to make $`\omega ^m`$ $`s`$-invariant by adding a trivial solution to it; we say that $`\omega ^m`$ has a nontrivial descent.
In case (b), we treat $`s\omega ^{m1}+d\omega ^{m2}=0`$ as Eq. (9.2) before: acting with $`s`$ on it gives $`d(s\omega ^{m2})=0`$; if $`m20`$, this implies $`s\omega ^{m2}+d\omega ^{m3}=0`$ for some $`(m3)`$-form thanks to (9.1). Again there are two possibilities: either $`\omega ^{m2}`$ is trivial and can be removed through suitable redefinitions such that $`s\omega ^{}{}_{}{}^{m1}=0`$; or it is nontrivial. In the latter case one continues the procedure until one arrives at $`s\omega ^{\underset{¯}{m}}=0`$ at some nonvanishing form-degree $`\underset{¯}{m}`$ (possibly after suitable redefinitions), or until the form-degree drops to zero and one gets the equation $`d(s\omega ^0)=0`$.
From the equation $`d(s\omega ^0)=0`$, one derives, using once again (9.1), $`s\omega ^0=\alpha `$ for some constant $`\alpha `$. If one assumes that the equations of motion are consistent - which one better does! - , then $`\alpha `$ must vanish and the conclusion is the same as in the previous case. This is seen by decomposing $`s\omega ^0=\alpha `$ into pieces with definite antifield number and pure ghost number. Since $`\alpha `$ is a pure number and has thus vanishing antifield number and pure ghost number, the decomposition yields in particular the equation $`\delta a=\alpha `$ where $`a`$ is the piece contained in $`\omega ^0`$ which has antifield number $`1`$ and pure ghost number 0. Due to $`\delta a0`$, this makes only sense if $`\alpha =0`$ because otherwise the equations of motion would be inconsistent (as one could have, e.g., $`0=1`$ on-shell<sup>9</sup><sup>9</sup>9Such an inconsistency would arise, for instance, if one had a neutral scalar field $`\mathrm{\Phi }`$ with Lagrangian $`L=\mathrm{\Phi }`$. This Lagrangian yields the equation of motion $`1=0`$ and must be excluded. Having $`1=s\omega `$ would of course completely kill the cohomology of $`s`$. We have not investigated whether inconsistent Lagrangians (in the above sense) are eliminated by the general conditions imposed on $`L`$ in the introduction, and so, we make the assumption separately. Note that a similar difficulty does not arise in the descent associated with $`\gamma `$; an equation like $`1=\gamma \omega `$ is simply impossible, independently of the Lagrangian, because $`1`$ has vanishing pure ghost number while $`\gamma \omega `$ has pure ghost number equal to or greater than $`1`$. For $`s`$, the relevant grading is the total ghost number and can be negative.).
We conclude: when (9.1) holds, Eq. (9.2) implies in physically meaningful theories that there are forms $`\omega ^p`$, $`p=\underset{¯}{m},\mathrm{},m`$ fulfilling
$$s\omega ^p+d\omega ^{p1}=0\text{for}p=\underset{¯}{m}+1,\mathrm{},m,s\omega ^{\underset{¯}{m}}=0$$
(9.4)
with $`\underset{¯}{m}\{0,\mathrm{},m\}`$. Eqs. (9.4) are called the descent equations. We call the forms $`\omega ^p`$, $`p=\underset{¯}{m},\mathrm{},m1`$ descendants of $`\omega ^m`$, and $`\omega ^{\underset{¯}{m}}`$ the bottom form of the descent equations.
Furthermore we have seen that $`\omega ^m`$ and its descendants are determined only up to coboundaries of $`H(s|d)`$. In fact, for given cohomology class $`H^{,m}(s|d)`$ represented by $`\omega ^m`$, this is the only ambiguity in the solution of the descent equations, modulo constant forms at form-degree 0. This is so because a trivial solution of the consistency condition can only have trivial descendants, except that $`\omega ^0`$ can contain a constant. Indeed, assume that $`\omega ^p`$ is trivial, $`\omega ^p=s\eta ^p+d\eta ^{p1}`$. Inserting this in $`s\omega ^p+d\omega ^{p1}=0`$ gives $`d(\omega ^{p1}s\eta ^{p1})=0`$ and thus, by (9.1), $`\omega ^{p1}=s\eta ^{p1}+d\eta ^{p2}+\delta _0^{p1}\alpha `$ where $`\alpha `$ can occur only when $`p1=0`$. Hence, when $`\omega ^p`$ is trivial, its first descendant $`\omega ^{p1}`$ is necessarily trivial too, except for a possible pure number when $`p=1`$. By induction this applies to all further descendants too.
##### Shortest descents.
The ambiguity in the solution of the descent equations implies in particular that all nonvanishing forms which appear in the descent equations can be chosen such that none of them is trivial in $`H(s|d)`$ because otherwise we can “shorten” the descent equations. In particular, there is thus a “shortest descent” (i.e., a maximal value of $`\underset{¯}{m}`$) for every nontrivial cohomology class $`H^{,m}(s|d)`$. A shortest descent is realized precisely when all the forms in the descent equations are nontrivial. An equivalent characterization of a shortest descent is that the bottom form $`\omega ^{\underset{¯}{m}}`$ is nontrivial in $`H^{,\underset{¯}{m}}(s|d)`$ if $`\underset{¯}{m}>0`$, respectively that it is nontrivial in $`H^{,0}(s|d)`$ even up to a constant if $`\underset{¯}{m}=0`$ (i.e., that $`\omega ^{\underset{¯}{m}}s\eta ^{\underset{¯}{m}}+d\eta ^{\underset{¯}{m}1}+\delta _0^{\underset{¯}{m}1}\alpha `$, $`\alpha `$). The latter statement holds because the triviality of any nonvanishing form in the descent equations implies necessarily that all its descendants, and thus in particular $`\omega ^{\underset{¯}{m}}`$, are trivial too except for a number that can contribute to $`\omega ^0`$. Of course, the shortest descent is not unique since one may still make trivial redefinitions which do not change the length of a descent.
### 9.3 Lifts and obstructions
We have seen that the bottoms $`\omega ^{\underset{¯}{m}}`$ of the descent equations associated with solutions $`\omega ^m`$ of the consistency conditions $`s\omega ^m+d\omega ^{m1}=0`$ are cocycles of $`s`$, $`s\omega ^{\underset{¯}{m}}=0`$, which are non trivial in $`H^{,0}(s|d)`$ (even up to a constant if $`\underset{¯}{m}=0`$). In particular, they are non trivial in $`H(s)`$.
One can conversely ask the following questions. Given a non trivial cocycle of $`H(s)`$: (i) Is it trivial in $`H(s|d)`$?; (ii) Can it be viewed as bottom of a non trivial descent? These questions were raised for the first time in and turn out to contain the key to the calculation of $`H(s|d)`$ in theories of the Yang-Mills type.
We say that an $`s`$-cocycle $`\omega ^p`$ can be “lifted” $`k`$ times if there are forms $`\omega ^{p+1},\mathrm{},\omega ^{p+k}`$ such that $`d\omega ^p+s\omega ^{p+1}=0`$, …, $`d\omega ^{p+k1}+s\omega ^{p+k}=0`$. Contrary to the descent, which is never obstructed, the lift of an element of $`H(s)`$ can be obstructed because the cohomology of $`s`$ is non trivial. Let $`a`$ be an $`s`$-cocycle and let us try to construct an element “above it”. To that end, one must compute $`da`$ and see whether it is $`s`$-exact. It is clear that $`da`$ is $`s`$-closed; the obstructions to it being $`s`$-exact are thus in $`H(s)`$. Two things can happen. Either $`da`$ is not $`s`$-exact,
$$da=m$$
(9.5)
with $`m`$ a non trivial cocycle of $`H(s)`$. Or $`da`$ is $`s`$-exact, in which case one has
$$da+sb=0$$
(9.6)
for some $`b`$. Of course, $`b`$ is defined up to a cocycle of $`s`$.
In the first case, it is clear that “the obstruction” $`m`$ to lifting $`a`$, although non trivial in $`H(s)`$ is trivial in $`H(s|d)`$. Furthermore $`a`$ itself cannot be trivial in $`H(s|d)`$ since trivial elements $`a=su+dv`$ can always be lifted ($`da=s(du)`$).
In the second case, one may try to lift $`a`$ once more. Thus one computes $`db`$. Again, it is easy to verify that $`db`$ is an $`s`$-cocycle. Therefore, either $`db`$ is not $`s`$-exact,
$$db+sc=n\text{ “Case A”}$$
(9.7)
for some non trivial cocycle $`n`$ of $`H(s)`$ (we allow here for the presence of the exact term $`sc`$ \- which can be absorbed in $`n`$ \- because usually, one has natural representatives of the classes of $`H(s)`$, and $`db`$ may differ from such a representative $`n`$ by an $`s`$-exact term). Or $`db`$ is $`s`$-exact,
$$db+sc=0\text{ “Case B”}$$
(9.8)
for some $`c`$.
Note that in case A, $`b`$ is defined up to the addition of an $`s`$-cocycle, so the “obstruction” $`n`$ to lifting $`a`$ a second time is really present only if $`n`$ cannot be written as $`dt+sq`$ where $`t`$ is an $`s`$-cocycle, i.e., if $`n`$ is not in fact the obstruction to the first lift of some $`s`$-cocycle. The obstructions to second lifts are therefore in the space $`H(s)/\mathrm{Im}d`$ of the cohomology of $`s`$ quotientized by the space of obstructions to first lifts. If the obstruction to lifting $`a`$ a second time is really present, then $`a`$ is clearly non trivial in $`H(s|d)`$. And in any case, $`n`$ is trivial in $`H(s|d)`$.
In case B, one can continue and try to lift $`a`$ a third time. This means computing $`dc`$. The analysis proceeds as above and is covered by the results of the next subsection.
### 9.4 Length of chains and structure of $`H(s|d)`$
By following the above procedure, one can construct a basis of $`H(s)`$ which displays explicitly the lift structure and the obstructions.
###### Theorem 9.1
If $`H^p(d,\mathrm{\Omega })=\delta _0^p`$ for $`p=0,\mathrm{},n1`$ and the equations of motions are consistent, there exists a basis
$`\{[1],[h_{i_r}^0],[\widehat{h}_{i_r}],[e_{\alpha _s}^0]\}`$ (9.9)
of $`H(s)`$ such that the representatives fulfill
$`sh_{i_r}^{r+1}+dh_{i_r}^r=\widehat{h}_{i_r},`$
$`sh_{i_r}^r+dh_{i_r}^{r1}=0,`$
$`\mathrm{}`$ (9.10)
$`sh_{i_r}^1+dh_{i_r}^0=0,`$
$`sh_{i_r}^0=0`$
and
$`\mathrm{form}\mathrm{degree}e_{\alpha _s}^s=n,`$
$`se_{\alpha _s}^s+de_{\alpha _s}^{s1}=0,`$
$`\mathrm{}`$ (9.11)
$`se_{\alpha _s}^1+de_{\alpha _s}^0=0,`$
$`se_{\alpha _s}^0=0,`$
for some forms $`h_{i_r}^q`$, $`q=1,\mathrm{},r+1`$ and $`e_{\alpha _s}^p`$, $`p=1,\mathrm{},s`$. Here, $`[a]`$ denotes the class of the $`s`$-cocycle $`a`$ in $`H(s)`$.
We recall that a set $`\{f_A\}`$ of $`s`$-cocycles is such that the set $`\{[f_A]\}`$ forms a basis of $`H(s)`$ if and only if the following two properties hold: (i) any $`s`$-cocycle is a linear combination of the $`f_A`$’s, up to an $`s`$-exact term; and (ii) if $`\lambda ^Af_A=sg`$, then the coefficients $`\lambda ^A`$ all vanish.
The elements of the basis (9.9) have the following properties: The $`h_{i_r}^0`$ can be lifted $`r`$ times, until one hits an obstruction given by $`\widehat{h}_{i_r}`$. By contrast, the $`e_{\alpha _s}^0`$ can be lifted up to maximum degree without meeting any obstruction. We stress that the superscripts of $`h_{i_r}^q`$ and $`e_{\alpha _s}^p`$ in the above theorem do not indicate the form-degree but the increase of the form-degree relative to $`h_{i_r}^0`$ and $`e_{\alpha _s}^0`$ respectively. The form-degree of $`h_{i_r}^0`$ is not determined by the above formulae except that it is smaller than $`nr`$. $`e_{\alpha _s}^0`$ has form-degree $`ns`$.
We shall directly construct bases with such properties in the Yang-Mills setting, for various (sub)algebras fulfilling (9.1), thereby proving explicitly their existence in the concrete cases relevant for our purposes. The proof of the theorem in the general case is given in appendix 9.A following . We refer the interested reader to the pioneering work of for a proof involving more powerful homological tools (“exact couples”).
For a basis of $`H(s)`$ with the above properties, the eqs (9.10) provide optimum lifts of the $`h_{i_r}^0`$. The $`\widehat{h}_{i_r}`$ represent true obstructions; by using the ambiguities in the successive lifts of $`h_{i_r}^0`$, one cannot lift $`h_{i_r}^0`$ more than $`r`$ times. This is seen by using a recursive argument. It is clear that the $`h_{i_0}^0`$ cannot be lifted at all since the $`\widehat{h}_{i_0}`$ are independent in $`H(s)`$. Consider next $`h_{i_1}^0`$ and the corresponding chain, $`sh_{i_1}^0=0`$, $`dh_{i_1}^0+sh_{i_1}^1=0`$, $`dh_{i_1}^1+sh_{i_1}^2=\widehat{h}_{i_1}`$. Suppose that the linear combination $`\alpha ^{i_1}h_{i_1}^0`$ could be lifted more than once, which would occur if and only if $`\alpha ^{i_1}\widehat{h}_{i_1}`$ was the obstruction to the single lift of an $`s`$-cocycle, i.e., $`\alpha ^{i_1}\widehat{h}_{i_1}=da+sb`$, $`sa=0`$. Since (9.9) provides a basis of $`H(s)`$, one could expand $`a`$ in terms of $`\{1,h_{i_r}^0,\widehat{h}_{i_r},e_{\alpha _s}^0\}`$ (up to an $`s`$-exact term), $`a=\alpha ^{i_0}h_{i_0}^0+\mathrm{}`$. Computing $`da`$ using (9.10) and (9.11), and inserting the resulting expression into $`\widehat{h}_{i_1}=da+sb`$, one gets $`\widehat{h}_{i_1}=\alpha ^{i_0}\widehat{h}_{i_0}+s()`$, leading to a contradiction since the $`\widehat{h}_{i_1}`$ and $`\widehat{h}_{i_0}`$ are independent in cohomology. The argument can be repeated in the same way for bottoms leading to longer lifts and is left to the reader.
It follows from this analysis that the $`h_{i_r}^0`$ are $`s`$-cocycles that are non-trivial in $`H(s|d)`$ \- while, of course, the $`\widehat{h}_{i_r}`$ are trivial. In fact, the advantage of a basis of the type of (9.9) for $`H(s)`$ is that it gives immediately the cohomology of $`H(s|d)`$.
###### Theorem 9.2
If $`\{[1],[h_{i_r}^0],[\widehat{h}_{i_r}],[e_{\alpha _s}^0]\}`$ is a basis of $`H(s)`$ with the properties of theorem 9.1, then an associated basis of $`H(s|d)`$ is given by
$`\{[1],[h_{i_r}^q],[e_{\alpha _s}^p]:q=0,\mathrm{},r,p=0,\mathrm{},s\}`$ (9.12)
where in this last list, $`[]`$ denotes the class in $`H(s|d)`$.
##### Proof:
The proof is given in the appendix 9.B.
The theorem shows in particular that the $`e_{\alpha _s}^0`$, just like the $`h_{i_r}^0`$, are non trivial $`s`$-cocycles that remain non trivial in $`H(s|d)`$. This property holds even though they can be lifted all the way to maximum form-degree $`n`$ (while the lifts of the $`h_{i_r}^0`$ are obstructed before). The basis of $`H(s|d)`$ is given by the non trivial bottoms $`h_{i_r}^0`$, $`e_{\alpha _s}^0`$ and all the terms in the descent above them (up to the obstructions in the case of $`h_{i_r}^0`$).
### 9.5 Descent equations with weaker assumptions on $`H(d)`$
So far we have assumed that (9.1) holds. We shall now briefly discuss the modifications when (9.1) is replaced by an appropriate weaker prerequisite. Applications of these modifications are described below.
Let $`\{\alpha _{i_p}^p\}`$ be a set of $`p`$-forms representing $`H^p(d)`$ (hence, the superscript of the $`\alpha `$’s indicates the form-degree, the subscript $`i_p`$ labels the inequivalent $`\alpha `$’s for fixed $`p`$). That is, any $`d`$-closed $`p`$-form is a linear combination of the $`\alpha _{i_p}^p`$ with constant coefficients $`\lambda ^{i_p}`$, modulo a $`d`$-exact form,
$$d\omega ^p=0\omega ^p=\lambda ^{i_p}\alpha _{i_p}^p+d\eta ^{p1},d\alpha _{i_p}^p=0,$$
(9.13)
and no nonvanishing linear combination of the $`\alpha _{i_p}^p`$ is $`d`$-exact.
Now, in order to derive the descent equations, it is quite crucial that the equality $`d(s\omega ^p)=0`$ implies $`s\omega ^p+d\omega ^{p1}=0`$. However, if $`H^p(d)`$ is non trivial, we must allow for a combination of the $`\alpha _{i_p}^p`$ on the right hand side, and this spoils the descent. This phenomenon cannot occur if no $`\alpha _{i_p}^p`$ is $`s`$-exact modulo $`d`$. Thus, in order to be able to use the tools provided by the descent equations, we shall assume that the forms non trivial in $`H^p(d)`$ remain non trivial in $`H(s|d)`$ for $`p<n`$. More precisely, it is assumed that the $`\alpha _{i_p}^p`$ with $`p<n`$ have the property that no nonvanishing linear combination of them is trivial in $`H(s|d)`$,
$$p<n:\lambda ^{i_p}\alpha _{i_p}^p=s\eta ^p+d\eta ^{p1}\lambda ^{i_p}=0i_p.$$
(9.14)
This clearly implies the central property of the descent,
$$p<n:d(s\omega ^p)=0\omega ^{p1}:s\omega ^p+d\omega ^{p1}=0.$$
(9.15)
Indeed, by (9.13), $`d(s\omega ^p)=0`$ implies $`s\omega ^p+d\omega ^{p1}=\lambda ^{i_p}\alpha _{i_p}^p`$ for some $`\omega ^{p1}`$ and $`\lambda ^{i_p}`$. (9.14) implies now $`\lambda ^{i_p}=0`$ whenever $`p<n`$.
When (9.14) holds, the discussion of the descent equations proceeds as before. The only new feature is the fact that the $`\alpha _{i_p}^p`$ yield additional nontrivial classes of $`H(s|d)`$ and $`H(s)`$. Indeed, $`d\alpha _{i_p}^p=0`$ implies $`d(s\alpha _{i_p}^p)=0`$ and thus, due to (9.15), $`s\alpha _{i_p}^p+d\alpha _{i_p}^{p1}=0`$ for some $`\alpha _{i_p}^{p1}`$ (which may vanish). Hence, the $`\alpha _{i_p}^p`$ are cocycles of $`H(s|d)`$ and they are nontrivial by (9.14). In particular, some of the $`\alpha _{i_p}^p`$ may have a nontrivial descent. Theorems (9.1) and (9.2) get modified because the $`\alpha _{i_p}^p`$ and their nontrivial descendants (if any) represent classes of $`H(s|d)`$ in addition to the $`[h_{i_r}^q]`$ and $`[e_{\alpha _s}^p]`$ ($`q=0,\mathrm{},r`$, $`p=0,\mathrm{},s`$), while $`H(s)`$ receives additional classes represented by nontrivial bottom forms corresponding to the $`\alpha _{i_p}^p`$.
##### Applications.
1. The above discussion is important to cover the space of local forms which are not allowed to depend explicitly on the $`x^\mu `$. Indeed, in that space $`H^p(d)`$ is represented for $`p<n`$ by the constant forms $`c_{\mu _1\mathrm{}\mu _p}dx^{\mu _1}\mathrm{}dx^{\mu _p}`$ (theorem 4.3). Now, the equations of motion may be such that some of the constant forms become trivial in $`H(s|d)`$. We know that this cannot happen in form-degree zero, but nothing prevents it from happening in higher form-degrees. For instance, for a single abelian gauge field with Lagrangian $`L=(1/4)F^{\mu \nu }F_{\mu \nu }+k^\mu A_\mu `$ with constant $`k^\mu `$, the equations of motion read $`_\nu F^{\nu \mu }+k^\mu =0`$ and imply that the constant $`(n1)`$-form $`k=\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}k^{\mu _n}`$ is trivial in $`H(s|d)`$, $`sA^{}+dF+()^nk=0`$. Hence, for this Lagrangian (9.14) is not fulfilled. Of course, the example is academic and the Lagrangian is not Lorentz-invariant. The triviality of $`k`$ in $`H(s|d)`$ is a consequence of the linear term in the Lagrangian.
(9.14) is fulfilled in the space of $`x`$-independent local forms for Lagrangians having no linear part in the fields, for which the equations of motion reduce identically to $`0=0`$ when the fields are set to zero. It is also fulfilled if one restricts one’s attention to the space of Poincaré invariant local forms. \[And it is also trivially fulfilled in the space of all local forms with a possible explicit $`x`$-dependence, as we have seen\]. For this reason, (9.14) does not appear to be a drastic restriction in the space of $`x`$-independent local forms.
Note that the classification of the elements (and the number of these elements) in a basis of $`H(s)`$ having the properties of theorem 9.1 depends on the context. For instance, for a single abelian gauge field with ghost $`C`$, $`dx^\mu C`$ is non trivial in the algebra of local forms with no explicit $`x`$-dependence, and so can be taken as a $`h_{i_1}^0`$; but it becomes trivial if one allows for an explicit $`x`$-dependence, $`dx^\mu C=d(x^\mu C)+s(x^\mu A)`$, and so can be regarded in that case as a $`\widehat{h}_{i_0}`$.
2. Another instance where the descent equations with the above assumption on $`H(d)`$ play a rôle is the cohomology $`H(\delta |d,_\chi )`$, where $`_\chi `$ is the space of $`𝒢`$-invariant local forms depending only on the variables $`\chi _\mathrm{\Delta }^u`$ defined in section 8.1 and on the $`x^\mu `$ and $`dx^\mu `$. $`H^p(d,_\chi )`$ is represented for $`p<n`$ by the “characteristic classes”, i.e., by $`𝒢`$-invariant polynomials $`P(F)`$ in the curvature 2-forms $`F^I`$ (these polynomials are $`d`$-exact in the space of all local forms, but they are not $`d`$-exact in $`_\chi `$). Using this result on $`H^p(d,_\chi )`$, one proves straightforwardly by means of the descent equations that $`H(\delta |d,_\chi )`$ is isomorphic to the characteristic cohomology in the space of gauge invariant local forms (“equivariant characteristic cohomology”) which will play an important rôle in the analysis of the consistency condition performed in section 11.
3. Finally we mention that the above discussion was used within the computation of the local BRST cohomology in Einstein-Yang-Mills theory . In that case $`H^p(d)`$ is nontrivial in certain form-degrees $`p<n`$ due to the nontrivial De Rham cohomology of the manifold in which the vielbein fields take their values. The corresponding $`\alpha _{i_p}^p`$ fulfill (9.14) and have a nontrivial descent (in contrast, the constant forms and the characteristic classes met in the two instances discussed before do not descend).
### 9.6 Cohomology of $`s+d`$
The descent equations establish a useful relation between $`H^{,n}(s|d)`$, $`H(d)`$ and the cohomology of the differential $`\stackrel{~}{s}`$ that combines $`s`$ and $`d`$,
$$\stackrel{~}{s}=s+d.$$
(9.16)
Note that $`\stackrel{~}{s}`$ squares to zero thanks to $`s^2=sd+ds=d^2=0`$ and defines thus a cohomology $`H(\stackrel{~}{s})`$ in the space of formal sums of forms with various degrees,
$$\stackrel{~}{\omega }=\underset{p}{}\omega ^p.$$
We call such sums total forms. Cocycles of $`H(\stackrel{~}{s})`$ are defined through
$$\stackrel{~}{s}\stackrel{~}{\omega }=0,$$
(9.17)
while coboundaries take the form $`\stackrel{~}{\omega }=\stackrel{~}{s}\stackrel{~}{\eta }`$. Consider now a cocycle $`\stackrel{~}{\omega }=_{p=\underset{¯}{m}}^m\omega ^p`$ of $`H(\stackrel{~}{s})`$. The cocycle condition (9.17) decomposes into
$$d\omega ^m=0,s\omega ^m+d\omega ^{m1}=0,\mathrm{},s\omega ^{\underset{¯}{m}}=0.$$
These are the descent equations with top-form $`\omega ^m`$ and the supplementary condition $`d\omega ^m=0`$. The extra condition is of course automatically fulfilled if $`m=n`$. Hence, assuming that Eq. (9.14) holds, every cohomology class of $`H^{,n}(s|d)`$ gives rise to a cohomology class of $`H(\stackrel{~}{s})`$. Evidently, this cohomology class of $`H(\stackrel{~}{s})`$ is nontrivial if its counterpart in $`H^{,n}(s|d)`$ is nontrivial (since $`\stackrel{~}{\omega }=\stackrel{~}{s}\stackrel{~}{\eta }`$ implies $`\omega ^n=s\eta ^n+d\eta ^{n1}`$). All additional cohomology classes of $`H(\stackrel{~}{s})`$ (i.e. those without counterpart in $`H^{,n}(s|d)`$) correspond precisely to the cohomology classes of $`H(d)`$ in form degrees $`<n`$ and thus to the $`\alpha _{i_p}^p`$ with $`p<n`$ in the notation used above. Indeed, $`d\omega ^m=0`$ implies $`\omega ^m=\lambda ^{i_m}\alpha _{i_m}^m+d\eta ^{m1}`$ by (9.13). Furthermore, as shown above, each $`\alpha _{i_m}^m`$ gives rise to a solution of the descent equations and thus to a representative of $`H(\stackrel{~}{s})`$. These representatives are nontrivial and inequivalent because of (9.14). This yields the following isomorphism whenever (9.14) holds:
$$H(\stackrel{~}{s})H^{,n}(s|d)H^{n1}(d)\mathrm{}H^0(d).$$
(9.18)
This isomorphism can be used, in particular, to determine $`H^{,n}(s|d)`$ by computing $`H(\stackrel{~}{s})`$. The cohomology of $`s`$ modulo $`d`$ in lower form degrees is however not given by $`H(\stackrel{~}{s})`$.
### 9.7 Appendix 9.A: Proof of theorem 9.1
In order to prove the existence of (9.9), we note first that the space $`\mathrm{\Omega }`$ of local forms admits the decomposition $`\mathrm{\Omega }=E_0GsG`$ for some space $`E_0H(s,\mathrm{\Omega })`$ and some space $`G`$.
Following , we define recursively differentials $`d_r:H(d_{r1})H(d_{r1})`$ for $`r=0,\mathrm{},n`$, with $`d_1s`$ as follows: the space $`H(d_{r1})`$ is given by equivalence classes $`[X]_{r1}`$ of elements $`X\mathrm{\Omega }`$ such that there exist $`c_1,\mathrm{},c_r`$ satisfying $`sX=0,dX+sc_1=0,\mathrm{},dc_{r1}+sc_r=0`$ (i.e., $`X`$ can be lifted at least $`r`$ times). We define $`c_0X`$. The equivalence relation $`[]_{r1}`$ is defined by $`X_{r1}Y`$ iff $`XY=sZ+d(v_0^0+\mathrm{}+v_{r1}^{r1})`$ where $`sv_j^j+dv_j^{j1}=0,\mathrm{},sv_j^0=0`$. $`d_r`$ is defined by $`d_r[X]_{r1}=[dc_r]_{r1}`$. Let us check that this definition makes sense. Applying $`d`$ to $`dc_{r1}+sc_r=0`$ gives $`sdc_r=0`$, so that $`[dc_r]_{r1}H(d_{r1})`$ (one can simply choose the required $`c_1^{},\mathrm{},c_r^{}`$ to be zero because of $`d(dc_r)=0`$). Furthermore, $`[dc_r]_{r1}`$ does not depend on the ambiguity in the definition of $`X,c_1,\mathrm{},c_r`$. Indeed, the ambiguity in the definition of $`X`$ is $`sZ+d(v_0^0+\mathrm{}+v_{r1}^{r1})`$, the ambiguity in the definition of $`c_1`$ is $`dZ+m_{r1}^0`$, where $`sm_{r1}^0=0,dm_{r1}^0+sm_{r1}^1=0,\mathrm{},dm_{r1}^{r2}+sm_{r1}^{r1}=0`$. Similarly, the ambiguity in $`c_2`$ is $`m_{r1}^1+m_{r2}^0`$, where $`sm_{r2}^0=0,dm_{r2}^0+sm_{r2}^1=0,\mathrm{},dm_{r2}^{r3}+sm_{r2}^{r2}=0`$. Going on in the same way, one finds that the ambiguity in $`c_r`$ is $`m_{r1}^{r1}+\mathrm{}+m_0^0`$, where $`sm_j^j+dm_j^{j1}=0`$. This means that the ambiguity in $`dc_r`$ is $`d(m_{r1}^{r1}+\mathrm{}+m_0^0)`$, which is zero in $`H(d_{r1})`$. (For $`r=0`$, the ambiguity in $`dc_0dX`$ is $`sdZ`$, which is zero in $`H(s)`$.) Finally, $`d_r^2=0`$ follows from $`d^2=0`$.
The cocycle condition $`d_r[X]_{r1}=0`$ reads explicitly $`dc_r=sW+d(w_0^0+\mathrm{}+w_{r1}^{r1})`$, where $`sw_j^j+dw_j^{j1}=0,\mathrm{},sw_j^0=0`$. This means that $`sX=0`$, $`dX+s(c_1w_{r1}^0)=0`$, $`d(c_1w_{r1}^0)+s(c_2w_{r1}^1w_{r2}^0)=0`$, $`\mathrm{}`$, $`d(c_rw_{r1}^{r1}\mathrm{}w_0^0)+s(W)=0`$. The coboundary condition $`[X]_{r1}=d_r[Y]_{r1}`$ gives $`X=db_r+sZ+d(v_0^0+\mathrm{}+v_{r1}^{r1})`$, where $`sY=0,dY+sb_1=0,\mathrm{},db_{r1}+sb_r=0`$. The choices $`c_1^{}=c_1w_{r1}^0,\mathrm{},c_r^{}=c_rw_{r1}^{r1}\mathrm{}w_0^0`$, $`c_{r+1}^{}=W`$, respectively $`v_{}^{}{}_{j}{}^{j}=v_j^j`$ for $`j=0,\mathrm{},r1`$ and $`v_r^r=b_r`$, $`v_r^{r1}=b_{r1}`$, $`\mathrm{}`$, $`v_r^1=b_1`$, $`v_r^0=Y`$, show that $`H(d_r)`$ is defined by the same equations as $`H(d_{r1})`$ with $`r1`$ replaced by $`r`$, as it should.
Because the maximum form degree is $`n`$, one has $`d_n0`$ and the construction stops.
It is now possible to define spaces $`E_rE_{r1}E_0\mathrm{\Omega }`$ and spaces $`F_{r1}E_{r1}\mathrm{\Omega }`$ for $`r=1,\mathrm{},n`$ such that $`E_{r1}=E_rd_{r1}F_{r1}F_{r1}`$ with $`E_rH(d_{r1})`$. This leads to the decomposition
$`E_0=E_n(d_{n1}F_{n1}F_{n1})(d_{n1}F_{n1}F_{n1})\mathrm{}(d_0F_{n1}F_0).`$ (9.19)
Note that this decomposition involves choices of representatives $`(E_i)`$ and of supplementary subspaces $`(F_i)`$ and is not “canonical”. But it does exist.
The $`e_{\alpha _s}^0`$ are elements of a basis of $`E_n`$. They can be lifted $`s`$ times to form degree $`n`$, i.e., they are of form degree $`ns`$. The element $`1`$ also belongs to $`E_n`$. The $`\widehat{h}_{i_r}`$ and $`h_{i_r}^0`$ are elements of a basis of $`d_rF_r`$ and $`F_r`$ respectively.
### 9.8 Appendix 9.B: Proof of theorem 9.2
We verify here that if a basis (9.9) with properties (9.10) and (9.11) exists, then a basis of $`H(s|d)`$ is given by (9.12).
##### The set (9.12) is complete.
Suppose that $`s\omega ^l+d\omega ^{l1}=0,`$ $`s\omega ^{l1}+d\omega ^{l2}=0,\mathrm{},`$ $`s\omega ^0=0`$ where the superscript indicates the length of the descent (number of lifts) rather than the form-degree. We shall prove that then
$$\omega ^l=C+\underset{0ql}{}\underset{rlq}{}\lambda _q^{i_r}h_{i_r}^{lq}+\underset{0pl}{}\underset{slp}{}\mu _p^{\alpha _s}e_{\alpha _s}^{lp}+s\eta ^l+d[\eta ^{l1}+\underset{r0}{}\nu ^{(l)i_r}h_{i_r}^r],$$
(9.20)
with $`C`$ a constant and $`\eta ^1=0`$.
To prove this, we proceed recursively in the length $`l`$ of the descent. For $`l=0`$, we have $`s\omega ^0=0`$. Using the assumption that the $`h^0`$’s, $`\widehat{h}`$’s, $`e^0`$’s and the number 1 provide a basis of $`H(s)`$, this gives $`\omega ^0=C+_{r0}\lambda _0^{i_r}h_{i_r}^0+_{s0}\mu _0^{\alpha _s}e_{\alpha _s}^0+_{r0}\nu ^{(0)i_r}\widehat{h}_{i_r}+s\stackrel{~}{\eta }^0=_{r0}\lambda _0^{i_r}h_{i_r}^0+_{s0}\mu _0^{\alpha _s}e_{\alpha _s}^0+s\eta ^0+d(_{r0}\nu ^{(0)i_r}h_{i_r}^r)`$, where $`\eta ^0=\stackrel{~}{\eta }^0+_{r0}\nu ^{(0)i_r}h_{i_r}^{r+1}`$. This is (9.20) for $`l=0`$.
We assume now that (9.20) holds for $`l=L`$. Then we have for $`l=L+1`$, that in $`s\omega ^{L+1}+d\omega ^L=0`$, $`\omega ^L`$ is given by (9.20) with $`l`$ replaced by $`L`$. Using (9.10) and (9.11), one gets $`d\omega _L=s[_{0qL}_{rLq+1}\lambda _q^{i_r}h_{i_r}^{Lq+1}_{0pl}_{sLp+1}\mu _p^{\alpha _s}e_{\alpha _s}^{Lp+1}d\eta ^L_{0qL}\lambda _q^{i_{Lq}}h_{i_{Lq}}^{Lq+1}]+_{0qL}\lambda _q^{i_{Lq}}\widehat{h}_{i_{Lq}}`$. Injecting this into $`s\omega ^{L+1}+d\omega ^L=0`$, we find, using the properties of the basis and the first relation of (9.10), on the one hand that $`\lambda _q^{i_{Lq}}=0`$, and on the other hand that $`\omega ^{L+1}=C+_{0qL}_{rLq+1}\lambda _q^{i_r}h_{i_r}^{Lq+1}+_{0pl}_{sLp+1}\mu _p^{\alpha _s}e_{\alpha _s}^{Lp+1}+_{r0}\lambda _{L+1}^{i_r}h_{i_r}^0+_{s0}\mu _{L+1}^{\alpha _s}e_{\alpha _s}^0+s\eta ^{L+1}+d[\eta ^L+_{r0}\nu ^{(L+1)i_r}h_{i_r}^r]`$, which is precisely (9.20) for $`l=L+1`$.
##### The elements of (9.12) are cohomologically independent.
Suppose now that
$$C+\underset{0ql}{}\underset{rlq}{}\lambda _q^{i_r}h_{i_r}^{lq}+\underset{0pl}{}\underset{slp}{}\mu _p^{\alpha _s}e_{\alpha _s}^{lp}=s\stackrel{~}{\eta }^{(l)}+d\eta ^{(l)}.$$
(9.21)
We have to show that this implies that $`\lambda _q^{i_r}=0`$, $`\mu _p^{\alpha _s}=0`$ and $`C=0`$.
Again, we proceed recursively on the length of the descent. For $`l=0`$, we have $`C+_{r0}\lambda _0^{i_r}h_{i_r}^0+_{s0}\mu _0^{\alpha _s}e_{\alpha _s}^0=s\stackrel{~}{\eta }^{(0)}+d\eta ^{(0)}`$. Applying $`s`$ and using the triviality of the cohomology of $`d`$, we get that $`\eta ^{(0)}`$ is an $`s`$ modulo $`d`$ cocycle, $`s\eta ^{(0)}+d()=0`$ (without constant since the equations of motion are consistent). Suppose that the descent of $`\eta ^{(0)}`$ stops after $`l^{}`$ steps with $`0l^{}n1`$, i.e., $`\eta ^{(0)}\eta ^{(0)l^{}}`$ with $`s\eta ^{(0)l^{}}+d\eta ^{(0)l^{}1}=0,\mathrm{},`$ $`s\eta ^{(0)0}=0`$ (again no constant here). It follows that $`\eta ^{(0)l^{}}`$ is given by (9.20) with $`l`$ replaced by $`l^{}`$. Evaluating then $`d\eta ^{(0)l^{}}`$ in the equation for $`l=0`$ and using the properties of the basis implies that $`\lambda _0^{i_r}=\mu _0^{\alpha _s}=C=0`$.
Suppose now that the result holds for $`l=L`$. If we apply $`s`$ to (9.21) at $`l=L+1`$ and use the triviality of the cohomology of $`d`$ we get $`_{0qL}_{rL+1q}\lambda _q^{i_r}h_{i_r}^{Lq}+_{0pL}_{sL+1p}\mu _p^{\alpha _s}e_{\alpha _s}^{Lp}=s\eta ^{(L+1)}+d()`$. The induction hypothesis implies then that $`\lambda _q^{i_r}=0=\mu _p^{\alpha _s}`$ for $`0qL`$ and $`0pL`$. This implies that the relation at $`l=L+1`$ reduces to $`_{r0}\lambda _{L+1}^{i_r}h_{i_r}^0+_{s0}\mu _{L+1}^{\alpha _s}e_{\alpha _s}^0=s\stackrel{~}{\eta }^{(l)}+d\eta ^{(l)}`$. As we have already shown, this implies that $`\lambda _{L+1}^{i_r}=0=\mu _{L+1}^{\alpha _s}`$.
## 10 Cohomology in the small algebra
### 10.1 Definition of small algebra
The “small algebra” $``$ is by definition the algebra of polynomials in the undifferentiated ghosts $`C^I`$, the gauge field 1-forms $`A^I`$ and their exterior derivatives $`dC^I`$ and $`dA^I`$.
$$=\{\text{polynomials in }C^I,A^I,dC^I,dA^I\}.$$
It is stable under $`d`$ and $`s`$ ($`b`$ $``$ $`db`$, $`sb`$). This is obvious for $`d`$ and holds for $`s`$ thanks to
$`sC^I={\displaystyle \frac{1}{2}}ef_{JK}^{}{}_{}{}^{I}C^KC^J,sA^I=dC^Ief_{JK}^{}{}_{}{}^{I}C^JA^K,`$
$`sdC^I=ef_{JK}^{}{}_{}{}^{I}C^JdC^K,sdA^I=ef_{JK}^{}{}_{}{}^{I}(C^KdA^JA^JdC^K).`$ (10.1)
Accordingly, the cohomological groups $`H(s|d,)`$ of $`s`$ modulo $`d`$ in $``$ are well defined<sup>10</sup><sup>10</sup>10Note that $`s`$ and $`\gamma `$ coincide in the small algebra, which contains no antifield..
The small algebra $``$ is only a very small subspace of the complete space of all local forms (in fact $``$ is finite dimensional whereas the space of all local forms is infinite dimensional). Nevertheless it provides a good deal of the BRST cohomology in Yang-Mills theories, in that it contains all the antifield-independent solutions of the consistency condition $`sa+db=0`$ that descend non-trivially, in a sense to be made precise in section 11. Furthermore, it will also be proved there that the representatives of $`H(s|d,)`$ remain nontrivial in the full cohomology, with only very few possible exceptions.
For this reason, the calculation of $`H(s|d,)`$ is an essential part of the calculation of $`H(s|d)`$ in the full algebra. This calculation was done first in (in fact in the universal algebra defined below).
Let us briefly outline the construction of $`H(s|d,)`$ before going into the details. It is based on an analysis of the descent equations described in Section 9. The central task in this approach is the explicit construction of a particular basis of $`H(s,)`$ as in theorem 9.1 which provides $`H(s|d,)`$ by theorem 9.2. Technically it is of great help that the essential steps of this construction can be carried out in a free differential algebra associated with $``$. This is shown first.
### 10.2 Universal algebra
The free differential algebra associated with $``$ is denoted by $`𝒜`$. It was called the “universal algebra” in . It has the same set of generators as $``$, but these are not constrained by the condition coming from the spacetime dimension that there is no exterior form of form-degree higher than $`n`$.
Explicitly, $`𝒜`$ is generated by anticommuting variables $`C_𝒜^I`$, $`A_𝒜^I`$ and commuting variables $`(dC)_𝒜^I`$, $`(dA)_𝒜^I`$ which correspond to $`C^I`$, $`A^I`$, $`dC^I`$ and $`dA^I`$ respectively, $`𝒜`$ is the space of polynomials in these variables,
$$𝒜=\{\text{polynomials in }C_𝒜^I,A_𝒜^I,(dC)_𝒜^I,(dA)_𝒜^I\}.$$
These variables are subject only to the commutation/anticommutation relations $`C_𝒜^IC_𝒜^J=C_𝒜^JC_𝒜^I`$ etc but are not constrained by the further condition that the forms are zero whenever their form-degree exceeds $`n`$. Thus, the free differential algebra $`𝒜`$ is independent of the spacetime dimension; hence its name “universal”.
The fundamental difference between $`𝒜`$ and $``$ is that the $`A_𝒜^I`$, $`(dC)_𝒜^I`$ and $`(dA)_𝒜^I`$ are variables by themselves, whereas their counterparts $`A^I`$, $`dC^I`$ and $`dA^I`$ are composite objects containing the differentials $`dx^\mu `$ and jet space variables (fields and their derivatives), $`A^I=dx^\mu A_\mu ^I`$, $`dC^I=dx^\mu _\mu C^I`$, $`dA^I=dx^\mu dx^\nu _\mu A_\nu ^I`$.
The universal algebra $`𝒜`$ is infinite dimensional, whereas $``$ is finite dimensional since the spacetime dimension bounds the form-degree of elements in $``$. By contrast, $`𝒜`$ contains elements with arbitrarily high degree in the $`(dC)_𝒜^I`$ and $`(dA)_𝒜^I`$.
The usefulness of $`𝒜`$ for the computations in the small algebra rests on the fact that $`𝒜`$ and $``$ are isomorphic at all form-degrees smaller than or equal to the spacetime dimension $`n`$. To make this statement precise, we first define a ”$`p`$-degree” which is the form-degree in $`𝒜`$,
$$p=A_𝒜^I\frac{}{A_𝒜^I}+(dC)_𝒜^I\frac{}{(dC)_𝒜^I}+2(dA)_𝒜^I\frac{}{(dA)_𝒜^I},$$
and clearly coincides with the form-degree in $``$ for the corresponding objects. The $`p`$-degree is indicated by a superscript. $`𝒜`$ and $``$ decompose into subspaces with definite $`p`$-degrees,
$$=\underset{p=0}{\overset{n}{}}^p,𝒜=\underset{p=0}{\overset{\mathrm{}}{}}𝒜^p.$$
We then introduce the natural mappings $`\pi ^p`$ from $`𝒜^p`$ to $`^p`$ which just replace each generator $`C_𝒜^I`$, $`A_𝒜^I`$, $`(dC)_𝒜^I`$, $`(dA)_𝒜^I`$ by its counterpart in $``$,
$$\pi ^p:\{\begin{array}{ccc}\hfill 𝒜^p& & ^p\hfill \\ \hfill a(C_𝒜,A_𝒜,(dC)_𝒜,(dA)_𝒜)& & a(C,A,dC,dA).\hfill \end{array}$$
(10.2)
These mappings are defined for all $`p`$, with $`^p0`$ for $`p>n`$. Evidently they are surjective (each element of $`^p`$ is in the image of $`\pi ^p`$). The point is that they are also injective for all $`pn`$. Indeed, the kernel of $`\pi ^p`$ is trivial for $`pn`$ by the following lemma:
###### Lemma 10.1
For all $`pn`$, the image of $`a^p𝒜^p`$ under $`\pi ^p`$ vanishes if and only if $`a^p`$ itself vanishes,
$$pn:\pi ^p(a^p)=0a^p=0.$$
(10.3)
Lemma 10.1 holds due to the fact that the components $`A_\mu ^I`$, $`_\mu C^I`$, $`_{[\mu }A_{\nu ]}^I`$ of $`A^I`$, $`dC^I`$, $`dA^I`$ are algebraically independent variables in the jet space (see section 4) and occur in elements of $``$ always together with the corresponding differential(s). For instance, consider $`a^p=k_{I_1\mathrm{}I_p}A_𝒜^{I_1}\mathrm{}A_𝒜^{I_p}`$ where the $`k_{I_1\mathrm{}I_p}`$ are constant coefficients. Without loss of generality the $`k_{I_1\mathrm{}I_p}`$ are antisymmetric since the $`A_𝒜^I`$ anticommute. Hence, one has $`a^p=p!_{I_i<I_{i+1}}k_{I_1\mathrm{}I_p}A_𝒜^{I_1}\mathrm{}A_𝒜^{I_p}`$ and $`\pi ^p(a^p)=p!_{I_i<I_{i+1}}k_{I_1\mathrm{}I_p}A^{I_1}\mathrm{}A^{I_p}`$. Vanishing of $`\pi ^p(a^p)`$ in the jet space requires in particular that the coefficients of $`(dx^1A_1^{I_1})\mathrm{}(dx^pA_p^{I_p})`$ vanish, for all sets $`\{I_1,\mathrm{},I_p:I_i<I_{i+1}\}`$. This requires vanishing of all coefficients $`k_{I_1\mathrm{}I_p}`$, and thus $`a^p=0`$. The general case is a straightforward extension of this example.
We can thus conclude:
###### Corollary 10.1
The mappings $`\pi ^p`$ are bijective for all $`pn`$ and establish thus isomorphisms between $`𝒜^p`$ and $`^p`$ for all $`pn`$.
In order to use these isomorphisms, $`𝒜`$ is equipped with differentials $`s_𝒜`$ and $`d_𝒜`$ which are the counterparts of $`s`$ and $`d`$. Accordingly, $`s_𝒜`$ and $`d_𝒜`$ are defined on the generators of $`𝒜`$ as follows:
$$\begin{array}{ccc}Z_𝒜& s_𝒜Z_𝒜& d_𝒜Z_𝒜\\ & & \\ \text{}C_𝒜^I& \frac{1}{2}ef_{JK}^{}{}_{}{}^{I}C_𝒜^KC_𝒜^J& (dC)_𝒜^I\\ \text{}A_𝒜^I& (dC)_𝒜^Ief_{JK}^{}{}_{}{}^{I}C_𝒜^JA_𝒜^K& (dA)_𝒜^I\\ \text{}(dC)_𝒜^I& ef_{JK}^{}{}_{}{}^{I}C_𝒜^J(dC)_𝒜^K& 0\\ \text{}(dA)_𝒜^I& ef_{JK}^{}{}_{}{}^{I}(C_𝒜^K(dA)_𝒜^JA_𝒜^J(dC)_𝒜^K)& 0\end{array}$$
(10.4)
The definition of $`s_𝒜`$ is extended to all polynomials in the generators by the graded Leibniz rule, $`s_𝒜(ab)=(s_𝒜a)b+()^{ϵ_a}a(s_𝒜b)`$, and the definition of $`d_𝒜`$ is analogously extended. With these definitions, $`s_𝒜`$ and $`d_𝒜`$ are anticommuting differentials in $`𝒜`$,
$$s_𝒜^2=d_𝒜^2=s_𝒜d_𝒜+d_𝒜s_𝒜=0.$$
One can therefore define $`H(d_𝒜,𝒜)`$ and $`H(s_𝒜,𝒜)`$ (and $`H(s_𝒜|d_𝒜,𝒜)`$ as well).
By construction one has
$$p:\pi ^ps_𝒜=s\pi ^p,\pi ^{p+1}d_𝒜=d\pi ^p(\text{on}𝒜^p).$$
(10.5)
This just means that $`\pi ^p`$ and $`\pi ^{p+1}`$ map $`s_𝒜a^p`$ and $`d_𝒜a^p`$ to $`s\pi ^p(a^p)`$ and $`d\pi ^p(a^p)`$ respectively, for every $`a^p𝒜^p`$ (note that $`s\pi ^p(a^p)`$ and $`d\pi ^p(a^p)`$ vanish for $`p>n`$ and $`pn`$ respectively).
Now, $`\pi ^p`$ can be inverted for $`pn`$ since it is bijective (see corollary 10.1). Hence, (10.5) gives
$$\begin{array}{ccc}\hfill pn:& s=\pi ^ps_𝒜(\pi ^p)^1\hfill & (\text{on}^p),\hfill \\ & d=\pi ^{p+1}d_𝒜(\pi ^p)^1\hfill & (\text{on}^p),\hfill \\ & s_𝒜=(\pi ^p)^1s\pi ^p\hfill & (\text{on}𝒜^p),\hfill \\ \hfill p<n:& d_𝒜=(\pi ^{p+1})^1d\pi ^p\hfill & (\text{on}𝒜^p)\hfill \end{array}$$
where the last relation holds only for $`p<n`$ (but not for $`p=n`$) since $`\pi ^{n+1}`$ has no inverse due to $`\pi ^{n+1}(𝒜^{n+1})=0`$. We conclude:
###### Corollary 10.2
$`H(s,^p)`$ is isomorphic to $`H(s_𝒜,𝒜^p)`$ for all $`pn`$, and $`H(d,^p)`$ is isomorphic to $`H(d_𝒜,𝒜^p)`$ for all $`p<n`$. At the level of the representatives $`[b^p]^p`$ and $`[a^p]𝒜^p`$ of these cohomologies, the isomorphisms are given by the mappings (10.2),
$`pn:`$ $`H(s,^p)H(s_𝒜,𝒜^p),[b^p]=\pi ^p([a^p]);`$
$`p<n:`$ $`H(d,^p)H(d_𝒜,𝒜^p),[b^p]=\pi ^p([a^p]).`$ (10.6)
In the following this corollary will be used to deduce the cohomologies of $`s`$ and $`d`$ in $``$ (except for $`H(d,^n)`$) from their counterparts in $`𝒜`$.
### 10.3 Cohomology of $`d`$ in the small algebra
It is very easy to compute $`H(d_𝒜,𝒜)`$ since all generators of $`𝒜`$ group in contractible pairs for $`d_𝒜`$ given by $`(C_𝒜^I,(dC)_𝒜^I)`$ and $`(A_𝒜^I,(dA)_𝒜^I)`$, see (10.4). One concludes by means of a contracting homotopy (cf. Section 2.7):
###### Lemma 10.2
$`H(d_𝒜,𝒜^p)`$ vanishes for all $`p>0`$ and $`H(d_𝒜,𝒜^0)`$ is represented by the constants (pure numbers),
$$H(d_𝒜,𝒜^p)=\delta _0^p.$$
(10.7)
By corollary 10.2, this implies
###### Corollary 10.3
$`H(d,^p)`$ vanishes for $`0<p<n`$ and $`H(d,^0)`$ is represented by the constants,
$$H(d,^p)=\delta _0^p\text{for}p<n.$$
(10.8)
Corollary 10.3 guarantees that we can apply theorems 9.1 and 9.2 of Section 9 to compute $`H(s|d,)`$ since (9.1) holds.
### 10.4 Cohomology of $`s_𝒜`$
The cohomology of $`s_𝒜`$ can be derived using the techniques of Section 8. One first gets rid of the exterior derivatives of the ghosts and of the gauge potentials by introducing a new basis of generators of $`𝒜`$, which are denoted by $`\{u^I,v^I,w^i\}`$ with
$`u^I=A_𝒜^I,v^I=(dC)_𝒜^Ief_{JK}^{}{}_{}{}^{I}C_𝒜^JA_𝒜^K,`$
$`\{w^i\}=\{C_𝒜^I,F_𝒜^I\},F_𝒜^I=(dA)_𝒜^I+{\displaystyle \frac{1}{2}}ef_{JK}^{}{}_{}{}^{I}A_𝒜^JA_𝒜^K.`$ (10.9)
Note that $`v^I`$ and $`F_𝒜^I`$ replace the former generators $`(dC)_𝒜^I`$ and $`(dA)_𝒜^I`$ respectively and that $`F_𝒜^I`$ corresponds of course to the field strength 2-forms $`F^I=\frac{1}{2}dx^\mu dx^\nu F_{\mu \nu }^I=\pi ^2(F_𝒜^I)`$. Note also that the change of basis preserves the polynomial structure: $`𝒜`$ is the space of polynomials in the new generators.
Using (10.4), one easily verifies that $`s_𝒜`$ acts on the new generators according to
$`s_𝒜u^I=v^I,s_𝒜v^I=0,`$
$`s_𝒜C_𝒜^I={\displaystyle \frac{1}{2}}ef_{JK}^{}{}_{}{}^{I}C_𝒜^KC_𝒜^J,s_𝒜F_𝒜^I=ef_{JK}^{}{}_{}{}^{I}C_𝒜^JF_𝒜^K.`$ (10.10)
The $`u^I`$ and $`v^I`$ form thus contractible pairs for $`s_𝒜`$ and drop therefore from $`H(s_𝒜,𝒜)`$. Hence, $`H(s_𝒜,𝒜)`$ reduces to $`H(s_𝒜,𝒜_w)`$ where $`𝒜_w`$ is the space of polynomials in the $`C_𝒜^I`$ and $`F_𝒜^I`$.
From (10.10) and our discussion in section 8, it follows that $`H(s_𝒜,𝒜_w)`$ is nothing but the Lie algebra cohomology of $`𝒢`$ in the representation space of the polynomials in the $`F_𝒜^I`$, tansforming under the extension of the coadjoint representation. As shown in section 8, it is generated by the ghost polynomials $`\theta _r(C_𝒜)`$ and by $`𝒢`$-invariant polynomials in the $`F_𝒜^I`$.
We can make the description of $`H(s_𝒜,𝒜)`$ completely precise here, because the space of invariant polynomials in the $`F_𝒜^I`$ is completely known. Indeed, the space of $`𝒢`$-invariant polynomials in the $`F_𝒜^I`$ is generated by a finite set of such polynomials given by
$$f_r(F_𝒜)=\mathrm{Tr}(F_𝒜^{m(r)}),F_𝒜=F_𝒜^IT_I,r=1,\mathrm{},R,R=\mathrm{𝑟𝑎𝑛𝑘}(𝒢)$$
(10.11)
where we follow the notations of subsection 8.5: $`r`$ labels the independent Casimir operators of $`𝒢`$, $`m(r)`$ is the order of the $`r`$th Casimir operator, and $`\{T_I\}`$ is the same matrix representation of $`𝒢`$ used for constructing $`\theta _r(C)`$ in Eq. (8.18). More precisely one has (see e.g. ):
(i) Every $`𝒢`$-invariant polynomial in the $`F_𝒜^I`$ is a polynomial $`P(f_1(F_𝒜),\mathrm{},f_R(F_𝒜))`$ in the $`f_r(F_𝒜)`$.
(ii) A polynomial $`P(f_1(F_𝒜),\mathrm{},f_R(F_𝒜))`$ in the $`f_r(F_𝒜)`$ vanishes as a function of the $`F_𝒜^I`$ if and only if $`P(f_1,\mathrm{},f_R)`$ vanishes as a function of commuting independent variables $`f_r`$ (i.e., in the free differential algebra of variables $`f_1,\mathrm{},f_R`$).
Note that (i) states the completeness of $`\{f_r(F_𝒜)\}`$ while (ii) states that the $`f_r(F_𝒜)`$ are algebraically independent. Analogous properties hold for the $`\theta _r(C_𝒜)`$ (they generate the space of $`𝒢`$-invariant polynomials in the anticommuting variables $`C_𝒜^I`$). In particular, the algebraic independence of the $`\theta _r(C_𝒜)`$ and $`f_r(F_𝒜)`$ implies that a polynomial in the $`\theta _r(C_𝒜)`$ and $`f_r(F_𝒜)`$ vanishes in $`𝒜`$ if and only if it vanishes already as a polynomial in the free differential algebra of anticommuting variables $`\theta _r`$ and commuting variables $`f_r`$. Summarizing, we have:
###### Lemma 10.3
$`H(s_𝒜,𝒜)`$ is freely generated by the $`\theta _r(C_𝒜)`$ and $`f_r(F_𝒜)`$:
(i) Every $`s_𝒜`$-closed element of $`𝒜`$ is a polynomial in the $`\theta _r(C_𝒜)`$ and $`f_r(F_𝒜)`$ up to an $`s_𝒜`$-exact element of $`𝒜`$,
$`s_𝒜a=0,a𝒜`$
$`a=P(\theta _1(C_𝒜),\mathrm{},\theta _R(C_𝒜),f_1(F_𝒜),\mathrm{},f_R(F_𝒜))+s_𝒜a^{},a^{}𝒜.`$ (10.12)
(ii) No nonvanishing polynomial $`P(\theta _1,\mathrm{},\theta _R,f_1,\mathrm{},f_R)`$ gives rise to an $`s_𝒜`$-exact polynomial in $`𝒜`$,
$`P(\theta _1(C_𝒜),\mathrm{},\theta _R(C_𝒜),f_1(F_𝒜),\mathrm{},f_R(F_𝒜))=s_𝒜a,a𝒜`$
$`P(\theta _1,\mathrm{},\theta _R,f_1,\mathrm{},f_R)=0.`$ (10.13)
By lemma 10.3, a basis of $`H(s_𝒜,𝒜)`$ is obtained from a basis of all polynomials in anticommuting variables $`\theta _r`$ and commuting variables $`f_r`$, $`r=1,\mathrm{},R`$. Such a basis is simply given by all monomials of the following form:
$$\theta _{r_1}\mathrm{}\theta _{r_K}f_{s_1}\mathrm{}f_{s_N}:K,N0,r_i<r_{i+1},s_is_{i+1}.$$
(10.14)
Here it is understood that $`\theta _{r_1}\mathrm{}\theta _{r_0}1`$ if $`K=0`$, and $`f_{s_1}\mathrm{}f_{s_0}1`$ if $`N=0`$. The requirements $`r_i<r_{i+1}`$ and $`s_is_{i+1}`$ take the commutation relations (Grassmann parities) of the variables into account.
### 10.5 Cohomology of $`s`$ in the small algebra
(10.14) induces a basis of $`H(s,)`$ thanks to corollary 10.2. However, this basis is not best suited for our ultimate goal, the determination of $`H(s|d,)`$, because it is not split into what we called $`h_{i_r}^0`$, $`e_{\alpha _s}^0`$ and $`\widehat{h}_{i_r}`$ in theorem 9.1. Therefore we will now construct a better suited basis, using (10.14) as a starting point. For this purpose we order the Casimirs according to their degrees in the $`F`$’s, namely, we assume that the Casimir labels $`r=1,\mathrm{},R`$ are such that, for any two such labels $`r`$ and $`r^{}`$,
$$r<r^{}m(r)m(r^{}).$$
(10.15)
Note that the ordering (10.15) is ambiguous if two or more Casimir operators have the same order. This ambiguity will not matter, i.e., any ordering that satisfies (10.15) is suitable for our purposes.
The set of all monomials (10.14) is now split into three subsets. The first subset is just given by $`\{1\}`$. The second subset contains all those monomials which have the property that the lowest appearing Casimir label is carried by a $`\theta `$. These monomials are denoted by $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)`$,
$`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)=\theta _{r_1}\mathrm{}\theta _{r_K}f_{s_1}\mathrm{}f_{s_N},`$
$`K1,N0,r_i<r_{i+1},s_is_{i+1},r_1=\mathrm{min}\{r_i,s_i\}.`$ (10.16)
Note that $`r_1=\mathrm{min}\{r_i,s_i\}`$ requires $`r_1s_1`$ if $`N0`$ while it does not impose an extra condition if $`N=0`$ (it is already implied by $`r_i<r_{i+1}`$ if $`N=0`$).
The third subset contains all remaining monomials. In these monomials at least one of the $`f`$’s has a lower Casimir label than all the $`\theta `$’s. Denoting the lowest occurring label again by $`r_1`$, the monomials of the third set can thus be written as
$`\overline{N}_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta ,f)=f_{r_1}\theta _{r_2}\mathrm{}\theta _{r_K}f_{s_1}\mathrm{}f_{s_N},`$
$`K1,N0,r_i<r_{i+1},s_is_{i+1},r_1=\mathrm{min}\{r_i,s_i\}.`$ (10.17)
Thus, for instance, in the case of $`G=U(1)\times SU(2)`$, there are two $`f`$’s and two $`\theta `$’s, namely, $`f_1=F^{u(1)}`$, $`f_2=\mathrm{Tr}_{\mathrm{su}(2)}F^2`$, $`\theta _1=C^{\mathrm{u}(1)}`$ and $`\theta _2=\mathrm{Tr}_{\mathrm{su}(2)}C^3`$. The monomial $`\theta _1f_2`$ belongs to the second subset, i.e., it is an “$`M`$”, because the label on $`\theta `$ is clearly smaller than that on $`f`$. The monomial $`\theta _2f_1`$, by contrast, belongs to the third set. We shall see that $`\theta _1f_2`$ is non trivial in $`H(s|d)`$ – and can be regarded as an $`h_{i_r}^0`$ –, while $`\theta _2f_1`$ is $`s`$-exact modulo $`d`$ and is a $`\widehat{h}_{i_r}`$ (it arises as an obstruction in the lift of $`\theta _1\theta _2`$).
We now define the following polynomials:
$$N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta ,f)=\underset{r:m(r)=m(r_1)}{}f_r\frac{M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)}{\theta _r}.$$
(10.18)
$`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}`$ is the sum of $`\overline{N}_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}`$ and a linear combination of $`M`$’s,
$$N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta ,f)=\overline{N}_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta ,f)\underset{\genfrac{}{}{0pt}{}{i:i2,}{m(r_i)=m(r_1)}}{}()^iM_{r_1\mathrm{}\widehat{r}_i\mathrm{}r_K|r_is_1\mathrm{}s_N}(\theta ,f)$$
where $`\widehat{r}_i`$ means omission of $`r_i`$. This shows that the set of all $`M`$’s and $`N`$’s, supplemented by the number 1, is a basis of polynomials in the $`\theta _r`$ and $`f_r`$ (as the same holds for the $`M`$’s and $`\overline{N}`$’s). Due to lemma 10.3 this provides also a basis of $`H(s_𝒜,𝒜)`$ after substituting the $`\theta _r(C_𝒜)`$ and $`f_r(F_𝒜)`$ for the $`\theta _r`$ and $`f_r`$:
###### Corollary 10.4
A basis of $`H(s_𝒜,𝒜)`$ is given by
$$\{B_\alpha \}=\{1,M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C_𝒜),f(F_𝒜)),N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C_𝒜),f(F_𝒜))\}.$$
(10.19)
Again, ”basis” is meant here in the cohomological sense: (i) every $`s_𝒜`$-closed element of $`𝒜`$ is a linear combination of the $`B_\alpha `$ up to an $`s_𝒜`$-exact element ($`s_𝒜a=0`$ $``$ $`a=\lambda ^\alpha B_\alpha +s_𝒜a^{}`$); (ii) no nonvanishing linear combination of the $`B_\alpha `$ is $`s_𝒜`$-exact ($`\lambda ^\alpha B_\alpha =s_𝒜a`$ $``$ $`\lambda ^\alpha =0\alpha `$).
Note that each $`B_\alpha `$ has a definite $`p`$-degree $`p_\alpha `$ which equals twice its degree in the $`F_𝒜^I`$. Hence, by corollary 10.2, those $`B_\alpha `$ with $`p_\alpha n`$ provide a basis of $`H(s,)`$. The requirement $`p_\alpha n`$ selects those $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}`$ with $`\mathrm{\Sigma }_{i=1}^N2m(s_i)n`$ and those $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}`$ with $`2m(r_1)+\mathrm{\Sigma }_{i=1}^N2m(s_i)n`$. We conclude:
###### Corollary 10.5
A basis of $`H(s,)`$ is given by $`\{1,M_a,N_i\}`$ where
$`\{M_a\}`$ $``$ $`\{M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F)):\mathrm{\Sigma }_{i=1}^N2m(s_i)n\},`$
$`\{N_i\}`$ $``$ $`\{N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F)):2m(r_1)+\mathrm{\Sigma }_{i=1}^N2m(s_i)n\}.`$ (10.20)
### 10.6 Transgression formulae
To derive $`H(s|d,)`$ from corollary 10.5, we need to construct the lifts of the elements $`M`$ of the previous basis. The most expedient way to achieve this task is to use the celebrated “transgression formula” (also called Russian formula)
$$(s+d)q_r(C+A,F)=\mathrm{Tr}\left(F^{m(r)}\right)=f_r(F),$$
(10.21)
where $`C=C^IT_I`$, $`A=A^IT_I`$, $`F=F^IT_I`$, and
$$q_r(C+A,F)=m(r)_0^1𝑑t\mathrm{Tr}\left((C+A)\left[tF+e(t^2t)(C+A)^2\right]^{m(r)1}\right).$$
(10.22)
Here $`t`$ is just an integration variable and should not be confused with the spacetime coordinate $`x^0`$. A derivation of Eqs. (10.21) and (10.22) is given at the end of this subsection. $`q_r(C+A,F)`$ is nothing but the Chern-Simons polynomial $`q_r(A,F)`$ with $`C+A`$ substituting for $`A`$. It fulfills $`dq_r(A,F)=f_r(F)`$, as can be seen from (10.21) in ghost number $`0`$.
The usefulness of (10.21) for the determination of $`H(s|d,)`$ rests on the fact that it relates $`\theta _r(C)`$ and $`f_r(F)`$ via a set of equations obtained by decomposing (10.21) into parts with definite form-degrees,
$`d[\theta _r]^{2m(r)1}=f_r(F),`$ (10.23)
$`s[\theta _r]^p+d[\theta _r]^{p1}=0`$ $`\text{for}0<p<2m(r),`$
$`s[\theta _r]^0=0`$
where $`[\theta _r]^p`$ is the $`p`$-form contained in $`q_r(C+A,F)`$,
$$q_r(C+A,F)=\underset{p=0}{\overset{2m(r)1}{}}[\theta _r]^p.$$
(10.24)
The 0-form contained in $`q_r(C+A,F)`$ is nothing but $`\theta _r(C)`$,
$`[\theta _r]^0`$ $`=`$ $`m(r)\mathrm{Tr}(C^{\mathrm{\hspace{0.17em}2}m(r)1})e^{m(r)1}{\displaystyle _0^1}𝑑t(t^2t)^{m(r)1}`$ (10.25)
$`=`$ $`(e)^{m(r)1}{\displaystyle \frac{m(r)!(m(r)1)!}{(2m(r)1)!}}\mathrm{Tr}(C^{\mathrm{\hspace{0.17em}2}m(r)1})=\theta _r(C).`$
Note that $`f_r(F)`$ and some of the $`[\theta _r]^p`$ vanish in sufficiently low spacetime dimension (when $`n<2m(r)`$) but that $`q_r(C+A,F)`$ never vanishes completely since it contains $`\theta _r(C)`$. The same formulae hold in the universal algebra $`𝒜`$, but there, of course, none of the $`f_r`$ vanishes.
Consider now the polynomials $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(q(C+A,F),f(F))`$ arising from the $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)`$ in Eq. (10.16) by substituting the $`q_r(C+A,F)`$ and $`f_r(F)`$ for the corresponding $`\theta _r`$ and $`f_r`$. Analogously to (10.24), these polynomials decompose into pieces of various form-degrees,
$`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(q(C+A,F),f(F))={\displaystyle \underset{p=\underset{¯}{s}}{\overset{\overline{p}}{}}}[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p,`$
$`\underset{¯}{s}=\mathrm{\Sigma }_{i=1}^K2m(s_i),\overline{p}=\underset{¯}{s}+\mathrm{\Sigma }_{i=1}^K(2m(r_i)1)`$ (10.26)
where some or all $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ may vanish in sufficiently low spacetime dimension. Due to (10.25), one has
$$[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}}=M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F)).$$
(10.27)
The polynomials (10.26) give rise to transgression equations that generalize Eqs. (10.23). These equations are obtained by evaluating $`(s+d)M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(q(C+A,F),f(F))`$: one gets a sum of terms in which one of the $`q_r(C+A,F)`$ in $`M`$ is replaced by the corresponding $`f_r(F)`$ as a consequence of (10.21) (note that one has $`(s+d)f_r(F)=(s+d)^2q_r(C+A,F)=0`$ due to $`(s+d)^2=0`$, i.e., $`(s+d)`$ acts nontrivially only on the $`q`$’s contained in $`M`$). Hence, $`(s+d)M`$ is obtained by applying the operation $`_rf_r/q_r`$ to $`M`$. This makes it easy to identify the part of lowest form-degree contained in the resulting expression: it is $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))`$ given in Eq. (10.18) thanks to the ordering (10.15) of the Casimir labels. One thus gets generalized transgression equations
$`s[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r_1)}+d[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r_1)1}=N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F)),`$
$`s[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+q}+d[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+q1}=0\text{for}0<q<2m(r_1),`$
$`s[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}}=0.`$ (10.28)
Note that (10.23) is just a special case of (10.28), arising for $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)\theta _r`$.
##### Derivation of Eqs. (10.21) and (10.22).
The derivation is performed in the free differential algebra $`𝒜`$.
As shown in subsection 10.3, the cohomology of $`d_𝒜`$ is trivial in the algebra $`𝒜`$ of polynomials in $`A_𝒜^I,(dA)_𝒜^I,C_𝒜^I,(dC)_𝒜^I`$. The contracting homotopy is explicitly given by $`\rho =A_𝒜^I\frac{}{(dA)_𝒜^I}+C_𝒜^I\frac{}{(dC)_𝒜^I}`$. For a $`d_𝒜`$-cocycle $`f(A_𝒜^I,(dA)_𝒜^I,C_𝒜^I,(dC)_𝒜^I)`$ with $`f(0,0,0,0)=0`$, we get
$$f(A_𝒜^I,(dA)_𝒜^I,C_𝒜^I,(dC)_𝒜^I)=d_𝒜_0^1\frac{dt}{t}[\rho f](tA_𝒜^I,t(dA)_𝒜^I,tC_𝒜^I,t(dC)_𝒜^I).$$
(10.29)
We then consider the change of generators $`A_𝒜^I,(dA)_𝒜^IA_𝒜^I,F_𝒜^I`$ in $`𝒜`$ (while $`C_𝒜^I,(dC)_𝒜^I`$ remain unchanged). The differential $`d_𝒜`$ acts on $`A_𝒜^I`$ and $`F_𝒜^I`$ according to
$`d_𝒜A_𝒜^I`$ $`=`$ $`F_𝒜^I{\displaystyle \frac{1}{2}}ef_{JK}^{}{}_{}{}^{I}A_𝒜^JA_𝒜^K,`$
$`d_𝒜F_𝒜^I`$ $`=`$ $`ef_{JK}^{}{}_{}{}^{I}A_𝒜^JF_𝒜^K.`$ (10.30)
For a $`d_𝒜`$-cocycle $`f(A_𝒜^I,F_𝒜^I)`$ with $`f(0,0)=0`$, the homotopy formula (10.29) gives
$$f(A_𝒜^I,F_𝒜^I)=d_𝒜\left[A_𝒜^L_0^1𝑑t[\frac{f(A_𝒜^I,F_𝒜^I)}{F_𝒜^L}](tA_𝒜^I,tF_𝒜^I+\frac{1}{2}e(t^2t)f_{JK}^{}{}_{}{}^{I}A_𝒜^JA_𝒜^K)\right].$$
(10.31)
This formula generalizes straightforwardly to an analogous one for the differential
$`\stackrel{~}{s}_𝒜=s_𝒜+d_𝒜.`$ (10.32)
Indeed, with $`\stackrel{~}{C}_𝒜^I=C_𝒜^I+A_𝒜^I`$, one gets
$`\stackrel{~}{s}_𝒜\stackrel{~}{C}_𝒜^I`$ $`=`$ $`{\displaystyle \frac{1}{2}}ef_{JK}^{}{}_{}{}^{I}\stackrel{~}{C}_𝒜^K\stackrel{~}{C}_𝒜^J+F_𝒜^I,`$
$`\stackrel{~}{s}_𝒜F_𝒜^I`$ $`=`$ $`ef_{JK}^{}{}_{}{}^{I}\stackrel{~}{C}_𝒜^JF_𝒜^K.`$ (10.33)
Comparing (10.30) and (10.33), we see that the differential algebras $`(d_𝒜,\mathrm{\Lambda }(A_𝒜^I,F_𝒜^I))`$ and $`(\stackrel{~}{s}_𝒜,\mathrm{\Lambda }(\stackrel{~}{C}_𝒜^I,F_𝒜^I))`$ are isomorphic. It follows that
$$f(\stackrel{~}{C}_𝒜^I,F_𝒜^I)=\stackrel{~}{s}_𝒜\left[\stackrel{~}{C}_𝒜^L_0^1𝑑t[\frac{f(\stackrel{~}{C}_𝒜^I,F_𝒜^I)}{F_𝒜^L}](t\stackrel{~}{C}_𝒜^I,tF_𝒜^I+\frac{1}{2}e(t^2t)f_{JK}^{}{}_{}{}^{I}\stackrel{~}{C}_𝒜^J\stackrel{~}{C}_𝒜^K)\right],$$
(10.34)
for an $`\stackrel{~}{s}_𝒜`$-cocycle $`f(\stackrel{~}{C}_𝒜^I,F_𝒜^I)`$ with $`f(0,0)=0`$. Taking $`f=\mathrm{Tr}F_𝒜^{m(r)}`$, this yields (10.21) and (10.22) through the mappings (10.2).
### 10.7 $`H(s|d)`$ in the small algebra
We are now in the position to determine $`H(s|d,)`$. Namely, Eqs. (10.28) imply that the basis of $`H(s,)`$ given in corollary 10.5 has the properties described in theorems 9.1 and 9.2 of section 9. To verify this, consider Eqs. (10.28) first in the cases $`\underset{¯}{s}+2m(r_1)n`$. In these cases, Eqs. (10.28) reproduce Eqs. (9.10) in theorem 9.1, with the identifications
$`\underset{¯}{s}+2m(r_1)n:`$
$`\widehat{h}_{i_r}`$ $``$ $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))`$
$`h_{i_r}^q`$ $``$ $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+q},q=0,\mathrm{},r`$
$`r`$ $`=`$ $`2m(r_1)1.`$
In particular this yields $`h_{i_r}^0M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F))`$ for $`\underset{¯}{s}+2m(r_1)n`$ by Eq. (10.27).
Next consider Eqs. (10.28) in the cases $`n2m(r_1)<\underset{¯}{s}n`$. This reproduces Eqs. (9.11) in theorem 9.1, with the identifications
$`n2m(r_1)<\underset{¯}{s}n:`$
$`e_{\alpha _s}^q`$ $``$ $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+q},q=0,\mathrm{},s`$
$`s`$ $`=`$ $`n\underset{¯}{s}.`$
In particular this yields $`e_{\alpha _s}^0M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F))`$ for $`n2m(r_1)<\underset{¯}{s}n`$.
Hence, the basis of $`H(s,)`$ given in corollary 10.5 has indeed the properties described in theorem 9.1. By theorem 9.2, a basis of $`H(s|d,)`$ is thus given by the $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ specified in the above equations. The whole set of these representatives can be described more compactly through $`p=\underset{¯}{s},\mathrm{},\overline{m}`$ where $`\overline{m}=\mathrm{min}\{\underset{¯}{s}+2m(r_1)1,n\}`$.
We can summarize the result in the form of a receipe. Given the gauge group $`G`$ and the spacetime dimension $`n`$, one obtains $`H(s|d,)`$ as follows:
1. Specify the independent Casimir operators of $`G`$ and label them such that
$$r<r^{}m(r)m(r^{})$$
(10.35)
where $`m(r)`$ is the order of the $`r`$th Casimir operator.
2. Specify the following monomials:
$`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta ,f)=\theta _{r_1}\mathrm{}\theta _{r_K}f_{s_1}\mathrm{}f_{s_N}:`$
$`K1,N0,r_i<r_{i+1},s_is_{i+1},`$
$`r_1=\mathrm{min}\{r_i,s_i\},{\displaystyle \underset{i=1}{\overset{N}{}}}2m(s_i)n.`$ (10.36)
3. Replace in (10.36) the $`\theta _r`$ and $`f_r`$ by the corresponding $`q_r(C+A,F)`$ and $`f_r(F)`$ given in (10.22) and (10.21) and decompose the resulting polynomials in the $`q_r(C+A,F)`$ and $`f_r(F)`$ into pieces of definite form-degree,
$$M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(q(C+A,F),f(F))=\underset{p}{}[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p.$$
(10.37)
4. A basis of $`H(s|d,)`$ is then given by the number 1 and the following $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$:
$`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p:`$ $`p=\underset{¯}{s},\mathrm{},\overline{m},`$ (10.38)
$`\underset{¯}{s}=\mathrm{\Sigma }_{i=1}^N2m(s_i),`$
$`\overline{m}=\mathrm{min}\{\underset{¯}{s}+2m(r_1)1,n\}.`$
A similar results hold in the universal algebra $`𝒜`$, but in this case, there is no $`e_{\alpha _s}^0`$ but only $`h_{i_r}^0`$: all lifts are obstructed at some point. This implies, in particular, that any solution of the consistency condition in $`𝒜`$ can be seen as coming from an obstruction living above. For instance, the Adler-Bardeen-Bell-Jackiw anomaly in four dimensions, which is a four-form, comes from the six-form $`\mathrm{Tr}F_𝒜^3`$ through the Russian formula. This makes sense only in the universal algebra, although the anomaly itself is meaningful both in $`𝒜`$ and $``$.
### 10.8 $`H^{0,n}(s|d)`$ and $`H^{1,n}(s|d)`$ in the small algebra
Physically important representatives of $`H(s|d,)`$ are those with form-degree $`n`$ and ghost numbers 0 or 1 as they provide possible counterterms and gauge anomalies respectively. To extract these representatives from Eqs. (10.35) through (10.38), one uses that $`q_r(C+A,F)`$ and $`f_r(F)`$ have total degree (= form-degree + ghost number) $`2m(r)1`$ and $`2m(r)`$ respectively. The total degree of $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ is thus $`_{i=1}^K(2m(r_i)1)+_{i=1}^N2m(s_i)=\underset{¯}{s}+_{i=1}^K(2m(r_i)1)`$. Hence, representatives with form-degree $`n`$ and ghost number $`g`$ fulfill
$$n+g=\underset{¯}{s}+\underset{i=1}{\overset{K}{}}(2m(r_i)1).$$
Furthermore, representatives with form-degree $`n`$ fulfill
$$n\underset{¯}{s}+2m(r_1)1$$
because of the requirement $`\overline{m}=\mathrm{min}\{\underset{¯}{s}+2m(r_1)1,n\}`$ in Eq. (10.38). Combining these two conditions, one gets
$$\underset{i=2}{\overset{K}{}}(2m(r_i)1)g.$$
(10.39)
Note that here the sum runs from $`2`$ to $`K`$, and that we have $`K1`$ by (10.36). Hence, for $`g=0`$, (10.39) selects the value $`K=1`$. The representatives of $`H^{0,n}(s|d,)`$ arise thus from (10.37) by setting $`K=1`$ and selecting the ghost number 0 part. These representatives are
$$[M_{r|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r)1}=[\theta _r]^{2m(r)1}f_{s_1}(F)\mathrm{}f_{s_N}(F).$$
(10.40)
Note that (10.36) imposes $`rs_1s_2\mathrm{}s_N`$ if $`N0`$. $`[\theta _r]^{2m(r)1}`$ is nothing but the Chern-Simons form corresponding to $`f_r(F)`$, see equation (10.23). (10.40) is thus a Chern-Simons form too, corresponding to $`f_r(F)f_{s_1}(F)\mathrm{}f_{s_N}(F)`$. All representatives (10.40) have odd form-degree and occur thus only in odd spacetime dimensions.
For $`g=1`$, (10.39) leaves two possibilities: $`K=1`$, or $`K=2`$ where the latter case requires in addition $`m(r_2)=1`$. For $`K=1`$, this yields the following representatives of $`H^{1,n}(s|d,)`$,
$$[M_{r|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r)2}=[\theta _r]^{2m(r)2}f_{s_1}(F)\mathrm{}f_{s_N}(F).$$
(10.41)
Again, (10.36) imposes $`rs_1s_2\mathrm{}s_N`$ if $`N0`$. By (10.22) one has
$$[\theta _r]^{2m(r)2}=\mathrm{Tr}(CF^{m(r)1}+\mathrm{}).$$
All representatives (10.41) have even form-degree. They represent the consistent chiral gauge anomalies in even spacetime dimensions.
The remaining representatives of $`H^{1,n}(s|d,)`$ have $`K=2`$ and $`m(r_2)=1`$. $`m(r_2)=1`$ requires $`m(r_1)=1`$ by (10.35) and (10.36). The Casimir operators of order 1 are the abelian generators (see Section 8.5). The corresponding $`q_r(C+A,F)`$ coincide with the abelian $`C+A`$,
$$\{q_r(C+A,F):m(r)=1\}=\{\text{abelian}C^I+A^I\}.$$
(10.42)
The representatives of $`H^{1,n}(s|d,)`$ with $`K=2`$ read
$$[M_{IJ|s_1\mathrm{}s_N}]^{\underset{¯}{s}+1}=(C^IA^JC^JA^I)f_{s_1}(F)\mathrm{}f_{s_N}(F)(\text{abelian}I,J).$$
(10.43)
(10.36) imposes $`I<J`$, $`s_is_{i+1}`$ and that $`f_{s_1}(F)`$ is not an abelian $`F^K`$ with $`K<I`$. Note that the representatives (10.43) have odd form-degree and are only present if the gauge group contains at least two abelian factors. They yield candidate gauge anomalies in odd spacetime dimensions.
### 10.9 Examples
To illustrate the results, we shall now spell out $`H^{0,n}(s|d,)`$ and $`H^{1,n}(s|d,)`$ for specific choices of $`n`$ and $`G`$. We list those $`\theta _r(C)`$ (up to the normalization factor) and $`f_r(F)`$ needed to construct $`H^{0,n}(s|d,)`$ and $`H^{1,n}(s|d,)`$ and give a complete set of the inequivalent representatives (“Reps.”) of these cohomological groups and the corresponding obstructions (“Obs.”) in the universal algebra $`𝒜`$ (except that in the last example we leave it to the reader to spell out the obstructions as it is similar to the second example). The inclusion of $`SO(1,9)`$ and $`SO(1,10)`$ in the last two examples is relevant in the gravitational context because the Lorentz group plays a rôle similar to the Yang-Mills gauge group when one includes gravity in the analysis.
$`n=4`$, $`G=U(1)\times SU(2)\times SU(3)`$
$$\begin{array}{ccccc}r& 1& 2& 3& 4\\ & & & & \\ \text{}m(r)& 1& 2& 2& 3\\ \text{}\theta _r(C)& C^{\mathrm{u}(1)}& \mathrm{Tr}_{\mathrm{su}(2)}C^3& \mathrm{Tr}_{\mathrm{su}(3)}C^3& \mathrm{Tr}_{\mathrm{su}(3)}C^5\\ \text{}f_r(F)& F^{\mathrm{u}(1)}& \mathrm{Tr}_{\mathrm{su}(2)}F^2& \mathrm{Tr}_{\mathrm{su}(3)}F^2& 0\end{array}$$
$`H^{0,4}(s|d,):`$ empty
$`H^{1,4}(s|d,):`$ $`\begin{array}{ccccc}\mathrm{Reps}.& C^{\mathrm{u}(1)}(F^{\mathrm{u}(1)})^2& C^{\mathrm{u}(1)}f_2(F)& C^{\mathrm{u}(1)}f_3(F)& [\theta _3]^4\\ & & \multicolumn{-1}{c}{}& & \\ \text{}\mathrm{Obs}.& (F_𝒜^{\mathrm{u}(1)})^3& F_𝒜^{\mathrm{u}(1)}f_2(F_𝒜)& F_𝒜^{\mathrm{u}(1)}f_3(F_𝒜)& f_4(F_𝒜)\end{array}`$
where $`[\theta _3]^4=\mathrm{Tr}_{\mathrm{su}(3)}[Cd(AdA+\frac{1}{2}e_{\mathrm{su}(3)}A^3)],f_4(F_𝒜)=\mathrm{Tr}_{\mathrm{su}(3)}(F_𝒜)^3`$
$`n=10`$, $`G=SO(32)`$
$$\begin{array}{cccc}r& 1& 2& 3\\ & & & \\ \text{}m(r)& 2& 4& 6\\ \text{}\theta _r(C)& \mathrm{Tr}C^3& \mathrm{Tr}C^7& \mathrm{Tr}C^{11}\\ \text{}f_r(F)& \mathrm{Tr}F^2& \mathrm{Tr}F^4& 0\end{array}$$
$`H^{0,10}(s|d,):`$ empty
$`H^{1,10}(s|d,):`$ $`\begin{array}{cccc}\mathrm{Reps}.& [\theta _1]^2(f_1(F))^2& [\theta _1]^2f_2(F)& [\theta _3]^{10}\\ & & \multicolumn{-1}{c}{}& \\ \text{}\mathrm{Obs}.& (f_1(F_𝒜))^3& f_1(F_𝒜)f_2(F_𝒜)& f_3(F_𝒜)\end{array}`$
where $`[\theta _1]^2=\mathrm{Tr}(CdA),[\theta _3]^{10}=\mathrm{Tr}[C(dA)^5+\mathrm{}],`$
$`f_3(F_𝒜)=\mathrm{Tr}(F_𝒜)^6`$
$`n=11`$, $`G=SO(1,10)`$
$$\begin{array}{cccc}r& 1& 2& 3\\ & & & \\ \text{}m(r)& 2& 4& 6\\ \text{}\theta _r(C)& \mathrm{Tr}C^3& \mathrm{Tr}C^7& \mathrm{Tr}C^{11}\\ \text{}f_r(F)& \mathrm{Tr}F^2& \mathrm{Tr}F^4& 0\end{array}$$
$`H^{0,11}(s|d,):`$ $`\begin{array}{cccc}\mathrm{Reps}.& [\theta _1]^3(f_1(F))^2& [\theta _1]^3f_2(F)& [\theta _3]^{11}\\ & & \multicolumn{-1}{c}{}& \\ \text{}\mathrm{Obs}.& (f_1(F_𝒜))^3& f_1(F_𝒜)f_2(F_𝒜)& f_3(F_𝒜)\end{array}`$
where $`[\theta _1]^3=\mathrm{Tr}(AdA+\frac{1}{3}eA^3),[\theta _3]^{11}=\mathrm{Tr}[A(dA)^5+\mathrm{}],`$
$`f_3(F_𝒜)=\mathrm{Tr}(F_𝒜)^6`$
$`H^{1,11}(s|d,):`$ empty
$`n=10`$, $`G=SO(1,9)\times SO(32)`$
$$\begin{array}{ccccccc}r& 1& 2& 3& 4& 6& 7\\ & & & & & & \\ \text{}m(r)& 2& 2& 4& 4& 6& 6\\ \text{}\theta _r(C)& \mathrm{Tr}_{\mathrm{so}(1,9)}C^3& \mathrm{Tr}_{\mathrm{so}(32)}C^3& \mathrm{Tr}_{\mathrm{so}(1,9)}C^7& \mathrm{Tr}_{\mathrm{so}(32)}C^7& \mathrm{Tr}_{\mathrm{so}(1,9)}C^{11}& \mathrm{Tr}_{\mathrm{so}(32)}C^{11}\\ \text{}f_r(F)& \mathrm{Tr}_{\mathrm{so}(1,9)}F^2& \mathrm{Tr}_{\mathrm{so}(32)}F^2& \mathrm{Tr}_{\mathrm{so}(1,9)}F^4& \mathrm{Tr}_{\mathrm{so}(32)}F^4& 0& 0\end{array}$$
$`H^{0,10}(s|d,):`$ empty
$`H^{1,10}(s|d,)(\text{Reps.}):`$ $`[\theta _1]^2(f_1(F))^2,[\theta _1]^2f_1(F)f_2(F),[\theta _1]^2(f_2(F))^2,`$
$`[\theta _1]^2f_3(F),[\theta _1]^2f_4(F),`$
$`[\theta _2]^2(f_2(F))^2,[\theta _2]^2f_3(F),[\theta _2]^2f_4(F),`$
$`[\theta _6]^{10},[\theta _7]^{10}`$
Remark: the Pfaffian of $`SO(1,9)`$ yields $`f_5(F)`$ with $`m(5)=5`$; however it does not contribute to $`H^{1,10}(s|d,)`$ in this case (it would contribute through $`C^{\mathrm{u}(1)}f_5(F)`$ if $`G`$ contained in addition a $`U(1)`$).
## 11 General solution of the consistency condition in Yang-Mills type theories
### 11.1 Assumptions
We shall now put the pieces together and determine $`H(s|d,\mathrm{\Omega })`$ completely in Yang-Mills type theories for two cases:
Case I: $`\mathrm{\Omega }=\{\text{all local forms}\},`$
Case II: $`\mathrm{\Omega }=\{\text{Poincaré-invariant local forms}\},`$ (11.1)
on the following assumptions:
(a) the gauge group does not contain abelian gauge symmetries under which all matter fields are uncharged – we shall call such special abelian symmetries “free abelian gauge symmetries” in the following;
(b) the spacetime dimension (denoted by $`n`$) is larger than 2;
(c) the theory is normal and the regularity conditions hold;
(d) in case II it is assumed that the Lagrangian itself is Poincaré-invariant (in case I it need not be Poincaré-invariant).
Assumption (c) is a technical one and has been explained in Sections 5 and 6. Assumptions (a) and (b) reflect special properties of free abelian gauge symmetries and 2-dimensional (pure) Yang-Mills theory which complicate somewhat the general analysis. These special properties of free abelian gauge symmetries are illustrated and dealt with in Section 13 where we compute the cohomology for a set of free abelian gauge fields. Two-dimensional pure Yang-Mills theory is treated separately in the appendix to this Section. Note that assumption (a) does not exclude abelian gauge symmetries. It excludes only the presence of free abelian gauge fields, or of abelian gauge fields that couple exclusively non-minimally to matter or gauge fields (i.e., through the field strengths and their derivatives only).
The space of all local forms is the direct product $`𝒫\mathrm{\Omega }(^n)`$ where $`𝒫`$ is the space of local functions of the fields, antifields, and all their derivatives, while $`\mathrm{\Omega }(^n)`$ is the space of ordinary differential forms $`\omega (x,dx)`$ in $`^n`$. Depending on the context and Lagrangian, $`𝒫`$ can be, for instance, the space of polynomials in the fields, antifields, and all their derivatives (when the Lagrangian is polynomial too), or it can be the space of local forms that depend polynomially on the derivatives of the fields and antifields but may depend smoothly on (some of) the undifferentiated fields (when the Lagrangian has the same property). The latter would be the case for instance for Yang-Mills theory coupled to a dilaton. It can also be the space of formal power series in the fields, antifields, and all their derivatives, with coefficients that depend on the coupling constants (in the case of effective theories). More generally speaking, the results and their derivation apply whenever the various cohomological results derived and discussed in the previous sections (especially those on $`H(s)`$ and $`H_{\mathrm{char}}(d)`$) hold, since these will be used within the computation below.
The space of Poincaré-invariant local forms is a subspace of $`𝒫\mathrm{\Omega }(^n)`$. It contains only those local forms which do not depend explicitly on the spacetime coordinates $`x^\mu `$ and are Lorentz-invariant. Lorentz-invariance requires here simply that all Lorentz indices (including the indices of derivatives and differentials, and the spinor indices of spacetime fermions) are contracted in an $`SO(1,n1)`$-invariant manner.
Two central ingredients in our computation of $`H(s|d)`$ are the results on the characteristic cohomology of $`d`$ in Section 6 and on the cohomology of $`s`$ in Section 8. We shall repeat them here, for case I and case II, and reformulate the result on $`H(s)`$ since we shall frequently use it in that formulation. For the characteristic cohomology of $`d`$ we have:
###### Corollary 11.1
$`H_{\mathrm{char}}^p(d,\mathrm{\Omega })`$ vanishes at all form-degrees $`0<p<n1`$ and is at form-degree 0 represented by the constants,
$`0<p<n1:`$ $`d\omega ^p0,\omega ^p\mathrm{\Omega }\omega ^pd\omega ^{p1},\omega ^{p1}\mathrm{\Omega };`$
$`p=0:`$ $`d\omega ^00,\omega ^0\mathrm{\Omega }\omega ^0\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}.`$ (11.2)
In case I this follows directly from the results of Section 6 thanks to assumptions (a), (b) and (c), where assumptions (a) and (b) are only needed for the vanishing of $`H_{\mathrm{char}}^{n2}(d,\mathrm{\Omega })`$ ($`H_{\mathrm{char}}^{n2}(d,\mathrm{\Omega })`$ does not vanish when free abelian gauge symmetries are present, and is not exhausted by the constants in 2-dimensional pure Yang-Mills theory; this makes these two cases special). The analysis and results of Section 6 extend to case II because both $`d`$ and $`\delta `$ are Lorentz invariant according to the definition of Lorentz invariance used here (i.e., $`d`$ and $`\delta `$ commute in $`\mathrm{\Omega }`$ with $`SO(1,n1)`$-rotations of all Lorentz indices). This is obvious for $`d`$ and holds for $`\delta `$ thanks to assumption (d) because that assumption guarantees the Lorentz-covariance of the equations of motion.
The results on $`H(s)`$ in section 8 can be reformulated as follows:
###### Corollary 11.2
Description of $`H(s,\mathrm{\Omega })`$:
$`s\omega =0,\omega \mathrm{\Omega }\omega =I^\alpha \mathrm{\Theta }_\alpha +s\eta ,I^\alpha ,\eta \mathrm{\Omega };`$ (11.3)
$`I^\alpha \mathrm{\Theta }_\alpha =s\omega ,I^\alpha ,\omega \mathrm{\Omega }I^\alpha 0\alpha .`$ (11.4)
Here, $`\{\mathrm{\Theta }_\alpha \}`$ is a basis of all polynomials in the $`\theta _r(C)`$,
$$\{\mathrm{\Theta }_\alpha \}=\{1,\underset{\genfrac{}{}{0pt}{}{i=1,\mathrm{},K}{r_i<r_{i+1}}}{}\theta _{r_i}(C):K=1,\mathrm{},\mathrm{𝑟𝑎𝑛𝑘}(𝒢)\},$$
(11.5)
and $``$ is the antifield independent gauge invariant subspace of $`\mathrm{\Omega }`$ given in the two cases under study respectively by
Case I: $`=\{𝒢\text{-invariant local functions of }F_{\mu \nu }^I\text{}\psi ^i\text{}D_\rho F_{\mu \nu }^I\text{}D_\rho \psi ^i\text{, …}\}\mathrm{\Omega }(^n)`$
Case II: $`=\{𝒢\text{-invariant and Lorentz-invariant local functions of}`$ (11.6)
$`dx^\mu \text{}F_{\mu \nu }^I\text{}\psi ^i\text{}D_\rho F_{\mu \nu }^I\text{}D_\rho \psi ^i\text{, …}\}.`$
Again, the precise definition of “local functions” depends on the context and Lagrangian.
Let us now briefly explain how and why corollary 11.2 reformulates the results in Section 8. We first treat case I. The results in Section 8 give (using a notation as in that section) $`H(s,\mathrm{\Omega })=\mathrm{\Omega }(^n)V_{\rho =0}^X\mathrm{\Lambda }(C)_{\rho ^C=0}`$ where $`V_{\rho =0}^X`$ is the space of $`𝒢`$-invariant local functions of the $`X_A^u`$ specified in Section 8.3, and $`\mathrm{\Lambda }(C)_{\rho ^C=0}`$ is the space of polynomials in the $`\theta _r(C)`$. Now, one has $`\mathrm{\Omega }(^n)V_{\rho =0}^X`$ and thus $`\mathrm{\Omega }(^n)V_{\rho =0}^X\mathrm{\Lambda }(C)_{\rho ^C=0}\mathrm{\Lambda }(C)_{\rho ^C=0}`$. This yields the implication $``$ in (11.3). The implication $``$ holds because all elements of $`\mathrm{\Lambda }(C)_{\rho ^C=0}`$ are $`s`$-closed. Furthermore, since $`\mathrm{\Omega }(^n)V_{\rho =0}^X`$ is only a subspace of $``$, one has $`=(\mathrm{\Omega }(^n)V_{\rho =0}^X)(\mathrm{\Omega }(^n)V_{\rho =0}^X)^{}`$. By the results in Section 8, all nonvanishing elements of $`\mathrm{\Omega }(^n)V_{\rho =0}^X\mathrm{\Lambda }(C)_{\rho ^C=0}`$ are nontrivial in $`H(s,\mathrm{\Omega })`$ whereas all elements of $`(\mathrm{\Omega }(^n)V_{\rho =0}^X)^{}\mathrm{\Lambda }(C)_{\rho ^C=0}`$ are trivial in $`H(s,\mathrm{\Omega })`$ and vanish on-shell. This gives (11.4).
We now turn to case II. Thanks to assumption (d), both $`\delta `$ and $`\gamma `$ commute with Lorentz transformations. Therefore the results of Section 8 hold analogously in the Lorentz-invariant subspace of $`𝒫\mathrm{\Omega }(^n)`$. This gives in case II $`H(s,\mathrm{\Omega })=V_{\rho =0}^{X,dx}\mathrm{\Lambda }(C)_{\rho ^C=0}`$ where $`V_{\rho =0}^{X,dx}`$ is the space of Lorentz-invariant and $`𝒢`$-invariant local functions of the $`X_A^u`$ and $`dx^\mu `$. Corollary 11.2 holds now by reasons analogous to case I.
Finally we shall often use the following immediate consequence of the isomorphism (7.5) (cf. proof of that isomorphism):
###### Corollary 11.3
An $`s`$-cocycle with nonnegative ghost number is $`s`$-exact whenever its antifield independent part vanishes on-shell,
$$s\omega =0,\omega \mathrm{\Omega },\mathrm{𝑔ℎ}(\omega )0,\omega _00\omega =sK,K\mathrm{\Omega }$$
where $`\omega _0`$ is the antifield independent part of $`\omega `$.
### 11.2 Outline of the derivation and result
We shall determine the general solution of the consistency condition
$$s\omega ^p+d\omega ^{p1}=0,\omega ^p,\omega ^{p1}\mathrm{\Omega }$$
(11.7)
for all values of the form-degree $`p`$ and of the ghost number (the ghost number will not be made explicit throughout this section, contrary to the form-degree). Since the precise formulation and derivation of the result are involved, we shall first outline the crucial steps of the computation and describe the various nontrivial solutions. The precise formulation of the result and its proof will be given in the following Section 11.3.
The computation relies on the descent equation technique described in Section 9, which can be used because $`d`$ has trivial cohomology at all form-degrees different from 0 and $`n`$,
$$H^p(d,\mathrm{\Omega })=\delta _0^p\text{for}p<n.$$
(11.8)
This holds in the space of all local forms (case I) by the algebraic Poincaré lemma (theorem 4.2). It also holds in the space of Poincaré invariant local forms because $`d`$ is Lorentz-invariant (cf. text after corollary 11.1). In fact, one may even deduce this directly from the proof of the algebraic Poincaré lemma given in Section 4 because the operators $`\rho `$ and $`P_m`$ used there are manifestly Lorentz-invariant, and because there are no Lorentz-invariant constant forms $`c_{\mu _1\mathrm{}\mu _p}dx^{\mu _1}\mathrm{}dx^{\mu _p}`$ with form degree $`0<p<n`$. In fact, that proof of the algebraic Poincaré lemma is not just an existence proof but provides an explicit construction of $`\omega ^{p1}`$ for given $`d`$-closed $`\omega ^p`$ such that $`\omega ^p=d\omega ^{p1}`$, both in case I and case II.
One distinguishes between solutions with a trivial descent and solutions with a nontrivial descent.
#### 11.2.1 Solutions with a trivial descent.
These are solutions to (11.7) that can be redefined by the addition of trivial solutions such that they solve $`s\omega ^p=0`$. The result on $`H(s)`$ (corollary 11.2) implies then that these solutions have the form
$$\omega ^p=I^{p\alpha }\mathrm{\Theta }_\alpha ,I^{p\alpha }$$
(11.9)
modulo trivial solutions. We note that (11.9) can be trivial in $`H(s|d)`$ even if $`I^{p\alpha }0`$.
#### 11.2.2 Lifts and equivariant characteristic cohomology
The solutions with a non trivial descent are those for which it is impossible to make $`\omega ^{p1}`$ vanish in (11.7) through the addition of trivial solutions. Their determination is more difficult. In particular it calls for the solution of a cohomological problem that we have not discussed so far. Namely, one has to determine the characteristic cohomology in the space $``$ defined in (11.6). This cohomology is well defined because $`d`$. Indeed, for $`I`$, one has $`dI=DI`$, where $`D=dx^\mu D_\mu `$, with $`D_\mu `$ the covariant derivative on the fields and $`D_\mu x^\nu =\delta _\mu ^\nu `$. Furthermore $`DI`$ is $`𝒢`$-invariant. We call this cohomology the “equivariant characteristic cohomology” and denote it by $`H_{\mathrm{char}}(d,)`$. It is related to, but different from the ordinary characteristic cohomology discussed in Section 6.
To understand the difference, assume that $`I`$ is weakly $`d`$-exact in $`\mathrm{\Omega }`$, i.e., $`Id\omega `$ for some $`\omega \mathrm{\Omega }`$. The equivariant characteristic cohomology poses the following question: is it possible to choose $`\omega `$ ? We shall answer this question in the affirmative with the exception when $`I`$ contains a “characteristic class”. Abusing slightly standard terminology, a characteristic class is in this context a $`𝒢`$-invariant polynomial in the curvature 2-forms $`F^I`$ and is thus, in particular, an element of the small algebra.<sup>11</sup><sup>11</sup>11It is a semantic coincidence that the word “characteristic” is used in the literature both in the context of the polynomials $`P(F)`$ and to term the weak cohomology of $`d`$. The invariant cohomology of $`d`$ without antifields and use of the equations of motion has been investigated in . The fact that it contains only the characteristic classes has been called the “covariant Poincaré lemma” in . Furthermore, we shall show that no characteristic class with form-degree $`<n`$ is trivial in $`H_{\mathrm{char}}(d,)`$ (at form-degree $`n`$ there may be exceptions).
Hence, $`H_{\mathrm{char}}(d,)`$ is the sum of a subspace of $`H_{\mathrm{char}}(d,\mathrm{\Omega })`$ (given by $`H_{\mathrm{char}}(d,\mathrm{\Omega })`$) and of the space of characteristic classes. Using corollary 11.1, one thus gets that $`H_{\mathrm{char}}(d,)`$ is at all form-degrees $`<n1`$ solely represented by characteristic classes (at form-degree $`n2`$ this is due to assumptions (a) and (b)). In contrast, there are in general additional nontrivial representatives of $`H_{\mathrm{char}}(d,)`$ at form-degrees $`n`$ and $`n1`$.<sup>12</sup><sup>12</sup>12An exception is 3-dimensional pure Chern-Simons theory (with semisimple gauge group) where $`H_{\mathrm{char}}(d,)`$ vanishes even in form-degrees $`n=3`$ and $`n1=2`$, see Section 14. At form-degree $`n`$ they are present because an $`n`$-form is automatically $`d`$-closed but not necessarily weakly $`d`$-exact. The additional representatives at form-degree $`n1`$ are gauge-invariant nontrivial Noether currents written as $`(n1)`$-forms.
The result on $`H_{\mathrm{char}}(d,)`$ is interesting in itself and a cornerstone of the local BRST cohomology in Yang-Mills type theories. The technical assumption of “normality” assumed throughout the calculation is made in order to be able to characterize completely $`H_{\mathrm{char}}(d,)`$. The properties of $`H_{\mathrm{char}}(d,)`$ are at the origin of the importance of the small algebra for the cohomology, as we now explain.
The equivariant characteristic cohomology arises as follows when discussing the descent equations. One has
$$s\omega ^p+d\omega ^{p1}=0,s\omega ^{p1}+d\omega ^{p2}=0,\mathrm{},s\omega ^{\underset{¯}{m}}=0.$$
Without loss of generality, one can assume that the bottom form $`\omega ^{\underset{¯}{m}}`$ is a nontrivial solution of the consistency condition. It is thus a solution with a trivial descent and can be taken of the form
$$\omega ^{\underset{¯}{m}}=I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha ,I^{\underset{¯}{m}\alpha }.$$
(11.10)
In the case of a nontrivial descent, $`\omega ^{\underset{¯}{m}}`$ satisfies additionally
$$s\omega ^{\underset{¯}{m}+1}+d\omega ^{\underset{¯}{m}}=0$$
(11.11)
which is the last but one descent equation. It turns out that this equation is a very restrictive condition on the bottom $`\omega ^{\underset{¯}{m}}`$. Few bottoms can be lifted at least once. In order to be “liftable”, all $`I^{\underset{¯}{m}\alpha }`$ must be representatives of the equivariant characteristic cohomology.
Indeed, one has
$$d(I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha )=(dI^{\underset{¯}{m}\alpha })\mathrm{\Theta }_\alpha s(I^{\underset{¯}{m}\alpha }[\mathrm{\Theta }_\alpha ]^1)$$
(11.12)
where we used $`sI^{\underset{¯}{m}\alpha }=0`$ (which holds due to $`I^{\underset{¯}{m}\alpha }`$) and
$$d\mathrm{\Theta }_\alpha +s[\mathrm{\Theta }_\alpha ]^1=0,[\mathrm{\Theta }_\alpha ]^1=A^I\frac{\mathrm{\Theta }_\alpha }{C^I}.$$
(11.13)
(11.13) is nothing but the equation with $`q=1`$ contained in Eqs. (10.28), for the particular case $`M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}M_{r_1\mathrm{}r_K}`$.
Now, (11.10) through (11.12) imply $`(dI^{\underset{¯}{m}\alpha })\mathrm{\Theta }_\alpha =s(\mathrm{})`$. By corollary 11.2 this implies that all $`dI^{\underset{¯}{m}\alpha }`$ vanish weakly,
$$\alpha :dI^{\underset{¯}{m}\alpha }0.$$
(11.14)
Furthermore, if $`I^{\underset{¯}{m}\alpha }dI^{\underset{¯}{m}1}`$ for some $`I^{\underset{¯}{m}1}`$ and some $`\alpha `$, the piece $`I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha `$ (no sum over $`\alpha `$ here) can be removed by subtracting a trivial term from $`\omega ^{\underset{¯}{m}}`$. Indeed, $`I^{\underset{¯}{m}\alpha }dI^{\underset{¯}{m}1}`$ implies that $`I^{\underset{¯}{m}\alpha }=dI^{\underset{¯}{m}1}+sK^{\underset{¯}{m}}`$ for some $`K^{\underset{¯}{m}}`$ by corollary 11.3 and thus that $`I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha `$ is trivial, $`I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha =d(I^{\underset{¯}{m}1}\mathrm{\Theta }_\alpha )+s(I^{\underset{¯}{m}1}[\mathrm{\Theta }_\alpha ]^1+K^{\underset{¯}{m}}\mathrm{\Theta }_\alpha )`$ (no sum over $`\alpha `$). Hence, without loss of generality one can assume that
$$\alpha :I^{\underset{¯}{m}\alpha }dI^{\underset{¯}{m}1\alpha },I^{\underset{¯}{m}1\alpha }.$$
(11.15)
By Eqs. (11.14) and (11.15), every $`I^{\underset{¯}{m}\alpha }`$ is a nontrivial representative of the equivariant characteristic cohomology. This cohomology qualifies thus to some extent the bottom forms that appear in nontrivial descents: all those $`I^{\underset{¯}{m}\alpha }`$ with $`\underset{¯}{m}<n1`$ can be assumed to be characteristic classes $`P(F)`$, while those with $`\underset{¯}{m}=n1`$ can additionally contain nontrivial gauge-invariant Noether currents (in the special case $`n=2`$ or when free gauge symmetries are present, there can be bottom forms of yet another type with form-degree $`n2`$).
#### 11.2.3 Solutions with a nontrivial descent
Of course, the previous discussion gives not yet a complete characterization of the bottom forms because $`I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha `$ may be trivial in $`H(s|d,\mathrm{\Omega })`$ even when all $`I^{\underset{¯}{m}\alpha }`$ represent nontrivial classes of the equivariant characteristic cohomology (the nontriviality of $`I^{\underset{¯}{m}\alpha }`$ in $`H_{\mathrm{char}}(d,)`$ is necessary but not sufficient for the nontriviality of $`I^{\underset{¯}{m}\alpha }\mathrm{\Theta }_\alpha `$). Furthermore one still has to investigate how far the nontrivial bottom forms can be maximally lifted (so far we have only discussed lifting bottom forms once). Nevertheless the above discussion gives already an idea of the result. Namely, one finds ultimately that the consistency condition has at most three types of solutions with a nontrivial descent:
1. Solutions which lie in the small algebra $``$. These solutions are linear combinations of those $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ in (10.38) with $`_{i=1}^N2m(s_i)<p`$ (the solutions with $`_{i=1}^N2m(s_i)=p`$ are $`s`$-closed and have thus a trivial descent). Here and in the following, such linear combinations are denoted by $`B^p`$,
$`B^p={\displaystyle \underset{\underset{¯}{s}=p2m(r_1)+1}{\overset{p1}{}}}\lambda ^{r_1\mathrm{}r_K|s_1\mathrm{}s_N}[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p,\underset{¯}{s}={\displaystyle \underset{i=1}{\overset{N}{}}}2m(s_i)`$ (11.16)
where the $`\lambda ^{r_1\mathrm{}r_K|s_1\mathrm{}s_N}`$ are constant coefficients. We shall prove that no nonvanishing $`B^p`$ is trivial in $`H(s|d,\mathrm{\Omega })`$. In other words: $`B^p`$ is trivial only if all coefficients $`\lambda ^{r_1\mathrm{}r_K|s_1\mathrm{}s_N}`$ vanish (because the $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ are linearly independent, see Section 10). The solutions $`B^p`$ descend to bottom forms involving characteristic classes $`P(F)`$.
2. Antifield dependent solutions which involve nontrivial global symmetries corresponding to gauge invariant nontrivial Noether currents. These solutions cannot be given explicitly in a model independent manner because the set of global symmetries is model dependent. To describe them we introduce the notation $`\{j_\mathrm{\Delta }^\mu \}`$ for a basis of those Noether currents which can be brought to a form such that the corresponding $`(n1)`$-forms are elements of $``$ (possibly by the addition of trivial currents, see Section 6),
$$j_\mathrm{\Delta }=\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}j_\mathrm{\Delta }^{\mu _n},dj_\mathrm{\Delta }0.$$
(11.17)
“Basis” means here that (a) every Noether current which has a representative in $``$ is a linear combination of the $`j_\mathrm{\Delta }^\mu `$ up to a current which is trivial in $`\mathrm{\Omega }`$, and (b) no nonvanishing linear combination of the $`j_\mathrm{\Delta }`$ is trivial in $`H_{\mathrm{char}}(d,\mathrm{\Omega })`$,
$`dI^{n1}0,I^{n1}I^{n1}\lambda ^\mathrm{\Delta }j_\mathrm{\Delta }+d\omega ^{n2};`$ (11.18)
$`\lambda ^\mathrm{\Delta }j_\mathrm{\Delta }d\omega ^{n2}\lambda ^\mathrm{\Delta }=0\mathrm{\Delta }.`$ (11.19)
Note that, in general, $`\{j_\mathrm{\Delta }\}`$ does not represent a complete basis of $`H_{\mathrm{char}}^{n1}(d,)`$ because our definition does not use the coboundary condition in $`H_{\mathrm{char}}^{n1}(d,)`$ ($`\{j_\mathrm{\Delta }\}`$ does not contain characteristic classes $`P(F)`$ while $`H_{\mathrm{char}}^{n1}(d,)`$ may contain characteristic classes when $`n`$ is odd). Note also that, in general, $`\{j_\mathrm{\Delta }^\mu \}`$ differs in case I and II. For instance, the various components $`T_{0}^{}{}_{}{}^{\mu }`$, …, $`T_{n1}^{}{}_{}{}^{\mu }`$ of the energy momentum tensor $`T_{\nu }^{}{}_{}{}^{\mu }`$ can normally be redefined such that they are gauge invariant<sup>13</sup><sup>13</sup>13An example where $`T_{0}^{}{}_{}{}^{\mu }`$ cannot be made gauge invariant is example 3 in Section 12.1.3. and provide then $`n`$ elements of $`\{j_\mathrm{\Delta }\}`$ in case I; however, they are not contravariant Lorentz-vectors and therefore they do not provide elements of $`\{j_\mathrm{\Delta }\}`$ in case II. Similarly, in globally supersymmetric Yang-Mills models, the supersymmetry currents provide normally elements of $`\{j_\mathrm{\Delta }\}`$ in case I, but not in case II.
Since $`j_\mathrm{\Delta }`$ is gauge invariant, one has $`dj_\mathrm{\Delta }=Dj_\mathrm{\Delta }`$ and therefore $`dj_\mathrm{\Delta }`$ is gauge invariant and $`s`$-invariant too. By corollary 11.3, $`dj_\mathrm{\Delta }0`$ implies thus the existence of a volume form $`K_\mathrm{\Delta }`$ such that
$$sK_\mathrm{\Delta }+dj_\mathrm{\Delta }=0.$$
(11.20)
$`K_\mathrm{\Delta }`$ encodes the global symmetry corresponding to $`j_\mathrm{\Delta }`$ (see Section 6 and also Section 7 for the connection between $`s`$ and $`\delta `$). We now define the $`n`$-forms
$$V_{\mathrm{\Delta }\alpha }=K_\mathrm{\Delta }\mathrm{\Theta }_\alpha +j_\mathrm{\Delta }[\mathrm{\Theta }_\alpha ]^1$$
(11.21)
with $`[\mathrm{\Theta }_\alpha ]^1`$ as in (11.13). These forms solve the consistency condition. Indeed, Eqs. (11.13) and (11.20) give immediately
$$sV_{\mathrm{\Delta }\alpha }+d[j_\mathrm{\Delta }\mathrm{\Theta }_\alpha ]=0.$$
(11.22)
This also shows that $`j_\mathrm{\Delta }\mathrm{\Theta }_\alpha `$ is a bottom form corresponding to $`V_{\mathrm{\Delta }\alpha }`$ (as one has $`s(j_\mathrm{\Delta }\mathrm{\Theta }_\alpha )=0`$ due to $`j_\mathrm{\Delta }`$).
3. In special cases (i.e., for special Lagrangians), there are additional nontrivial antifield dependent solutions. They emerge from nontrivial conserved currents which can not be made gauge invariant by the addition of trivial currents. These “accidental” solutions complicate the derivation of the general solution of the consistency condition but, even though exceptional, they must be covered since we do not make restrictions on the Lagrangian besides the technical ones explained above.
We shall prove that, if $`n>2`$ and free abelian gauge symmetries are absent, such currents exist if and only if characteristic classes with maximal form-degree $`n`$ are trivial in $`H_{\mathrm{char}}(d,)`$ (examples are given in Section 12). Assume that $`\{P_A(F)\}`$ is a basis for characteristic classes of this type, i.e., assume that every characteristic class with form-degree $`n`$ which is trivial in $`H_{\mathrm{char}}(d,)`$ is a linear combination of the $`P_A(F)`$ and the $`P_A(F)`$ are linearly independent,
$`P_A(F)dI_A^{n1},I_A^{n1};`$ (11.23)
$`P^n(F)dI^{n1},I^{n1}P^n=\lambda ^AP_A(F);`$ (11.24)
$`\lambda ^AP_A(F)=0\lambda ^A=0A.`$ (11.25)
Note that $`P_A(F)dI_A^{n1}`$ implies $`d(I_A^{n1}q_A^{n1})0`$ where $`q_A^{n1}`$ is a Chern-Simons $`(n1)`$-form fulfilling $`P_A(F)=dq_A^{n1}`$. The $`I_A^{n1}q_A^{n1}`$ are thus conserved $`(n1)`$-forms; they are the afore-mentioned Noether currents that cannot be made gauge invariant. $`P_A(F)dI_A^{n1}`$ is $`s`$-invariant (as it is in $``$) and vanishes weakly; hence, corollary 11.3 guarantees the existence of a volume form $`K_A`$ such that
$$P_A(F)=dI_A^{n1}+sK_A.$$
(11.26)
This gives $`sK_A+d(I_A^{n1}q_A^{n1})=0`$, i.e. $`K_A`$ is a cocycle of $`H^{1,n}(s|d)`$ (it contains the global symmetry corresponding to the Noether current $`I_A^{n1}q_A^{n1}`$).
Every linear combination of $`n`$-forms $`P_A(F)\mathrm{\Theta }_\alpha `$ belongs to the cohomology of $`s`$ in the small algebra and can therefore be expanded in the $`N_i`$ and $`M_a`$ in (10.20). We now consider only those linear combinations which can be written solely in terms of $`N_i`$ (which must have form degree $`n`$). A basis for these linear combinations is denoted by $`\{N_\mathrm{\Gamma }\}`$:
$`N_\mathrm{\Gamma }=k_\mathrm{\Gamma }^iN_i=k_\mathrm{\Gamma }^{A\alpha }P_A(F)\mathrm{\Theta }_\alpha ;`$ (11.27)
$`\lambda ^iN_i=\lambda ^{A\alpha }P_A(F)\mathrm{\Theta }_\alpha \lambda ^iN_i=\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma };`$ (11.28)
$`\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }=0\lambda ^\mathrm{\Gamma }=0\mathrm{\Gamma }.`$ (11.29)
In particular one can choose the basis $`\{N_\mathrm{\Gamma }\}`$ such that it contains $`\{P_A(F)\}`$ (since $`\{N_i\}`$ contains a basis of all $`P(F)`$).
Now, on the one hand, one gets
$`N_\mathrm{\Gamma }`$ $`=`$ $`k_\mathrm{\Gamma }^{A\alpha }(dI_A^{n1}+sK_A)\mathrm{\Theta }_\alpha `$ (11.30)
$`=`$ $`k_\mathrm{\Gamma }^{A\alpha }\left[d(I_A^{n1}\mathrm{\Theta }_\alpha )+s(I_A^{n1}[\mathrm{\Theta }_\alpha ]^1+K_A\mathrm{\Theta }_\alpha )\right]`$
where we used (11.26) and (11.13). On the other hand, $`N_\mathrm{\Gamma }`$ is a linear combination of those $`N_i`$ with form-degree $`n`$ and thus of the form
$$N_\mathrm{\Gamma }=sb_\mathrm{\Gamma }^n+dB_\mathrm{\Gamma }^{n1},b_\mathrm{\Gamma }^n,B_\mathrm{\Gamma }^{n1}$$
(11.31)
by the first equation in (10.28) \[$`b_\mathrm{\Gamma }^n`$ is a linear combination of the $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^n`$ with $`n=2m(r_1)+_{i=1}^N2m(s_i)`$ and is therefore not of the form (11.16), in contrast to $`B_\mathrm{\Gamma }^{n1}`$\]. Subtracting (11.30) from (11.31), one gets
$$sW_\mathrm{\Gamma }+d[B_\mathrm{\Gamma }^{n1}k_\mathrm{\Gamma }^{A\alpha }I_A^{n1}\mathrm{\Theta }_\alpha ]=0$$
(11.32)
where
$$W_\mathrm{\Gamma }=b_\mathrm{\Gamma }^nk_\mathrm{\Gamma }^{A\alpha }(K_A\mathrm{\Theta }_\alpha +I_A^{n1}[\mathrm{\Theta }_\alpha ]^1).$$
(11.33)
$`W_\mathrm{\Gamma }`$ descends to a bottom form in the small algebra which involves characteristic classes $`P(F)`$, namely to the same bottom form to which $`B_\mathrm{\Gamma }^{n1}`$ descends. From the point of view of the descent equations this means the following. In the small algebra, the bottom-form corresponding to $`B_\mathrm{\Gamma }^{n1}`$ can only be lifted to form-degree $`(n1)`$; there is no way to lift it to an $`n`$-form in the small algebra because this lift is obstructed by $`N_\mathrm{\Gamma }`$. However, there is no such obstruction in the full algebra because the characteristic classes contained in $`N_\mathrm{\Gamma }`$ are trivial in $`H_{\mathrm{char}}^n(d,)`$ ($`W_\mathrm{\Gamma }`$ is not entirely in the small algebra: in particular it contains antifields through the $`K_A`$).
### 11.3 Main result and its proof
According to the discussion in Section 11.2 it may appear natural to determine first $`H_{\mathrm{char}}(d,)`$ and afterwards $`H(s|d,\mathrm{\Omega })`$. In fact that strategy was followed in previous computations . However, it is more efficient to determine $`H_{\mathrm{char}}(d,)`$ and $`H(s|d,\mathrm{\Omega })`$ at a stroke. The reason is that these two cohomologies are strongly interweaved. In fact, not only does one need $`H_{\mathrm{char}}(d,)`$ to compute $`H(s|d,\mathrm{\Omega })`$; but also, $`H_{\mathrm{char}}(d,)`$ at form-degree $`p`$ can be computed by means of $`H(s|d,\mathrm{\Omega })`$ at lower former degrees using descent equation techniques. That makes it possible to determine both cohomologies simultaneously in a recursive manner, starting at form-degree 0 where $`H(s|d,\mathrm{\Omega })`$ reduces to $`H(s,\mathrm{\Omega })`$, and then proceeding successively to higher form-degrees. This strategy streamlines the derivation as compared to previously used approaches (but reaches of course identical conclusions!), and is reflected in the formulation and proof of the theorem given below. The theorem is formulated such that it applies both to case I and to case II; however, the different meaning of $`\mathrm{\Omega }`$, $``$ (and thus also $`j_\mathrm{\Delta }`$ and $`V_{\mathrm{\Delta }\alpha }`$) in these cases should be kept in mind.
To formulate and prove the theorem, we use the same notation as in Section 11.2. In addition we introduce, similarly to (11.16), the notation $`M^p`$, $`N^p`$ and $`b^p`$ for linear combinations of the $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}}`$, $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))`$, and $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r_1)}`$ with form-degree $`p`$ respectively,
$`M^p`$ $``$ $`{\displaystyle \underset{\underset{¯}{s}=p}{}}\lambda ^{r_1\mathrm{}r_K|s_1\mathrm{}s_N}M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F)),`$ (11.34)
$`N^p`$ $``$ $`{\displaystyle \underset{\underset{¯}{s}+2m(r_1)=p}{}}\lambda ^{r_1\mathrm{}r_Ks_1\mathrm{}s_N}N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))`$ (11.35)
$`b^p`$ $``$ $`{\displaystyle \underset{\underset{¯}{s}+2m(r_1)=p}{}}\lambda ^{r_1\mathrm{}r_K|s_1\mathrm{}s_N}[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^{\underset{¯}{s}+2m(r_1)}`$ (11.36)
where we used once again the notation $`\underset{¯}{s}=_{i=1}^N2m(s_i)`$. Note that we have, for every $`B^p`$ as in (11.16),
$`sB^p=d(B^{p1}+M^{p1}),dB^p=s(B^{p+1}+b^{p+1})+N^{p+1}`$ (11.37)
for some $`B^{p1}`$, $`M^{p1}`$, $`B^{p+1}`$, $`b^{p+1}`$ and $`N^{p+1}`$ by Eqs. (10.28).
We can now formulate the result as follows.
###### Theorem 11.1
Let $`\omega ^p\mathrm{\Omega }`$ with $`\mathrm{\Omega }`$ as in (11.1), $`I^p,I^{p\alpha }`$ with $``$ as in (11.6). Let $`P^p(F)`$ denote characteristic classes ($`p`$ indicating the form-degree respectively), and $``$ denoting equivalence in $`H(s|d,\mathrm{\Omega })`$ (i.e. $`\omega ^{}{}_{}{}^{p}\omega ^p`$ means $`\omega ^{}{}_{}{}^{p}=\omega ^p+s\eta ^p+d\eta ^{p1}`$ for some $`\eta ^p,\eta ^{p1}\mathrm{\Omega }`$). For Yang-Mills type theories without free abelian gauge symmetries, the following statements hold in all spacetime dimensions $`n>2`$:
(i) At all form-degrees $`p<n`$, the general solution of the consistency condition is given, up to trivial solutions, by the sum of a term $`I^\alpha \mathrm{\Theta }_\alpha `$ and a solution in the small algebra as in Eq. (11.16); at form-degree $`p=n`$ it contains in addition a linear combination of the $`V_{\mathrm{\Delta }\alpha }`$ and $`W_\mathrm{\Gamma }`$ given in (11.21) and (11.33) respectively,
$$s\omega ^p+d\omega ^{p1}=0\omega ^pI^{p\alpha }\mathrm{\Theta }_\alpha +B^p+\delta _n^p(\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma }).$$
(11.38)
(ii) $`I^{p\alpha }\mathrm{\Theta }_\alpha +B^p+\delta _n^p(\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma })`$ is trivial in $`H(s|d,\mathrm{\Omega })`$ if and only if $`B^p`$ and all coefficients $`\lambda ^{\mathrm{\Delta }\alpha }`$, $`\lambda ^\mathrm{\Gamma }`$ vanish and $`I^{p\alpha }\mathrm{\Theta }_\alpha `$ is weakly equal to $`N^p+(dI^{p1\alpha })\mathrm{\Theta }_\alpha `$ for some $`N^p`$ and $`I^{p1\alpha }`$,<sup>14</sup><sup>14</sup>14Note that this requires $`I^{p\alpha }dI^{p1\alpha }+P^{p\alpha }(F)`$ with $`P^{p\alpha }(F)`$ such that $`P^{p\alpha }(F)\mathrm{\Theta }_\alpha =N^p`$.
$`I^{p\alpha }\mathrm{\Theta }_\alpha +B^p+\delta _n^p(\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma })0`$ (11.39)
$``$ $`B^p=0,\lambda ^{\mathrm{\Delta }\alpha }=0(\mathrm{\Delta },\alpha ),\lambda ^\mathrm{\Gamma }=0\mathrm{\Gamma },I^{p\alpha }\mathrm{\Theta }_\alpha N^p+(dI^{p1\alpha })\mathrm{\Theta }_\alpha .`$
(iii) If $`I^p`$ ($`p>0`$) is trivial in $`H_{\mathrm{char}}^p(d,\mathrm{\Omega })`$ then it is the sum of a characteristic class and a piece which is trivial in $`H_{\mathrm{char}}^p(d,)`$,
$$p>0:I^pd\omega ^{p1}I^pP^p(F)+dI^{p1}.$$
(11.40)
(iv) No nonvanishing characteristic class with form-degree $`p<n`$ is trivial in $`H_{\mathrm{char}}(d,)`$,
$$p<n:P^p(F)dI^{p1}P^p(F)=0.$$
(11.41)
Proof.
Step 1. To prove the theorem, we first verify that (i), (ii) and (iv) hold at form-degree 0. There are no $`B^0`$ or $`N^0`$ since the ranges of values for $`\underset{¯}{s}`$ in the sums in Eqs. (11.16) and (11.35) are empty for $`p=0`$. Hence, for $`p=0`$, (i) and (ii) reduce to $`s\omega ^0=0\omega ^0=I^{0\alpha }\mathrm{\Theta }_\alpha +s\eta ^0`$ and $`I^{0\alpha }\mathrm{\Theta }_\alpha =s\eta ^0I^{0\alpha }0`$ respectively and hold by the results on $`H(s)`$ (corollary 11.2). (iv) reduces for $`p=0`$ to $`\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}0\mathrm{𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡}=0`$ which holds for every meaningful Lagrangian (if it would not hold then the equations of motion were inconsistent, see Section 9).
Step 2. In the second (and final) step we show that (i) through (iv) hold for $`p=m`$ if they hold for $`p=m1`$, excluding $`m=n`$ in the case (iv).
1. By corollary 11.3, $`P^m(F)dI^{m1}`$ implies $`P^m(F)=dI^{m1}+sK^m`$ for some local $`K^m`$. On the other hand one has $`P^m(F)=dq^{m1}`$ for some Chern-Simons $`(m1)`$-form $`q^{m1}`$ which we choose to be the $`B^{m1}`$ corresponding to $`P^m(F)`$.<sup>15</sup><sup>15</sup>15Note that there can be an ambiguity in the choice of Chern-Simons forms. Indeed, consider $`P^m(F)=f_1(F)f_2(F)`$. One has $`P^m(F)=d[\alpha q_1(A,F)f_2(F)+(1\alpha )q_2(A,F)f_1(F)]`$ where $`\alpha `$ is an arbitrary number. Our prescription in Section 10 selects $`B^{m1}=q_1(A,F)f_2(F)`$ ($`\alpha =1`$). This extends to all characteristic classes $`P^m(F)`$: our prescription selects precisely one $`B^{m1}`$ among all Chern-Simons forms corresponding to $`P^m(F)`$. This gives $`sK^m+d(I^{m1}q^{m1})=0`$, i.e., $`K^m`$ is a cocycle of $`H^{1,m}(s|d,\mathrm{\Omega })`$. One has $`H^{1,m}(s|d,\mathrm{\Omega })H_1^m(\delta |d,\mathrm{\Omega })H_0^{m1}(d|\delta ,\mathrm{\Omega })=H_{\mathrm{char}}^{m1}(d,\mathrm{\Omega })`$ for $`m>1`$ and analogously $`H^{1,1}(s|d,\mathrm{\Omega })H_{\mathrm{char}}^0(d,\mathrm{\Omega })/`$ by theorems 7.1 and 6.2. By corollary 11.1 this gives $`H^{1,m}(s|d,\mathrm{\Omega })=0`$ (since we are assuming $`0<m<n`$) and thus $`K^m0`$. This implies $`I^{m1}q^{m1}0`$ by the standard properties of the descent equations<sup>16</sup><sup>16</sup>16If one of the forms in the descent equations is trivial, then all its descendants are trivial too, see Section 9.. Now, since we assume that (ii) holds for $`p=m1`$, we conclude from $`I^{m1}q^{m1}0`$ in particular that $`q^{m1}=0`$ (since $`q^{m1}`$ is a $`B^{m1}`$) and thus that $`P^m(F)=dq^{m1}=0`$ which is (iv) for $`p=m<n`$.
2. $`s\omega ^m+d\omega ^{m1}=0`$ implies descent equations (Section 9). In particular there is some $`\omega ^{m2}`$ such that $`s\omega ^{m1}+d\omega ^{m2}=0`$. Since we assume that (i) holds for $`p=m1`$, we conclude
$$\omega ^{m1}=I^{m1\alpha }\mathrm{\Theta }_\alpha +B^{m1}$$
(11.42)
for some $`I^{m1\alpha }`$ and $`B^{m1}`$ (without loss of generality, since trivial contributions to any form in the descent equations can be neglected, see Section 9). By (11.37) we have
$$dB^{m1}=s(\widehat{B}^m+b^m)+N^m$$
(11.43)
for some $`\widehat{B}^m`$, $`b^m`$, $`N^m`$. Using in addition (11.13), we get
$$d\omega ^{m1}=(dI^{m1\alpha })\mathrm{\Theta }_\alpha s(I^{m1\alpha }[\mathrm{\Theta }_\alpha ]^1+\widehat{B}^m+b^m)+N^m.$$
Inserting this in $`s\omega ^m+d\omega ^{m1}=0`$, we obtain
$$s(\omega ^mI^{m1\alpha }[\mathrm{\Theta }_\alpha ]^1\widehat{B}^mb^m)+[dI^{m1\alpha }+P^{m\alpha }(F)]\mathrm{\Theta }_\alpha =0$$
(11.44)
where we used that
$$N^m=P^{m\alpha }(F)\mathrm{\Theta }_\alpha $$
(11.45)
for some $`P^{m\alpha }(F)`$. Using corollary 11.2, we conclude from (11.44) that
$$dI^{m1\alpha }+P^{m\alpha }(F)0\alpha .$$
(11.46)
To go on, we must distinguish the cases $`m<n`$ and $`m=n`$.
$`m<n`$. We have just proved that (iv) holds for $`p=m`$ if $`m<n`$. Using this, we conclude from (11.46) that
$$m<n:P^{m\alpha }(F)=0\alpha .$$
(11.47)
Hence, $`N^m`$ vanishes, see (11.45). Therefore $`b^m`$ vanishes as well because $`b^m`$ is present in Eq. (11.43) only if $`N^m`$ is present too \[using Eqs. (10.28), one verifies this by making the linear combinations of the $`[M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}]^p`$ and $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}`$ explicit that enter in (11.37)\]. Hence, we have
$$m<n:N^m=b^m=0,dB^{m1}=s\widehat{B}^m.$$
(11.48)
Moreover, using (11.47) in (11.46), we get $`dI^{m1\alpha }0`$. Hence, $`I^{m1\alpha }`$ is weakly $`d`$-closed and has form-degree $`<n1`$ (since we are discussing the cases $`m<n`$). We conclude, using corollary 11.1, that $`I^{m1\alpha }d\omega ^{m2\alpha }`$ for some $`\omega ^{m2\alpha }`$ if $`m1>0`$, or $`I^{0\alpha }=\lambda ^\alpha `$ for some constants $`\lambda ^\alpha `$ if $`m1=0`$. Since we assume that (iii) holds for $`p=m1`$, we get
$$m<n:I^{m1\alpha }dI^{m2\alpha }+P^{m1\alpha }(F)\alpha $$
(11.49)
(with $`P^{0\alpha }(F)\lambda ^\alpha `$ if $`m1=0`$). Using corollary 11.3 we conclude from (11.49) that $`I^{m1\alpha }dI^{m2\alpha }P^{m1\alpha }(F)`$ is $`s`$-exact,
$$m<n:I^{m1\alpha }=sK^{m1\alpha }+dI^{m2\alpha }+P^{m1\alpha }(F)\alpha .$$
(11.50)
Using (11.50) in (11.42), we get
$$m<n:\omega ^{m1}=[P^{m1\alpha }(F)+sK^{m1\alpha }+dI^{m2\alpha }]\mathrm{\Theta }_\alpha +B^{m1}.$$
(11.51)
To deal with the first term on the right hand side of (11.51), we use that every $`P^{m1\alpha }(F)\mathrm{\Theta }_\alpha `$ can be written as $`P^{m1\alpha }(F)\mathrm{\Theta }_\alpha =N^{m1}+M^{m1}+\delta _0^{m1}\widehat{\lambda }`$ for some $`N^{m1}`$ and $`M^{m1}`$ and some constant $`\widehat{\lambda }`$ which can only contribute if $`m1=0`$. This is guaranteed because $`\{1,M_{r_1\mathrm{}r_K|s_1\mathrm{}s_N}(\theta (C),f(F)),N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))\}`$ is a basis of all $`P^\alpha (F)\mathrm{\Theta }_\alpha `$, see Section 10 (note that this is the place where we use the completeness property of this basis).
Now, $`N^{m1}`$ is trivial in $`H(s|d,\mathrm{\Omega })`$ since each $`N_{r_1\mathrm{}r_Ks_1\mathrm{}s_N}(\theta (C),f(F))`$ is trivial, see Eqs. (10.28). Furthermore, $`(sK^{m1\alpha }+dI^{m2\alpha })\mathrm{\Theta }_\alpha `$ is trivial too, due to
$$(sK^{m1\alpha }+dI^{m2\alpha })\mathrm{\Theta }_\alpha =s(K^{m1\alpha }\mathrm{\Theta }_\alpha +I^{m2\alpha }[\mathrm{\Theta }_\alpha ]^1)+d(I^{m2\alpha }\mathrm{\Theta }_\alpha )$$
where we used once again (11.13). Since trivial contributions to $`\omega ^{m1}`$ can be neglected (see above), we can thus assume, without loss of generality,
$$m<n:\omega ^{m1}=M^{m1}+B^{m1}+\delta _0^{m1}\widehat{\lambda }.$$
(11.52)
For every $`M^{m1}`$ there is a $`\stackrel{~}{B}^m`$ such that $`dM^{m1}=s\stackrel{~}{B}^m`$. This holds by Eqs. (10.28) (more precisely: the equation with $`q=1`$ there). Using in addition Eq. (11.48), we get
$$m<n:d\omega ^{m1}=sB^m,B^m=\stackrel{~}{B}^m+\widehat{B}^m.$$
Using this in $`s\omega ^m+d\omega ^{m1}=0`$, we get
$$m<n:s(\omega ^mB^m)=0.$$
From this we conclude, using corollary 11.2,
$$m<n:\omega ^mB^m+I^{m\alpha }\mathrm{\Theta }_\alpha .$$
(11.53)
This proves (11.38) for $`p=m`$ if $`m<n`$.
$`m=n`$. In this case we conclude from (11.46), using (11.23) and (11.24),
$$P^{n\alpha }=\lambda ^{A\alpha }P_A,d(I^{n1\alpha }+\lambda ^{A\alpha }I_A^{n1})0$$
(11.54)
for some constant coefficents $`\lambda ^{A\alpha }`$. Using (11.18), we conclude from (11.54) that $`I^{n1\alpha }+\lambda ^{A\alpha }I_A^{n1}\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }+d\omega ^{n2}`$ for some constant coefficients $`\lambda ^{\mathrm{\Delta }\alpha }`$ and some $`\omega ^{n2}`$. Hence, $`I^{n1\alpha }+\lambda ^{A\alpha }I_A^{n1}\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }`$ is weakly $`d`$-exact. Using (iii) for $`p=m1=n1`$ and then once again corollary 11.3, we conclude from (11.54)
$$I^{n1\alpha }=\lambda ^{A\alpha }I_A^{n1}+\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }+P^{n1\alpha }(F)+dI^{n2\alpha }+sK^{n1\alpha }$$
(11.55)
for some $`K^{n1\alpha }`$. Furthermore, because of (11.28) and (11.54), we have $`N^n=\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }`$ in Eq. (11.45), for some $`\lambda ^\mathrm{\Gamma }`$ such that $`\lambda ^{A\alpha }=\lambda ^\mathrm{\Gamma }k_\mathrm{\Gamma }^{A\alpha }`$. Now consider
$`\widehat{\omega }^n`$ $`:=`$ $`\omega ^n\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma }`$
$`\widehat{\omega }^{n1}`$ $`:=`$ $`\omega ^{n1}\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }\mathrm{\Theta }_\alpha \lambda ^\mathrm{\Gamma }[B_\mathrm{\Gamma }^{n1}k_\mathrm{\Gamma }^{A\alpha }I_A^{n1}\mathrm{\Theta }_\alpha ]`$ (11.56)
$`=`$ $`[P^{n1\alpha }(F)+dI^{n2\alpha }+sK^{n1\alpha }]\mathrm{\Theta }_\alpha +\widehat{B}^{n1}`$
where $`\widehat{B}^{n1}=B^{n1}\lambda ^\mathrm{\Gamma }B_\mathrm{\Gamma }^{n1}`$. One has $`s\widehat{\omega }^n+d\widehat{\omega }^{n1}=0`$, due to (11.22) and (11.32) (and $`s\omega ^n+d\omega ^{n1}=0`$). The last line in (11.56) is analogous to (11.51). By the same arguments that have led from (11.51) to (11.53), we conclude that
$$\widehat{\omega }^nB^n+I^{n\alpha }\mathrm{\Theta }_\alpha .$$
This yields (11.38) for $`p=m=n`$ due to $`\omega ^n=\widehat{\omega }^n+\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma }`$.
3. We shall treat the case $`m=n`$; the proof for $`m<n`$ is simpler and obtained from the one for $`m=n`$ by setting $`\lambda ^{\mathrm{\Delta }\alpha }=\lambda ^\mathrm{\Gamma }=0`$ and substituting $`m`$ for $`n`$ in the following formulae.
Consider the $`n`$-form $`\omega ^n=I^{n\alpha }\mathrm{\Theta }_\alpha +B^n+\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma }`$. Due to (11.37), (11.22) and (11.32), one has
$`s(I^{n\alpha }\mathrm{\Theta }_\alpha )=0`$ (11.57)
$`sB^n=d(B^{n1}+M^{n1})`$ (11.58)
$`sV_{\mathrm{\Delta }\alpha }=d[j_\mathrm{\Delta }\mathrm{\Theta }_\alpha ]`$ (11.59)
$`sW_\mathrm{\Gamma }=d[B_\mathrm{\Gamma }^{n1}k_\mathrm{\Gamma }^{A\alpha }I_A^{n1}\mathrm{\Theta }_\alpha ]`$ (11.60)
for some $`B^{n1}`$ and $`M^{n1}`$. Hence, one has $`s\omega ^n+d\omega ^{n1}=0`$ where $`\omega ^{n1}=\widehat{B}^{n1}+\widehat{I}^{n1\alpha }\mathrm{\Theta }_\alpha `$ with
$`\widehat{B}^{n1}=B^{n1}+\lambda ^\mathrm{\Gamma }B_\mathrm{\Gamma }^{n1}`$ (11.61)
$`\widehat{I}^{n1\alpha }\mathrm{\Theta }_\alpha =M^{n1}+\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }\mathrm{\Theta }_\alpha \lambda ^\mathrm{\Gamma }k_\mathrm{\Gamma }^{A\alpha }I_A^{n1}\mathrm{\Theta }_\alpha `$ (11.62)
We assume now that $`\omega ^n`$ is trivial,
$$I^{n\alpha }\mathrm{\Theta }_\alpha +B^n+\lambda ^{\mathrm{\Delta }\alpha }V_{\mathrm{\Delta }\alpha }+\lambda ^\mathrm{\Gamma }W_\mathrm{\Gamma }0.$$
(11.63)
Then $`\omega ^{n1}`$ is trivial too (see footnote 16),
$$\widehat{B}^{n1}+\widehat{I}^{n1\alpha }\mathrm{\Theta }_\alpha 0.$$
(11.64)
Since we assume that (ii) holds for $`p=n1`$, we conclude from Eq. (11.64)
$`B^{n1}=\lambda ^\mathrm{\Gamma }B_\mathrm{\Gamma }^{n1}(\widehat{B}^{n1}=0)`$ (11.65)
$`\widehat{I}^{n1\alpha }\mathrm{\Theta }_\alpha N^{n1}+(dI^{n2\alpha })\mathrm{\Theta }_\alpha .`$ (11.66)
(11.65) implies that both $`\lambda ^\mathrm{\Gamma }B_\mathrm{\Gamma }^{n1}`$ and $`B^{n1}`$ vanish. This is trivial if $`n`$ is odd because then no $`N_\mathrm{\Gamma }`$ is present (recall that $`N_\mathrm{\Gamma }`$ is a polynomial in the $`C^I`$ and $`F^I`$ and has thus even form-degree). If $`n`$ is even, then no $`M^{n1}`$ can be present in Eq. (11.58) (as $`M^{n1}`$ is a polynomial in the $`C^I`$ and $`F^I`$ too). By (11.31), we have $`d(\lambda ^\mathrm{\Gamma }B_\mathrm{\Gamma }^{n1})=s(\lambda ^\mathrm{\Gamma }b_\mathrm{\Gamma }^n)+\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }`$. Using this and (11.65) in Eq. (11.58), for $`n`$ even, one gets $`s(B^n+\lambda ^\mathrm{\Gamma }b_\mathrm{\Gamma }^n)=\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }`$, i.e., $`\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }`$ is $`s`$-exact in the small algebra. This implies $`\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }=0`$ because $`\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }`$ is a linear combination of nontrivial representatives of $`H(s,)`$ by construction (recall that it is a linear combination of the $`N_i`$ in corollary 10.5) and is thus $`s`$-exact in $``$ only if it vanishes. $`\lambda ^\mathrm{\Gamma }N_\mathrm{\Gamma }=0`$ implies that all coefficients $`\lambda ^\mathrm{\Gamma }`$ vanish because the $`N_\mathrm{\Gamma }`$ are linearly independent by assumption, see Eq. (11.29). Hence, we get indeed
$$\lambda ^\mathrm{\Gamma }=0\mathrm{\Gamma }$$
(11.67)
and thus also, by Eq. (11.65),
$$B^{n1}=0.$$
(11.68)
Using $`\lambda ^\mathrm{\Gamma }=0`$, Eqs. (11.62) and (11.66) give
$$N^{n1}M^{n1}\lambda ^{\mathrm{\Delta }\alpha }j_\mathrm{\Delta }\mathrm{\Theta }_\alpha (dI^{n2\alpha })\mathrm{\Theta }_\alpha .$$
(11.69)
We have
$$N^{n1}M^{n1}=P^{n1\alpha }(F)\mathrm{\Theta }_\alpha (C)$$
(11.70)
for some $`P^{n1\alpha }(F)`$. By assumption no nonvanishing linear combination of the $`j_\mathrm{\Delta }`$ is weakly $`d`$-exact, see Eq. (11.19). Since each $`P^{n1\alpha }(F)`$ is $`d`$-exact, (11.69) implies
$$\lambda ^{\mathrm{\Delta }\alpha }=0(\mathrm{\Delta },\alpha ).$$
(11.71)
Since we assume that (iv) holds for $`p=n1`$, we conclude from (11.69) through (11.71) also that all $`P^{n1\alpha }(F)`$ vanish and thus that $`N^{n1}M^{n1}=0`$. The latter implies that $`N^{n1}`$ and $`M^{n1}`$ vanish separately because they contain independent representatives of $`H(s,)`$,
$$N^{n1}=0,M^{n1}=0.$$
Using this and Eq. (11.68) in (11.58), the latter turns into $`sB^n=0`$. By the very definition (11.16), $`B^n`$ is a linear combination of terms with nonvanishing and linearly independent $`s`$-transformations. Hence, $`sB^n=0`$ holds if and only if $`B^n`$ itself vanishes. We conclude
$$B^n=0.$$
(11.72)
(11.67), (11.71) and (11.72) provide already the assertions for $`\lambda ^\mathrm{\Gamma }`$, $`\lambda ^{\mathrm{\Delta }\alpha }`$ and $`B^n`$ in part (ii) of the theorem. We still have to prove those for $`I^{n\alpha }\mathrm{\Theta }_\alpha `$. Using $`\lambda ^\mathrm{\Gamma }=\lambda ^{\mathrm{\Delta }\alpha }=B^n=0`$, (11.63) reads
$$I^{n\alpha }\mathrm{\Theta }_\alpha =s\eta ^n+d\eta ^{n1}$$
(11.73)
where we made the trivial terms explicit. Acting with $`s`$ on this equation gives $`d(s\eta ^{n1})=0`$ and thus
$$s\eta ^{n1}+d\eta ^{n2}=0$$
(11.74)
for some $`\eta ^{n2}`$, thanks to the algebraic Poincaré lemma. Since we assume that (i) holds for $`p=n1`$, we conclude from (11.74) that
$$\eta ^{n1}=\stackrel{~}{B}^{n1}+\stackrel{~}{I}^{n1\alpha }\mathrm{\Theta }_\alpha +s\stackrel{~}{\eta }^{n1}+d\stackrel{~}{\eta }^{n2}.$$
(11.75)
As above, we have
$`d\stackrel{~}{B}^{n1}=s(\stackrel{~}{B}^n+\stackrel{~}{b}^n)+\stackrel{~}{N}^n`$
$`d(\stackrel{~}{I}^{n1\alpha }\mathrm{\Theta }_\alpha )=(d\stackrel{~}{I}^{n1\alpha })\mathrm{\Theta }_\alpha s(\stackrel{~}{I}^{n1\alpha }[\mathrm{\Theta }_\alpha ]^1).`$
Using this in (11.73), we get
$$[I^{n\alpha }d\stackrel{~}{I}^{n1\alpha }\stackrel{~}{P}^{n\alpha }]\mathrm{\Theta }_\alpha =s(\eta ^n\stackrel{~}{B}^n\stackrel{~}{b}^nd\stackrel{~}{\eta }^{n1}\stackrel{~}{I}^{n1\alpha }[\mathrm{\Theta }_\alpha ]^1),$$
(11.76)
where
$$\stackrel{~}{P}^{n\alpha }\mathrm{\Theta }_\alpha =\stackrel{~}{N}^n.$$
(11.77)
Using corollary 11.2 we conclude from (11.76) that
$$I^{n\alpha }d\stackrel{~}{I}^{n1\alpha }\stackrel{~}{P}^{n\alpha }0.$$
(11.78)
(11.77) and (11.78) complete the demonstration of (ii).
4. $`I^md\omega ^{m1}`$ implies $`I^m0`$, i.e. $`I^m`$ is trivial in $`H(s|d,\mathrm{\Omega })`$. Indeed, $`I^md\omega ^{m1}`$ means that $`I^m=\delta \omega ^m+d\omega ^{m1}`$ for some $`\omega ^m`$ with antifield number 1. Hence, $`I^m`$ is a cocycle of $`H(s|d,\mathrm{\Omega })`$ (since it is $`s`$-closed due to $`I^m`$) and trivial in $`H(\delta |d,\mathrm{\Omega })`$. It is therefore also trivial in $`H(s|d,\mathrm{\Omega })`$ by theorem 7.1 (cf. proof of (7.8)). Now, $`I^m0`$ is just a special case of $`I^{p\alpha }\mathrm{\Theta }_\alpha 0`$ (due to $`1\{\mathrm{\Theta }_\alpha \}`$). Hence, using (ii) for $`p=m`$ (which we have already proved), we conclude $`I^mdI^{m1}+P^m(F)`$ for some $`I^{m1}`$ and some $`P^m(F)`$. Conversely, if $`m>0`$, we have $`P^m(F)=dq^{m1}`$ for some Chern-Simons form $`q^{m1}`$ and thus $`I^mdI^{m1}+P^m(F)`$ implies $`I^md\omega ^{n1}`$ with $`\omega ^{n1}=I^{m1}+q^{m1}`$.
### 11.4 Appendix 11.A: 2-dimensional pure Yang-Mills theory
Pure 2-dimensional Yang-Mills theory needs a special treatment because $`H_{char}^{n2}(d,\mathrm{\Omega })H_{char}^0(d,\mathrm{\Omega })`$ is not given by the global reducibility identities associated with the abelian gauge symmetries (theorem 6.8), but is much bigger and in fact infinite-dimensional (see explicit description at the end of the appendix). This feature disappears if one couples coloured matter fields. We discuss the pure Yang-Mills case for the sake of completeness contenting ourselves with case I, i.e., with the solution of the consistency condition $`s\omega ^p+d\omega ^{p1}=0`$ in the space of all local forms.
We consider the standard Lagrangian
$$L=\frac{1}{4}F_{\mu \nu }^IF_I^{\mu \nu }$$
where $`F_I^{\mu \nu }=g_{IJ}F^{\mu \nu J}`$ involves an invertible $`𝒢`$-invariant symmetric tensor $`g_{IJ}`$. The gauge group may contain abelian factors.
Due to $`n=2`$, we have $`F_{01}^I=F_{10}^I=(1/2)ϵ^{\mu \nu }F_{\mu \nu }^I=F^I`$ and the equations of motion set all covariant derivatives of $`F_{01I}`$ to zero. The result for $`H(s)`$ (corollary 11.2) implies thus immediately that
$$\omega ^0=I^\alpha (x,F_{01})\mathrm{\Theta }_\alpha +s\eta ^0$$
(11.79)
where the $`I^\alpha (x,F_{01})`$ are arbitrary $`𝒢`$-invariant local functions of the $`F_{01I}`$ and the $`x^\mu `$ (the latter can occur because we are discussing case I). (11.79) is therefore the general solution of the consistency condition for $`p=0`$.
To find the general solutions with $`p=1`$ and $`p=2`$, we use the descent equations and examine whether an $`\omega ^0`$ as in Eq. (11.79) can be lifted to solutions of the consistency condition with form-degree 1 or 2. In order to lift $`\omega ^0`$ to form-degree 1, it is necessary and sufficient that the $`dI^\alpha (x,F_{01})`$ vanish weakly, for all $`\alpha `$ (see Section 11.2). Since the $`I^\alpha (x,F_{01})`$ are $`𝒢`$-invariant, we have
$$dI^\alpha (x,F_{01})=dx^\mu \left[\frac{I^\alpha (x,F_{01})}{x^\mu }+(D_\mu F_{01I})\frac{I^\alpha (x,F_{01})}{F_{01I}}\right]dx^\mu \frac{I^\alpha (x,F_{01})}{x^\mu }$$
where we have used $`D_\mu F_{01I}0`$. No nonvanishing function of the undifferentiated $`F_{01I}`$ is weakly zero since the equations of motion contain derivatives of $`F_{01I}`$. Hence, $`\omega ^0`$ can be lifted to form-degree 1 if and only if $`I^\alpha /x^\mu =0`$, i.e., the $`I^\alpha `$ must not depend explicitly on the $`x^\mu `$. It turns out that this also suffices to lift $`\omega ^0`$ to form-degree 2. To show this, we introduce
$$\stackrel{~}{C}_I^{}:=C_I^{}+A_I^{}+F_I$$
where $`C_I^{}=d^2xC_I^{}`$, $`A_I^{}=dx^\mu ϵ_{\mu \nu }A_I^\nu `$ and $`F_I=\frac{1}{2}ϵ_{\mu \nu }F_I^{\mu \nu }`$. One has
$$(s+d)\stackrel{~}{C}_I^{}=(C^J+A^J)ef_{JI}^{}{}_{}{}^{K}\stackrel{~}{C}_K^{},$$
i.e., the $`\stackrel{~}{C}_I^{}`$ transform under $`(s+d)`$ according to the adjoint representation of $`𝒢`$ with “$`(s+d)`$-ghosts $`(C^I+A^I)`$”. $`𝒢`$-invariant functions of the $`\stackrel{~}{C}_I^{}`$ are thus $`(s+d)`$-closed,
$$(s+d)I^\alpha (\stackrel{~}{C}^{})=0.$$
Recall that the $`\mathrm{\Theta }_\alpha `$ are polynomials in the $`\theta _r(C)`$ and that the latter are related to $`𝒢`$-invariant polynomials $`f_r(F)`$ via the transgression formula (10.21) which decomposes into Eqs. (10.23). In two dimensions, all $`f_r(F)`$ with degree $`m(r)>1`$ in the $`F^I`$ vanish. The transgression formula gives thus
$$m(r)>1:(s+d)q_r(C+A,F)=0,q_r(C+A,F)=[\theta _r]^0+[\theta _r]^1+[\theta _r]^2.$$
For $`m(r)=1`$ one gets $`(s+d)(C^I+A^I)=F^I`$ where $`C^I`$, $`A^I`$ and $`F^I`$ are abelian. One has
$`\text{abelian }F^I\text{:}F^I`$ $`=`$ $`\frac{1}{2}dx^\mu dx^\nu F_{\mu \nu }^I`$
$`=`$ $`d(\frac{1}{2}x^\mu dx^\nu F_{\mu \nu }^I)+\frac{1}{2}x^\mu dx^\nu dx^\rho _\rho F_{\mu \nu }^I`$
$`=`$ $`d(\frac{1}{2}x^\mu dx^\nu F_{\mu \nu }^I)s(\frac{1}{2}d^2xx^\mu ϵ_{\mu \nu }A^{\nu I})`$
$`=`$ $`(s+d)(\frac{1}{2}x^\mu dx^\nu F_{\mu \nu }^I\frac{1}{2}d^2xx^\mu ϵ_{\mu \nu }A^{\nu I})`$
where we have used that one has $`_\mu F_{01}^I=s(ϵ_{\mu \nu }A^{\nu I})`$ for abelian $`F^I`$. Hence we have two different quantities whose $`(s+d)`$-transformation equals $`F^I`$ in the abelian case ($`C^I+A^I`$ and the quantity in the previous equation). The difference of these quantities is thus an $`(s+d)`$-closed extension of the abelian ghosts. Hence, we can complete every $`\theta _r(C)`$, whether nonabelian or abelian, to an $`(s+d)`$-invariant quantity $`\stackrel{~}{q}_r`$,
$`m(r)>1:`$ $`\stackrel{~}{q}_r=q_r(C+A,F)=[\theta _r]^0+[\theta _r]^1+[\theta _r]^2`$
$`m(r)=1:`$ $`\stackrel{~}{q}_r=C^I+A^I\frac{1}{2}(x^\mu dx^\nu F_{\mu \nu }^Id^2xx^\mu ϵ_{\mu \nu }A^{\nu I})(\text{abelian }I).`$
Due to $`(s+d)\stackrel{~}{q}_r=0`$ and $`(s+d)I^\alpha (\stackrel{~}{C}^{})=0`$, we have
$$(s+d)\left[I^\alpha (\stackrel{~}{C}^{})\mathrm{\Theta }_\alpha (\stackrel{~}{q})\right]=0,$$
(11.80)
where $`\mathrm{\Theta }_\alpha (\stackrel{~}{q})`$ arises from $`\mathrm{\Theta }_\alpha `$ by substituting the $`\stackrel{~}{q}_r`$ for the $`\theta _r(C)`$. The decomposition of (11.80) into pieces with definite form-degree reads
$`s[I^\alpha \mathrm{\Theta }_\alpha ]^2+d[I^\alpha \mathrm{\Theta }_\alpha ]^1=0,`$
$`s[I^\alpha \mathrm{\Theta }_\alpha ]^1+d[I^\alpha \mathrm{\Theta }_\alpha ]^0=0,`$
$`s[I^\alpha \mathrm{\Theta }_\alpha ]^0=0,`$
where $`[I^\alpha \mathrm{\Theta }_\alpha ]^p`$ is the $`p`$-form contained in $`I^\alpha (\stackrel{~}{C}^{})\mathrm{\Theta }_\alpha (\stackrel{~}{q})`$,
$$I^\alpha (\stackrel{~}{C}^{})\mathrm{\Theta }_\alpha (\stackrel{~}{q})=\underset{p=0}{\overset{2}{}}[I^\alpha \mathrm{\Theta }_\alpha ]^p.$$
Every $`I^\alpha (F_{01})\mathrm{\Theta }_\alpha =[I^\alpha \mathrm{\Theta }_\alpha ]^0`$ can thus indeed be lifted to solutions of the consistency condition with form-degrees 1 and 2.
It is now easy to complete the analysis. $`s\omega ^1+d\omega ^0=0`$ yields $`s(\omega ^1[I^\alpha \mathrm{\Theta }_\alpha ]^1d\eta ^0)=0`$. By the result on $`H(s)`$, the general solution of $`s\omega ^1+d\omega ^0=0`$ is accordingly
$$\omega ^1=[I^\alpha \mathrm{\Theta }_\alpha ]^1+dx^\mu I_\mu ^\alpha (x,F_{01})\mathrm{\Theta }_\alpha +s\eta ^1+d\eta ^0$$
(11.81)
where the $`I_\mu ^\alpha (x,F_{01})`$ are arbitrary $`𝒢`$-invariant local functions of the $`F_{01I}`$ and the $`x^\mu `$.
We know already that every $`[I^\alpha \mathrm{\Theta }_\alpha ]^1`$ can be lifted to $`[I^\alpha \mathrm{\Theta }_\alpha ]^2`$. In order to lift an $`\omega ^1`$ as in Eq. (11.81), it is therefore necessary that the piece $`\widehat{\omega }^1:=dx^\mu I_\mu ^\alpha (x,F_{01})\mathrm{\Theta }_\alpha `$ can be lifted too. By arguments analogous to those used above, this requires $`d_x\widehat{\omega }^1=0`$ where $`d_x=dx^\mu /x^\mu `$. Since $`H^1(d_x)`$ is trivial (ordinary Poincaré lemma in $`^2`$), this gives $`\widehat{\omega }^1=d_xJ^\alpha (x,F_{01})\mathrm{\Theta }_\alpha `$ for some $`𝒢`$-invariant local functions $`J^\alpha (x,F_{01})`$. This implies that $`\widehat{\omega }^1`$ is trivial in $`H(s|d)`$. Indeed, using $`D_\mu F_{01I}=\delta (ϵ_{\mu \nu }A_I^\nu )`$, the $`𝒢`$-invariance of $`I^\alpha `$ and $`J^\alpha `$, and Eq. (11.13), one obtains
$`d_xJ^\alpha (x,F_{01})\mathrm{\Theta }_\alpha =d[J^\alpha (x,F_{01})\mathrm{\Theta }_\alpha ](dx^\mu D_\mu F_{01I}){\displaystyle \frac{J^\alpha (x,F_{01})}{F_{01I}}}\mathrm{\Theta }_\alpha J^\alpha (x,F_{01})d\mathrm{\Theta }_\alpha `$
$`=d[J^\alpha (x,F_{01})\mathrm{\Theta }_\alpha ]+s[A_I^{}{\displaystyle \frac{J^\alpha (x,F_{01})}{F_{01I}}}\mathrm{\Theta }_\alpha +J^\alpha (x,F_{01})A^I{\displaystyle \frac{\mathrm{\Theta }_\alpha }{C^I}}].`$
Hence, those solutions $`\omega ^1`$ which can be lifted are of the form $`[I^\alpha \mathrm{\Theta }_\alpha ]^1+s\eta ^1+d\eta ^0`$. Inserting this in $`s\omega ^2+d\omega ^1=0`$ yields $`s(\omega ^2[I^\alpha \mathrm{\Theta }_\alpha ]^2d\eta ^1)=0`$. Every element of $`H(s)`$ with form-degree 2 is $`d_x`$-closed and thus $`d_x`$-exact, due to $`H^2(d_x)=0`$. Using arguments as before, one concludes that the general solution of $`s\omega ^2+d\omega ^1=0`$ is
$$\omega ^2=[I^\alpha \mathrm{\Theta }_\alpha ]^2+s\eta ^{\prime \prime 2}+d\eta ^{\prime \prime 1}.$$
(11.82)
Remark. Using the isomorphism $`H_{\mathrm{char}}^0(d,\mathrm{\Omega })H_2^2(\delta |d,\mathrm{\Omega })H^{2,2}(s|d,\mathrm{\Omega })`$ (see theorems 6.2 and 7.1), one deduces from the above result that $`H_{\mathrm{char}}^0(d,\mathrm{\Omega })`$ and $`H_{\mathrm{char}}^0(d,)`$ are represented by arbitrary $`𝒢`$-invariant polynomials in the $`F_{01I}`$. These cohomological groups are thus infinite dimensional. This explains the different results as compared to higher dimensions where the nontrivial representatives of $`H_{\mathrm{char}}^{n2}(d,\mathrm{\Omega })`$ correspond one-to-one to the free abelian gauge symmetries.
## 12 Discussion of the results for Yang-Mills type theories
Theorem 11.1 gives the general solution of the consistency condition $`sa+db=0`$ at all form-degrees and ghost numbers for theories of the Yang-Mills type without free abelian gauge symmetries (in the sense of subsection 11.1) and in spacetime dimensions greater than $`2`$. The case of free abelian symmetries is treated in the next section. In this section, we spell out the physical implications of the theorem by expliciting the results in the relevant ghost numbers.
To that end we shall use the notation $`f([F,\psi ]_D)`$ for functions that depend only on the Yang-Mills field strengths, the matter fields and their covariant derivatives,
$`f([F,\psi ]_D)f(F_{\mu \nu }^I,D_\rho F_{\mu \nu }^I,D_\rho D_\sigma F_{\mu \nu }^I,\mathrm{},\psi ^i,D_\mu \psi ^i,D_\mu D_\nu \psi ^i,\mathrm{}).`$
We recall that the results are valid for general Lagrangians of the Yang-Mills type, provided these fulfill the technicality assumptions of “regularity” and “normality” explained above. The results cover in particular the standard model and effective gauge theories.
### 12.1 $`H^{1,n}(s|d)`$: Global symmetries and Noether currents
#### 12.1.1 Solutions of the consistency condition at negative ghost number
We start with the discussion of the results at negative ghost number. First, we recall that the groups $`H^{q,n}(s|d)`$ are trivial for $`q>1`$. This implies that there is no characteristic cohomology in form degree $`<n1`$, i.e., no non trivial higher order conservation law. Any conserved local antisymmetric tensor $`A^{\mu _1\mathrm{}\mu _q}`$ ($`q>1`$) is trivial, i.e., equal on-shell to the divergence of a local antisymmetric tensor with one more index,
$`q>1:`$ $`_{\mu _1}A^{\mu _1\mathrm{}\mu _q}0,A^{\mu _1\mathrm{}\mu _q}=A^{[\mu _1\mathrm{}\mu _q]}`$
$``$ $`A^{\mu _1\mathrm{}\mu _q}_{\mu _0}B^{\mu _0\mu _1\mathrm{}\mu _q},B^{\mu _0\mu _1\mathrm{}\mu _q}=B^{[\mu _0\mu _1\mathrm{}\mu _q]}.`$
We stress again that the important point in this statement is that the $`B^{\mu _0\mu _1\mathrm{}\mu _q}`$ are local functions; the statement would otherwise be somewhat empty due to the ordinary Poincaré lemma for $`^n`$.
The only non-vanishising group at negative ghost number is $`H^{1,n}(s|d,\mathrm{\Omega })`$. The nontrivial representatives of $`H^{1,n}(s|d,\mathrm{\Omega })`$ are the generators of the nontrivial global symmetries, denoted by $`K_\mathrm{\Delta }`$ and $`K_A`$ in the previous section. Indeed, one must set $`\mathrm{\Theta }_\alpha =1`$ in order that Eqs. (11.21) and (11.33) yield solutions with ghost number $`1`$. The general solution of the consistency condition with ghost number $`1`$ is thus
$$\omega ^{1,n}\lambda ^\mathrm{\Delta }K_\mathrm{\Delta }+\lambda ^AK_A,$$
in form-degree $`n`$, where $`K_\mathrm{\Delta }`$ and $`K_A`$ are related to the gauge invariant conserved currents $`j_\mathrm{\Delta }`$ and to the characteristic classes $`P_A(F)`$ respectively, through
$`sK_\mathrm{\Delta }+dj_\mathrm{\Delta }=0,j_\mathrm{\Delta },`$
$`sK_A+dI_A^{n1}=P_A(F),I_A^{n1}.`$
That is, the coefficients of the antifields $`A_I^\mu `$ and $`\psi _i^{}`$ in the $`K`$’s determine the transformations of the corresponding field in the global symmetry associated with the conserved Noether currents.
#### 12.1.2 Structure of global symmetries and conserved currents.
A determination of a complete set of gauge invariant nontrivial conserved currents $`j_\mathrm{\Delta }`$ depends on the specific model under study. It also depends on the detailed form of the Lagrangian whether or not invariants $`I_A^{n1}`$ exist which are related to characteristic classes by Eq. (11.23). However, we can make the description of the $`K_\mathrm{\Delta }`$ and $`K_A`$ a little more precise without specifying $`L`$. Namely, as we prove in the appendix to this section, one can always choose the $`j_\mathrm{\Delta }`$ and $`I_A^{n1}`$ such that all $`K_\mathrm{\Delta }`$ and $`K_A`$ take the form
$`K_\mathrm{\Delta }`$ $`=`$ $`d^nx\left[A_I^\mu Q_{\mathrm{\Delta }\mu }^I(x,[F,\psi ]_D)+\psi _i^{}Q_\mathrm{\Delta }^i(x,[F,\psi ]_D)\right]`$ (12.1)
$`K_A`$ $`=`$ $`d^nx\left[A_I^\mu Q_{A\mu }^I(x,[F,\psi ]_D)+\psi _i^{}Q_A^i(x,[F,\psi ]_D)\right]`$ (12.2)
where the $`Q_{\mathrm{\Delta }\mu }^I(x,[F,\psi ]_D)`$ transform under $`𝒢`$ according to the coadjoint representation and the $`Q_\mathrm{\Delta }^i(x,[F,\psi ]_D)`$ according to the same representation as the $`\psi ^i`$. We assume here that we work in the space of all local forms (case I). An analogous statement holds in the space of Poincaré invariant local forms (case II) where (12.1) and (12.2) hold with Poincaré invariant $`K`$’s.
This result, and the relationship between the $`K`$’s and the conserved currents, enables us to draw the following conclusions (in Yang-Mills type theories without free abelian gauge symmetries, when the spacetime dimension exceeds 2):
1. In odd dimensional spacetime, every nontrivial conserved current is equivalent to a gauge invariant conserved current,
$$n=2k+1:_\mu j^\mu 0j^\mu j_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)$$
where $`j_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)`$ is $`𝒢`$-invariant and $``$ means “equal modulo trivial conserved currents”,
$$j^\mu h^\mu :j^\mu h^\mu +_\nu m^{[\nu \mu ]}.$$
2. In even dimensional spacetime a nontrivial conserved current is either equivalent to a completely gauge invariant current or to a current that is gauge invariant except for a Chern-Simons term,
$$n=2k:_\mu j^\mu 0j^\mu \{\begin{array}{cc}j_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)& \\ \text{or}& \\ I_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)+q_{\mathrm{CS}}^\mu (A,A)& \end{array}$$
where $`j_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)`$ and $`I_{\mathrm{inv}}^\mu (x,[F,\psi ]_D)`$ are $`𝒢`$-invariant, and $`q_{\mathrm{CS}}^\mu (A,A)`$ is dual to a Chern-Simons $`(n1)`$-form, i.e.,
$$q_{\mathrm{CS}}^\mu (A,A)=ϵ^{\mu \nu _1\mu _2\nu _2\mathrm{}\mu _k\nu _k}d_{I_1\mathrm{}I_k}A_{\nu _1}^{I_1}_{\mu _2}A_{\nu _2}^{I_2}\mathrm{}_{\mu _k}A_{\nu _k}^{I_k}+\mathrm{}$$
One can choose the basis of inequivalent conserved currents such that those currents which contain Chern-Simons terms correspond one-to-one to the characteristic classes $`P_A(F)`$ which are trivial in the equivariant characteristic cohomology. In particular, all conserved currents can be made strictly gauge invariant when no characteristic class is trivial in the equivariant characteristic cohomology.<sup>17</sup><sup>17</sup>17To our knowledge, it is still an open problem whether in standard Yang-Mills theory characteristic classes $`P(F)`$ with form-degree $`n`$ can be trivial in $`H_{\mathrm{char}}(d,)`$. (The problem occurs only in the space of forms with explict $`x^\mu `$-dependence.) In Section 13 of , we have claimed that the answer is negative for a polynomial dependence on $`x^\mu `$. However, the proof of the assertion given there is incorrect because the $`sl(n)`$-decomposition of the equations of motion used there does not yield pieces which are all weakly zero separately. If the answer were positive (contrary to our expectations), it would mean that non covariantizable currents could occur in standard Yang-Mills theory, contrary to the claim in theorem 2 in .
3. Every nontrivial global symmetry can be brought to a gauge covariant form. More precisely, let $`\delta _Q`$ be the generator of a global symmetry whose characteristics $`\delta _QA_\mu ^I=Q_\mu ^I`$ and $`\delta _Q\psi ^i=Q^i`$ are local functions of the fields ($`Q_\mu ^I`$ or $`Q^i`$ may depend explicitly on the $`x^\mu `$). Then one can bring $`\delta _Q`$ to a form (by subtracting trivial symmetries if necessary) such that the characteristics depend only on the $`x^\mu `$, the Yang-Mills field strengths and their covariant derivatives, and the matter fields and their covariant derivatives,
$$\delta _QA_\mu ^I=Q_\mu ^I(x,[F,\psi ]_D),\delta _Q\psi ^i=Q^i(x,[F,\psi ]_D),$$
where the $`Q_\mu ^I`$ transform under the coadjoint representation of $`𝒢`$ and the $`Q^i`$ transform under the same representation of $`𝒢`$ as the $`\psi ^i`$.
Note that the gauge covariant global symmetries commute with the gauge transformations (2.4): for instance, one has
$`[\delta _Q,\delta _ϵ]A_\mu ^I`$ $`=`$ $`\delta _Q(_\mu ϵ^I+ef_{JK}^{}{}_{}{}^{I}A_\mu ^Jϵ^K)\delta _ϵQ_\mu ^I`$
$`=`$ $`ef_{JK}^{}{}_{}{}^{I}Q_\mu ^Jϵ^K(eϵ^Kf_{KJ}^{}{}_{}{}^{I}Q_\mu ^J)=0.`$
Here we used $`\delta _Qϵ^I=0`$ where $`ϵ^I`$ are arbitrary fields. Of course, in general $`\delta _Q`$ would not commute with a special gauge transformation obtained by substituting functions of the $`A_\mu ^I`$, $`\psi ^i`$ and their derivatives for $`ϵ^I`$.
#### 12.1.3 Examples.
1. It should be noted that the gauge covariant form of a global symmetry is not always its most familiar version. In order to make a global symmetry gauge covariant, it may be necessary to add a trivial symmetry to it. We illustrate this feature now for conformal transformations. Consider 4-dimensional massless scalar electrodynamics,
$$n=4,L=\frac{1}{4}F_{\mu \nu }F^{\mu \nu }\frac{1}{2}(D_\mu \phi )D^\mu \overline{\phi }$$
where $`\phi `$ is a complex scalar field, $`\overline{\phi }`$ is the complex conjugate of $`\phi `$, and
$$F_{\mu \nu }=_\mu A_\nu _\nu A_\mu ,D_\mu \phi =_\mu \phi +\mathrm{i}eA_\mu \phi ,D_\mu \overline{\phi }=_\mu \overline{\phi }\mathrm{i}eA_\mu \overline{\phi }.$$
The action is invariant under the following infinitesimal conformal transformations
$`\delta _{\mathrm{conf}}A_\mu `$ $`=`$ $`\xi ^\nu _\nu A_\mu +(_\mu \xi ^\nu )A_\nu ,`$
$`\delta _{\mathrm{conf}}\phi `$ $`=`$ $`\xi ^\nu _\nu \phi +\frac{1}{4}(_\nu \xi ^\nu )\phi ,`$
$`\xi ^\mu `$ $`=`$ $`a^\mu +\omega ^{[\mu \nu ]}x_\nu +\lambda x^\mu +b^\mu x_\nu x^\nu 2x^\mu b^\nu x_\nu `$
where the $`a^\mu `$, $`\omega ^{[\mu \nu ]}`$, $`\lambda `$ and $`b^\mu `$ are constant parameters of conformal transformations and $`x_\mu =\eta _{\mu \nu }x^\nu `$. $`\delta _{\mathrm{conf}}`$ is not gauge covariant. To make it gauge covariant we add a trivial global symmetry to it, namely a special gauge transformation with “gauge parameter” $`ϵ=\xi ^\nu A_\nu `$. This special gauge transformation is $`\delta _{\mathrm{trivial}}A_\mu =_\mu (\xi ^\nu A_\nu )`$, $`\delta _{\mathrm{trivial}}\phi =\mathrm{i}e\xi ^\nu A_\nu \phi `$. $`\widehat{\delta }_{\mathrm{conf}}=\delta _{\mathrm{conf}}+\delta _{\mathrm{trivial}}`$ is gauge covariant,
$$\widehat{\delta }_{\mathrm{conf}}A_\mu =\xi ^\nu F_{\nu \mu },\widehat{\delta }_{\mathrm{conf}}\phi =\xi ^\nu D_\nu \phi +\frac{1}{4}(_\nu \xi ^\nu )\phi .$$
Since $`\widehat{\delta }_{\mathrm{conf}}`$ and $`\delta _{\mathrm{conf}}`$ differ only by a special gauge transformation, they are equivalent and yield the same variation of the Lagrangian,
$$\widehat{\delta }_{\mathrm{conf}}L=\delta _{\mathrm{conf}}L=_\mu (\xi ^\mu Lb^\mu \phi \overline{\phi }).$$
The Noether current corresponding to $`\widehat{\delta }_{\mathrm{conf}}`$ is gauge invariant,
$$j_{\mathrm{inv}}^\mu (x,[F,\phi ,\overline{\phi }]_D)=\underset{\mathrm{\Phi }=A_\nu ,\phi ,\overline{\phi }}{}(\widehat{\delta }_{\mathrm{conf}}\mathrm{\Phi })\frac{L}{(_\mu \mathrm{\Phi })}\xi ^\mu L+b^\mu \phi \overline{\phi }.$$
2. We shall now illustrate the unusual situation in which a nontrivial Noether current contains a Chern-Simons term. As a first example we consider 4-dimensional Yang-Mills theory with gauge group $`SU(2)`$ coupled nonminimally to a real $`SU(2)`$-singlet scalar field $`\varphi `$ via the following Lagrangian,
$$n=4,L=\frac{1}{4}F_{\mu \nu }^IF^{\mu \nu J}\delta _{IJ}\frac{1}{2}(_\mu \varphi )^\mu \varphi +\frac{1}{4}\varphi ϵ^{\mu \nu \rho \sigma }F_{\mu \nu }^IF_{\rho \sigma }^J\delta _{IJ}$$
where
$$F_{\mu \nu }^I=_\mu A_\nu ^I_\nu A_\mu ^I+eϵ_{IJK}A_\mu ^JA_\nu ^K.$$
The action is invariant under constant shifts of $`\varphi `$,
$`\delta _{\mathrm{shift}}\varphi =1,\delta _{\mathrm{shift}}A_\mu ^I=0\delta _{\mathrm{shift}}L=_\mu q_{\mathrm{CS}}^\mu (A,A),`$
$`q_{\mathrm{CS}}^\mu (A,A)=ϵ^{\mu \nu \rho \sigma }(\delta _{IJ}A_\nu ^I_\rho A_\sigma ^J+\frac{1}{3}eϵ_{IJK}A_\nu ^IA_\rho ^JA_\sigma ^K).`$
$`\delta _{\mathrm{shift}}`$ is obviously nontrivial and gauge covariant. The corresponding Noether current contains the Chern-Simons term $`q_{\mathrm{CS}}^\mu (A,A)`$ and is otherwise gauge invariant,
$$j^\mu =^\mu \varphi +q_{\mathrm{CS}}^\mu (A,A).$$
3. A variant of the previous example arises when one replaces the scalar field by the time coordinate $`x^0`$,
$$n=4,L=\frac{1}{4}F_{\mu \nu }^IF^{\mu \nu J}\delta _{IJ}+\frac{1}{4}x^0ϵ^{\mu \nu \rho \sigma }F_{\mu \nu }^IF_{\rho \sigma }^J\delta _{IJ}$$
with $`F_{\mu \nu }^I`$ as in the previous example. This example breaks of course Lorentz invariance and is given only for illustrative purposes. One can get rid of $`x^0`$ by integrating by parts the last term, at the price of introducing an undifferentiated $`A_\mu `$. The action is therefore invariant under temporal translations,
$$\delta _{\mathrm{time}}A_\mu ^I=F_{0\mu }^I\delta _{\mathrm{time}}L=_0L_\mu q_{\mathrm{CS}}^\mu (A,A)$$
with $`q_{\mathrm{CS}}^\mu (A,A)`$ as in the previous example. $`\delta _{\mathrm{time}}A_\mu ^I=F_{0\mu }^I`$ is already the gauge covariant version of temporal translations (one has $`\delta _{\mathrm{time}}A_\mu ^I=_0A_\mu ^I+\delta _{\mathrm{trivial}}A_\mu ^I`$ where $`\delta _{\mathrm{trivial}}A_\mu ^I`$ is a special gauge transformation with $`ϵ^I=A_0^I`$). Note that we used $`\delta _{\mathrm{time}}x^0=0`$, i.e., we transformed only the fields. The conserved Noether current corresponding to $`\delta _{\mathrm{time}}`$ is the component $`\nu =0`$ of the energy momentum tensor $`T_{\nu }^{}{}_{}{}^{\mu }`$. In the present case, it cannot be made fully gauge invariant but contains the Chern-Simons term $`q_{\mathrm{CS}}^\mu (A,A)`$,
$$T_{0}^{}{}_{}{}^{\mu }=F_{0\nu }^I\frac{L}{(_\mu A_\nu ^I)}\delta _0^\mu L+q_{\mathrm{CS}}^\mu (A,A).$$
### 12.2 $`H^{0,n}(s|d)`$: Deformations and BRST-invariant counterterms
We now turn to the local BRST cohomology at ghost number zero. This case covers deformations of the action and controls therefore the stability of the theory. We first make the results of theorem 11.1 more explicit for the particular value $`0`$ of the ghost number; we then discuss the implications.
The most general solution of the consistency condition with ghost number 0 and form-degree $`n`$ is
$$\omega ^{0,n}I^n+B^{0,n}+V^{0,n}+W^{0,n}$$
where:
1. $`I^n`$, i.e., $`I^n`$ is a strictly gauge invariant $`n`$-form,
Case I: $`I^n=d^nxI_{\mathrm{inv}}(x,[F,\psi ]_D)`$
Case II: $`I^n=d^nxI_{\mathrm{inv}}([F,\psi ]_D).`$
(we recall that in case I, one computes the cohomology in the algebra of forms having a possible explicit $`x`$-dependence; while the forms in case II have no explicit $`x`$-dependence and are Lorentz-invariant)
2. $`B^{0,n}`$ is a linear combination of the independent Chern-Simons $`n`$-forms, see Eq. (10.40). Solutions $`B^{0,n}`$ can thus only exist in odd spacetime dimensions.
3. $`V^{0,n}`$ are linear combinations of the solutions $`V_{\mathrm{\Delta }\alpha }`$ (11.21) related to global symmetries. In order that $`V_{\mathrm{\Delta }\alpha }`$ has ghost number 0, the $`\mathrm{\Theta }_\alpha `$ which appears in it must have ghost number 1. There are such $`\mathrm{\Theta }_\alpha `$ only when the gauge group has abelian factors, in which case the $`\mathrm{\Theta }_\alpha `$ are the abelian ghosts. In the absence of abelian factors, there are thus no $`V_{\mathrm{\Delta }\alpha }`$ at ghost number zero. Using Eq. (12.1), one gets explicitly
$`V^{0,n}`$ $`=`$ $`{\displaystyle \underset{I:\mathrm{abelian}}{}}\lambda _I^\mathrm{\Delta }(K_\mathrm{\Delta }C^I+j_\mathrm{\Delta }A^I)`$ (12.3)
$`=`$ $`d^nx{\displaystyle \underset{I:\mathrm{abelian}}{}}\lambda _I^\mathrm{\Delta }\left[A_J^\mu Q_{\mathrm{\Delta }\mu }^JC^I+\psi _i^{}Q_\mathrm{\Delta }^iC^I()^{ϵ_\mathrm{\Delta }}j_\mathrm{\Delta }^\mu A_\mu ^I\right]`$
where the $`j_\mathrm{\Delta }^\mu `$ are the nontrivial gauge invariant Noether currents, $`Q_{\mathrm{\Delta }\mu }^J`$ and $`Q_\mathrm{\Delta }^i`$ are the corresponding gauge covariant symmetries, and $`ϵ_\mathrm{\Delta }`$ is the parity of $`j_\mathrm{\Delta }^\mu `$ (e.g., $`ϵ_\mathrm{\Delta }=1`$ when $`j_\mathrm{\Delta }^\mu `$ is the conserved current of a global supersymmetry). We stress again that the sets of $`Q`$’s and $`j`$’s are different in case I and case II, see text after Eq. (11.19).
4. $`W^{0,n}`$ are linear combinations of solutions $`W_\mathrm{\Gamma }`$ (11.33) with ghost number 0; such solutions exist only for peculiar choices of Lagrangians discussed below - and again only when there are abelian factors.
##### Nontriviality of the solutions.
A solution $`I^n+B^{0,n}+V^{0,n}+W^{0,n}`$ is only trivial when $`B^{0,n}`$, $`V^{0,n}`$ and $`W^{0,n}`$ all vanish and $`I^n`$ is weakly $`d`$-exact,
$$I^n+B^{0,n}+V^{0,n}+W^{0,n}0B^{0,n}=V^{0,n}=W^{0,n}=0,I^nd\omega ^{n1}.$$
$`I^nd\omega ^{n1}`$ is equivalent to $`I^ndI^{n1}+P(F)`$ for some $`I^{n1}`$ and some characteristic class $`P(F)`$.
##### Semisimple gauge group.
When the gauge group $`G`$ is semisimple there are no solutions $`V^{0,n}`$ or $`W^{0,n}`$ at all because all these solutions require the presence of abelian gauge symmetries. Hence, when $`G`$ is semisimple, all representatives of $`H^{0,n}(s|d)`$ can be taken to be strictly gauge invariant except for the Chern-Simons forms in odd spacetime dimensions. In particular, the antifields can then be removed from all BRST-invariant counterterms and integrated composite operators by adding cohomologically trivial terms, and the gauge transformations are stable, i.e., they cannot be deformed in a continuous and nontrivial manner. This result implies, in particular, the structural stability of effective Yang-Mills theories in the sense of .
##### Comment.
We add a comment on Chern-Simons forms which should also elucidate a bit the distinction between case I and case II. Chern-Simons forms $`B^{0,n}`$ are Lorentz-invariant in $`n`$-dimensional spacetime and occur thus among the solutions both in case I and in case II. However, these are not the only solutions constructible out of Chern-Simons forms. For instance, in 4 dimensions there is the solution $`\omega ^{0,4}=B^{0,3}dx^0`$ where $`B^{0,3}`$ is a Chern-Simons 3-form. This solution is not Lorentz-invariant and is thus present only in case I but not in case II. Have we missed this solution? The answer is “no” because it is equivalent to the solution $`I^4=x^0P(F)`$ where $`P(F)=dB^{0,3}`$. Namely we have $`B^{0,3}dx^0=d(x^0B^{0,3})+x^0dB^{0,3}`$ and thus indeed $`B^{0,3}dx^0x^0P(F)`$. Note that in order to establish this equivalence it is essential that we work in the space of local forms that may depend explicitly on the $`x^\mu `$.
##### The exceptional solutions $`𝑾^{\mathrm{𝟎}\mathbf{,}𝒏}`$.
The existence of a solution $`W^{0,n}`$ requires a relation $`k^iN_i=k^{A\alpha }P_A(F)\mathrm{\Theta }_\alpha `$ at ghost number 1, cf. Eq. (11.27). The $`N_i`$ with ghost number number 1 are linear combinations of terms $`(C^IF^JC^JF^I)P(F)`$ where $`C^I`$, $`C^J`$, $`F^I`$, $`F^J`$ are abelian and $`P(F)`$ is some characteristic class. The $`\mathrm{\Theta }_\alpha `$ with ghost number 1 are the abelian ghosts, and the $`P_A(F)`$ are characteristic classes which are trivial in the equivariant characteristic cohomology $`H_{\mathrm{char}}^n(d,)`$. Hence, in order that a solution $`W^{0,n}`$ exists, a nonvanishing linear combination of terms $`(C^IF^JC^JF^I)P(F)`$ must be equal to a linear combination of the $`P_A(F)C^I`$, where $`C^I`$, $`C^J`$, $`F^I`$, $`F^J`$ are abelian. The gauge group must therefore contain at least two abelian factors and, additionally, there must be at least two different $`P_A(F)`$ containing abelian field strengths. This is really a very special situation not met in practice (to our knowledge), which must be included in the discussion because we allow here for general Lagrangians. We illustrate it with a simple example:
$$n=4,L=\underset{I=1}{\overset{2}{}}\left[\frac{1}{4}F_{\mu \nu }^IF^{\mu \nu I}\frac{1}{2}(_\mu \varphi ^I)^\mu \varphi ^I+\frac{1}{4}\varphi ^Iϵ^{\mu \nu \rho \sigma }F_{\mu \nu }^IF_{\rho \sigma }^2\right]$$
where $`F_{\mu \nu }^I=_\mu A_\nu ^I_\nu A_\mu ^I`$ are abelian field strengths and $`\varphi ^1`$ and $`\varphi ^2`$ are real scalar fields. In this case we have $`\{P_A(F)\}=\{F^1F^2,F^2F^2\}`$ with corresponding $`\{I_A^3\}=\{d\varphi ^1,d\varphi ^2\}`$ and $`\{K_A\}=\{\varphi _1^{},\varphi _2^{}\}`$. Furthermore we have one $`N_\mathrm{\Gamma }`$ with ghost number 1 given by $`ϵ_{JI}C^IF^JF^2`$ and corresponding $`b_\mathrm{\Gamma }^4`$ given by $`(1/2)ϵ_{IJ}A^IA^JF^2`$ ($`ϵ_{IJ}=ϵ_{JI}`$). (11.33) gives now the following solution:
$$W^{0,4}=d^4x\underset{I,J=1}{\overset{2}{}}ϵ_{IJ}[\frac{1}{4}ϵ^{\mu \nu \rho \sigma }A_\mu ^IA_\nu ^JF_{\rho \sigma }^2+A_\mu ^I^\mu \varphi ^J\varphi _I^{}C^J].$$
(12.4)
### 12.3 $`H^{1,n}(s|d)`$: Anomalies
We know turn to $`H^{1,n}(s|d)`$, i.e., to anomalies. The most general solution of the consistency condition with ghost number 1 and form-degree $`n`$ is
$$\omega ^{1,n}\underset{I:\mathrm{abelian}}{}C^II_I^n+B^{1,n}+V^{1,n}+W^{1,n}$$
where:
1. $`I_I^n`$, i.e., $`\{I_I^n\}`$ is a set of strictly gauge invariant $`n`$-forms.
2. When $`n`$ is even, $`B^{1,n}`$ is a linear combination of the celebrated chiral anomalies listed in Eq. (10.41), except for those which contain abelian ghosts (the chiral anomalies with abelian ghosts are already included in $`_{I:\mathrm{abelian}}C^II_I^n`$). When $`n`$ is odd, $`B^{1,n}`$ is a linear combination of the solutions (10.43) which exist only when the gauge group contains at least two abelian factors.
3. $`V^{1,n}`$ are linear combinations of the solutions $`V_{\mathrm{\Delta }\alpha }`$ (11.21) with ghost number 1 related to global symmetries; in order that $`V_{\mathrm{\Delta }\alpha }`$ has ghost number 1, the $`\mathrm{\Theta }_\alpha `$ which appears in it must have ghost number 2 and must thus be a product of two abelian ghosts. Hence, solutions $`V^{1,n}`$ exist only if the gauge group contains at least two abelian factors. They are given by
$$V^{1,n}=\underset{I,J:\mathrm{abelian}}{}\lambda _{IJ}^\mathrm{\Delta }\left[K_\mathrm{\Delta }C^IC^J+j_\mathrm{\Delta }(A^IC^JA^JC^I)\right].$$
(12.5)
Using (12.1), the antifield dependence of $`V^{1,n}`$ can be made explicit, analogously to (12.3).
4. A discussion of Eq. (11.27) similar to the one performed for $`W^{0,n}`$ shows that the solutions $`W^{1,n}`$ are even more exceptional than their counterparts in ghost number zero; they exist only in the following situation: the gauge group must contain at least three abelian factors and, additionally, there must be at least three different $`P_A(F)`$ containing abelian field strengths. An example is the following:
$$n=4,L=\underset{I=1}{\overset{3}{}}\left[\frac{1}{4}F_{\mu \nu }^IF^{\mu \nu I}\frac{1}{2}(_\mu \varphi ^I)^\mu \varphi ^I+\frac{1}{4}\varphi ^Iϵ^{\mu \nu \rho \sigma }F_{\mu \nu }^IF_{\rho \sigma }^3\right]$$
where $`F_{\mu \nu }^I=_\mu A_\nu ^I_\nu A_\mu ^I`$ are abelian field strengths, and $`\varphi ^I`$ are real scalar fields. A solution $`W^{1,4}`$ is
$$W^{1,4}=d^4x\underset{I,J,K=1}{\overset{3}{}}ϵ_{IJK}[\frac{1}{4}ϵ^{\mu \nu \rho \sigma }C^IA_\mu ^JA_\nu ^KF_{\rho \sigma }^3+C^IA_\mu ^J^\mu \varphi ^K\frac{1}{2}C^IC^J\varphi _K^{}].$$
(12.6)
##### Nontriviality of the solutions.
A solution $`_{I:\mathrm{abelian}}C^II_I^n+B^{1,n}+V^{1,n}+W^{1,n}`$ is only trivial when $`B^{1,n}`$, $`V^{1,n}`$ and $`W^{1,n}`$ all vanish and, additionally,
$$\underset{I:\mathrm{abelian}}{}C^II_I^n\underset{I:\mathrm{abelian}}{}C^IdI_I^{n1}+\underset{I,J:\mathrm{abelian}}{}(C^IF^JC^JF^I)P_{IJ}(F)$$
for some $`I_I^{n1}`$ and some characteristic classes $`P_{IJ}(F)`$.
##### Semisimple gauge group.
Note that all nontrivial solutions $`\omega ^{1,n}`$ involve abelian ghosts, except for the solutions $`B^{1,n}`$ in even dimensions. Hence, when the gauge group is semisimple, the candidate gauge anomalies are exhausted by the well-known nonabelian chiral anomalies in even dimensions. These live in the small algebra and can be obtained from the characteristic classes living in two dimensions higher through the Russian formula of section 10.6. Furthermore, these anomalies are in finite number (independently of power counting arguments) and do not depend on the specific form of the Lagrangian. For some groups, there may be none (“anomaly-safe groups”), in which case the consistency condition implies absence of anomalies, for any Lagrangian.
In $`4`$ dimensions, $`B^{1,4}`$ is the non abelian gauge anomaly :
$$B^{1,4}=d_{IJK}C^Id[A^JdA^K+\frac{1}{4}ef_{LM}^{}{}_{}{}^{J}A^KA^LA^M],$$
(12.7)
where $`d_{IJK}`$ is the general symmetric $`𝒢`$-invariant tensor. Hence, the gauge group is anomaly safe for any Lagrangian in 4 dimensions if there is no $`d_{IJK}`$-tensor.
### 12.4 The cohomological groups $`H^{g,n}(s|d)`$ with $`g>1`$
The results for ghost numbers $`g>1`$ are similiar to those for $`g=0`$ and $`g=1`$; one gets
$$\omega ^{g,n}I^{n\alpha _g}\mathrm{\Theta }_{\alpha _g}+B^{g,n}+V^{g,n}+W^{g,n}$$
where $`\{\mathrm{\Theta }_{\alpha _g}\}`$ is the subset of $`\{\mathrm{\Theta }_\alpha \}`$ containing those $`\mathrm{\Theta }`$’s with ghost number $`g`$. Of course, this subset depends on the gauge group $`G`$. It depends both on $`G`$ and on the spacetime dimension which solutions are present for given $`g`$. For instance, for $`G=SU(2)`$ and $`n=4`$, one has $`\omega ^{2,4}V^{2,4}`$. The highest ghost number for which nontrivial solutions exist is $`g=\mathrm{dim}(G)`$ because this is the ghost number of the product of all $`\theta _r(C)`$. Antifield dependent solutions $`V^{g,n}`$ exist up to ghost number $`g=\mathrm{dim}(G)1`$. As we have mentioned already several times, solutions $`W^{g,n}`$ exist only for exceptional Lagrangians.
### 12.5 Appendix 12.A: Gauge covariance of global symmetries
(12.1) and (12.2) can be achieved because the equations of motion are gauge covariant. This is seen as follows. Consider an $`n`$-form $`I`$ which vanishes weakly, $`I0`$. This is equivalent to $`I=\delta \widehat{K}`$ for some $`n`$-form $`\widehat{K}`$. The equations of motion are gauge covariant in the sense that one has $`\delta A_I^\mu =L_I^\mu (x,[F,\psi ]_D)`$ and $`\delta \psi _i^{}=L_i(x,[F,\psi ]_D)`$ where the $`L_I^\mu (x,[F,\psi ]_D)`$ are in the adjoint representation of $`𝒢`$ and the $`L_i(x,[F,\psi ]_D)`$ in the representation dual to the representation of the $`\psi ^i`$ (cf. section 8.1). In particular, $`\delta `$ is stable in the space of $`𝒢`$-invariant functions $`f_{\mathrm{inv}}(x,[F,\psi ,A^{},\psi ^{},C^{}]_D)`$, i.e., it maps this space into itself. We can thus choose
$`\widehat{K}`$ $`=`$ $`d^nx[A_I^\mu \widehat{Q}_\mu ^I(x,[F,\psi ]_D)+\psi _i^{}\widehat{Q}^i(x,[F,\psi ]_D)`$
$`+(D_\nu A_I^\mu )\widehat{Q}_\mu ^{I\nu }(x,[F,\psi ]_D)+(D_\nu \psi _i^{})\widehat{Q}^{i\nu }(x,[F,\psi ]_D)+\mathrm{}]`$
where $`\widehat{K}`$ is $`𝒢`$-invariant. Note that $`\widehat{K}`$ contains in general covariant derivatives of antifields. To deal with these terms, we write
$`\widehat{K}`$ $`=`$ $`d^nx[A_I^\mu Q_\mu ^I+\psi _i^{}Q^i+D_\nu R^\nu ],`$
$`Q_\mu ^I`$ $`=`$ $`\widehat{Q}_\mu ^ID_\nu \widehat{Q}_\mu ^{I\nu }+\mathrm{},`$
$`Q^i`$ $`=`$ $`\widehat{Q}^iD_\nu \widehat{Q}^{i\nu }+\mathrm{},`$
$`R^\nu `$ $`=`$ $`A_I^\mu \widehat{Q}_\mu ^{I\nu }+\psi _i^{}\widehat{Q}^{i\nu }+\mathrm{}.`$
Since $`R^\nu `$ is $`𝒢`$-invariant, we have $`D_\nu R^\nu =_\nu R^\nu `$ and thus
$`\widehat{K}`$ $`=`$ $`K+dR,`$
$`K`$ $`=`$ $`d^nx[A_I^\mu Q_\mu ^I+\psi _i^{}Q^i],`$
$`R`$ $`=`$ $`()^n\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}R^{\mu _n}.`$
Using this in $`I=\delta \widehat{K}`$, we get
$$I+d\delta R=sK$$
because the $`𝒢`$-invariance of $`K`$ implies $`\delta K=sK`$.
(12.1) follows by setting $`I=dj`$ where $`j`$ is a conserved current (we have $`dj=Dj`$). Namely the above formula gives in this case $`d(j\delta R)=sK`$. Note that $`j\delta R`$ is equivalent to $`j`$ and gauge invariant (due to $`\delta R0`$ and $`\delta R`$). Hence we can indeed choose the basis $`\{j_\mathrm{\Delta }\}`$ of the inequivalent gauge invariant currents such that $`dj_\mathrm{\Delta }=sK_\mathrm{\Delta }`$ with $`K_\mathrm{\Delta }`$ as in (12.1).
Now consider the equation $`P(F)dI^{n1}`$ with $`I^{n1}`$. Setting $`I=P(F)dI^{n1}`$, the above formula gives $`P(F)d(I^{n1}\delta R)=sK`$. Hence, we can choose all $`I_A^{n1}`$ such that $`P_A(F)=sK_A+dI_A^{n1}`$ with $`K_A`$ as in (12.2).
## 13 Free abelian gauge fields
### 13.1 Pecularities of free abelian gauge fields
We now compute the local BRST cohomology for a set of $`R`$ abelian gauge fields with a free Lagrangian of the Maxwell type,
$$L=\frac{1}{4}\underset{I=1}{\overset{R}{}}F_{\mu \nu }^IF^{\mu \nu I},F_{\mu \nu }^I=_\mu A_\nu ^I_\nu A_\mu ^I.$$
(13.1)
This question is relevant for determining the possible consistent interactions that can be defined among massless vector particles, where both groups $`H^{0,n}`$ and $`H^{1,n}`$ play a rôle, as we shall discuss in subsection 13.3 below.
As we have already mentioned, theorem 11.1 does not hold for (13.1). The reason is that the characteristic cohomology group $`H_{\mathrm{char}}^{n2}(d,\mathrm{\Omega })`$ does not vanish in the free model. Rather, by theorem 6.8, this cohomological group is represented in all spacetime dimensions $`n>2`$ by the Hodge-duals of the abelian curvature 2-forms $`F^I=dA^I`$,
$$F^I=\frac{1}{(n2)!\mathrm{\hspace{0.17em}2}}dx^{\mu _1}\mathrm{}dx^{\mu _{n2}}ϵ_{\mu _1\mathrm{}\mu _n}F^{\mu _{n1}\mu _nI}.$$
(13.2)
This modifies the results for the form-degrees $`p=n1`$ and $`p=n`$ as compared to theorem 11.1, by allowing solutions of a new type. These solutions are precisely those that appear in the non Abelian deformation of (13.1).
In contrast, the results for lower form-degrees remain valid as an inspection of the proof of the theorem shows since one still has $`H_{\mathrm{char}}^p(d,\mathrm{\Omega })=\delta _0^p`$ for $`p<n2`$.
The discussion of this section applies also to abelian gauge fields with self-couplings involving the curvature only (like in the Born-Infeld Lagrangian), or in the case of non-minimal interactions with matter through terms involving only the field strength (e.g., $`F_{\mu \nu }\overline{\psi }\gamma ^{[\mu }\gamma ^{\nu ]}\psi `$, where $`\psi `$ is a Dirac spinor). In that case, the matter fields do not transform under the abelian gauge symmetry so that the global reducibility identities behind theorem 6.8 are still present.
### 13.2 Results
We shall now work out the modifications for form-degrees $`p=n1`$ and $`p=n`$, assuming the spacetime dimension $`n`$ to be greater than $`2`$.
#### 13.2.1 Results in form-degree $`p=n1`$
Let $`\omega ^{n1}`$ be a cocycle of $`H^{,n1}(s|d,\mathrm{\Omega })`$,
$$s\omega ^{n1}+d\omega ^{n2}=0.$$
(13.3)
The same arguments as in the proof of theorem 11.1 until Eq. (11.48) included yield $`\omega ^{n2}=I^{n2\alpha }\mathrm{\Theta }_\alpha `$ where (i) $`I^{n2\alpha }`$ is gauge-invariant and fulfills $`dI^{n2\alpha }0`$; and (ii) the $`\mathrm{\Theta }_\alpha `$ form a basis of polynomials in the undifferentiated ghosts (in the purely abelian case each ghost polynomial is invariant; furthermore, Eq. (11.48) has no solution $`B^{n2}`$ in the purely abelian case because the $`B`$’s cannot be lifted, see below). The condition $`dI^{n2\alpha }0`$ gives now $`I^{n2\alpha }\lambda _I^\alpha F^I+P^{n2\alpha }(F)+dI^{n3\alpha }`$ where the linear combination $`\lambda _I^\alpha F^I`$ of the $`F^I`$ comes from $`H_{\mathrm{char}}^{n2}(d,\mathrm{\Omega })`$. It is here that the extra characteristic cohomology enters and gives extra terms in $`I^{n2\alpha }`$ compared with Eq. (11.49). These extra terms fulfill
$$dF^I=sA^I$$
(13.4)
where $`A^I`$ is the antifield dependent $`(n1)`$-form
$$A^I=\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}A_J^{\mu _n}\delta ^{JI}.$$
(13.5)
It is then straightforward to adapt Eqs. (11.50) through (11.53). This gives, instead of Eq. (11.53),
$$\omega ^{n1}\lambda _I^\alpha (A^I\mathrm{\Theta }_\alpha +(F^I)[\mathrm{\Theta }_\alpha ]^1)+B^{n1}+I^{n1\alpha }\mathrm{\Theta }_\alpha .$$
Since we are dealing with a purely abelian case, the $`\mathrm{\Theta }_\alpha `$ are just products of the undifferentiated ghosts. We can therefore write the result, up to trivial solutions, as
$`\omega ^{n1}A^IP_I(C)+(F^I)A^J_JP_I(C)+B^{n1}+I^{n1\alpha }P_\alpha (C)`$ (13.6)
where $`P_I(C)`$ and $`P_\alpha (C)`$ are arbitrary polynomials in the undifferentiated ghosts, and
$$_I\frac{}{C^I}.$$
Furthermore, the descent is particularly simple in the small algebra because a non trivial bottom can be lifted only once; at the next step, one meets an obstruction. This implies that the solutions $`B^{n1}`$ can occur in (13.6) only when the spacetime dimension $`n`$ is even: Eq. (11.16) gives in the purely abelian case only solutions with odd form-degrees, which are linear in the one-forms $`A^I`$,
$`B^{2N+1}`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{K}{}}}\lambda _{I_1\mathrm{}I_KJ_1\mathrm{}J_N}C^{I_1}\mathrm{}C^{I_{i1}}A^{I_i}C^{I_{i+1}}\mathrm{}C^{I_K}F^{J_1}\mathrm{}F^{J_N}`$
$`B^{2k}`$ $`=`$ $`0`$ (13.7)
where $`I_i<I_{i+1}`$, $`J_iJ_{i+1}`$, and (if $`N>0`$) $`I_1J_1`$. These solutions descend on the gauge-invariant term $`\lambda _{I_1\mathrm{}I_KJ_1\mathrm{}J_N}C^{I_1}\mathrm{}C^{I_K}F^{J_1}\mathrm{}F^{J_N}`$.
Eq. (13.6) gives the general solution of (13.3), both in the space of all local forms (case I) and in the space of Poincaré invariant local forms (case II), with $`I^{n1\alpha }`$ where $``$ is the respective gauge invariant subspace of local forms,
Case I: $`=\{\text{local functions of }F_{\mu \nu }^I\text{}_\rho F_{\mu \nu }^I\text{, …}\}\mathrm{\Omega }(^n)`$ (13.8)
Case II: $`=\{\text{Lorentz-invariant local functions of }dx^\mu \text{}F_{\mu \nu }^I\text{}_\rho F_{\mu \nu }^I\text{, …}\}.`$ (13.9)
\[As before, $`\mathrm{\Omega }(^n)`$ denotes the space of ordinary differential forms in $`^n`$.\]
#### 13.2.2 Results in form-degree $`p=n`$
Let $`\omega ^n`$ be a cocycle of $`H^{,n}(s|d,\mathrm{\Omega })`$,
$$s\omega ^n+d\omega ^{n1}=0.$$
(13.10)
By the standard arguments of the descent equation technique, $`\omega ^{n1}`$ is a cocycle of $`H^{,n1}(s|d,\mathrm{\Omega })`$ and trivial contributions to $`\omega ^{n1}`$ can be neglected without loss of generality. Hence, $`\omega ^{n1}`$ can be assumed to be of the form (13.6) and we have to analyse the restrictions imposed on it by the fact that it can be lifted once to give $`\omega ^n`$ through (13.10). To this end we compute $`d\omega ^{n1}`$.
To deal with the first two terms in (13.6), we use once again (13.4) as well as
$$dA^I=sC^I$$
(13.11)
where
$$C^I=d^nxC_J^{}\delta ^{JI}.$$
(13.12)
This yields<sup>18</sup><sup>18</sup>18Eq. (13.13) can be elegantly derived using the quantities $`\stackrel{~}{C}^I=C^I+A^I+F^I`$ and $`\stackrel{~}{C}^I=C^I+A^I`$. One has $`(s+d)\stackrel{~}{C}^I=0`$ and $`(s+d)\stackrel{~}{C}^I=F^I`$. This implies $`(s+d)[\stackrel{~}{C}^IP_I(\stackrel{~}{C})]=()^n(\stackrel{~}{C}^I)F^J_JP_I(C)`$ whose $`n`$-form part is Eq. (13.13).
$`d[A^IP_I(C)+(F^I)A^J_JP_I(C)]=()^n(F^I)F^J_JP_I(C)`$
$`s[C^IP_I(C)+(A^I)A^J_JP_I(C)+\frac{1}{2}(F^I)A^JA^K_K_JP_I(C)].`$ (13.13)
The remaining terms in (13.6) are dealt with as in the proof of part (i) of theorem 11.1. One gets
$$d[B^{n1}+I^{n1\alpha }P_\alpha (C)]=s[b^n+I^{n1\alpha }A^I_IP_\alpha (C)]+N^n+(dI^{n1\alpha })P_\alpha (C)$$
(13.14)
where $`N^n`$ is in the small algebra (it is an obstruction to a lift in the small algebra, see corollary 10.4 and theorem 11.1). Using Eqs. (13.13) and (13.14) in Eq. (13.10), one obtains
$$s(\omega ^n\mathrm{})=()^nF^JF^I_JP_I(C)+N^n+(dI^{n1\alpha })P_\alpha (C).$$
(13.15)
The right hand side of (13.15) has zero antighost number and does not contain the derivatives of the ghosts. Due to $`\gamma A_\mu ^I=_\mu C^I`$ and $`\gamma C^I=\gamma A_I^\mu =\gamma C_I^{}=0`$, $`\gamma (\omega ^n\mathrm{})`$ is a sum of field monomials each of which contains derivatives of the ghosts (unless it vanishes). Hence, (13.15) implies that the term on the right hand side is $`\delta `$-exact, i.e., that it vanishes weakly,
$$()^n(F^I)F^J_JP_I(C)+N^n+(dI^{n1\alpha })P_\alpha (C)0.$$
(13.16)
To analyse this condition, we must distinguish cases I and II.
We treat first case II, i.e. the space of Poincaré-invariant local forms, for which the analysis can be pushed to the end. In this case we have $`I^{n1\alpha }`$ with $``$ as in (13.9). Hence, $`dI^{n1\alpha }`$ is a sum of field monomials each of which contains a first or higher order derivative of at least one of the $`F_{\mu \nu }^I`$. Furthermore the equations of motion contain at least first order derivatives of the $`F_{\mu \nu }^I`$. Hence, in case II, the part of Eq. (13.16) which contains only undifferentiated $`F_{\mu \nu }^I`$ reads $`()^nF^JF^I_JP_I(C)+N^n=0`$.<sup>19</sup><sup>19</sup>19The same argument yields $`P(F)dI,IP(F)=0`$ in case II. This implies that both $`F^JF^I_JP_I(C)`$ and $`N^n`$ vanish since $`N^n`$ contains the $`F_{\mu \nu }^I`$ only via wedge products of the $`F^I`$ (all wedge products of the $`F^I`$ are total derivatives while no $`F^JF^I`$ is a total derivative). Eq. (13.16) yields thus
$$\text{Case II:}F^JF^I_JP_I(C)=0,N^n=0,(dI^{n1\alpha })P_\alpha (C)0.$$
(13.17)
Since $`F^JF^I=\frac{1}{2}d^nxF_{\mu \nu }^JF^{\mu \nu I}`$ is symmetric in $`I`$ and $`J`$, the first condition in (13.17) gives
$$\text{Case II:}_IP_J(C)+_JP_I(C)=0.$$
(13.18)
The general solution of Eq. (13.18) is obtained from the cohomology $`H(D,𝒞)`$ of the differential $`D=\xi ^I_I`$ in the space $`𝒞`$ of polynomials in commuting extra variables $`\xi ^I`$ and anticommuting variables $`C^I`$. Indeed, by contracting Eq. (13.18) with $`\xi ^I\xi ^J`$, it reads $`Da=0`$ where $`a=\xi ^IP_I(C)`$. Using the contracting homotopy $`\varrho =C^I/\xi ^I`$ (see appendix 2.7), one easily proves that $`H(D,𝒞)`$ is represented solely by pure numbers (“Poincaré lemma for $`D`$”; note that $`D`$ is similar to $`dx^\mu /x^\mu `$ except that the “differentials” $`\xi ^I`$ commute while the “coordinates” $`C^I`$ anticommute). In particular this implies that $`Da=0a=DP(C)`$ for $`a=\xi ^IP_I(C)`$, i.e.,
$$\text{Case II:}P_I(C)=_IP(C)$$
(13.19)
for some polynomial $`P(C)`$ in the $`C^I`$. The analysis of Eq. (13.10) can now be finished along the lines of the proof of theorem 11.1. One obtains in case II that the general solution of Eq. (13.10) is, up to trivial solutions, given by
Case II: $`\omega ^n=[C^I_I+(A^I)A^J_J_I+\frac{1}{2}(F^I)A^JA^K_K_J_I]P(C)`$ (13.20)
$`+B^n+I^{n\alpha }P_\alpha (C)+[K_\mathrm{\Delta }+j_\mathrm{\Delta }A^I_I]P^\mathrm{\Delta }(C)`$
where $`P(C)`$, $`P_\alpha (C)`$ and $`P^\mathrm{\Delta }(C)`$ are arbitrary polynomials in the undifferentiated ghosts, $`B^n`$ occurs only in odd dimensional spacetime due to (13.7), $`j_\mathrm{\Delta }`$ are gauge-invariant and Poincaré-invariant conserved $`(n1)`$-forms, see Eqs. (11.17) through (11.19), and $`K_\mathrm{\Delta }`$ contains the global symmetry corresponding to $`j_\mathrm{\Delta }`$ and satisfies $`sK_\mathrm{\Delta }+dj_\mathrm{\Delta }=0`$. There are no solutions $`W_\mathrm{\Gamma }`$ because all characteristic classes $`P(F)`$, including those with form-degree $`n`$, are nontrivial in the equivariant characteristic cohomology $`H_{\mathrm{char}}(d,)`$ with $``$ as in (13.9) (cf. footnote 19).
Let us finally discuss case I, i.e., the space of all local forms (with a possible, explicit $`x`$-dependence). In this case Eq. (13.16) holds for $`I^{n1\alpha }`$ with $``$ as in (13.8). In contrast to case II, $`dI^{n1\alpha }`$ may thus contain field monomials which involve only undifferentiated $`F_{\mu \nu }^I`$ because $`I^{n1\alpha }`$ may depend explicitly on the spacetime coordinates $`x^\mu `$. Therefore the arguments that have led us to Eq. (13.19) do not apply in case I. In fact one finds in all spacetime dimensions $`n4`$ that Eq. (13.19) need not hold in case I. Rather, if $`n4`$, (13.10) does not impose any restriction on the $`P_I(C)`$ at all, i.e., the terms in $`\omega ^{n1}`$ related to the $`P_I(C)`$ can be lifted to a solution $`\omega ^n`$ for any set $`\{P_I(C)\}`$. This solution is $`d^nxa`$ where
$`a`$ $`=`$ $`(C^I+A^{\mu I}A_\mu ^J_J\frac{1}{2}F^{\mu \nu I}A_\mu ^JA_\nu ^K_K_J)P_I(C)`$ (13.21)
$`+\frac{2}{n4}F_{\mu \nu }^I(x^\mu A^{\nu J}+x^\mu F^{\rho \nu J}A_\rho ^K_K+\frac{1}{4}F^{\mu \nu J}x^\rho A_\rho ^K_K)_{(I}P_{J)}(C).`$
$`d^nxa`$ fulfills (13.19), i.e., $`s(d^nxa)+da^{n1}=0`$ where $`a^{n1}`$ is indeed of the form (13.6),
$`a^{n1}`$ $`=`$ $`[A^I+(F^I)A^J_J]P_I(C)+\frac{1}{(n1)!}dx^{\mu _1}\mathrm{}dx^{\mu _{n1}}ϵ_{\mu _1\mathrm{}\mu _n}I^{\mu _n},`$
$`I^\mu `$ $`=`$ $`\frac{2}{n4}\left(\frac{1}{4}x^\mu F_{\nu \rho }^IF^{\nu \rho J}+F^{\mu \nu I}F_{\nu \rho }^Jx^\rho \right)_{(I}P_{J)}(C).`$
Note that both $`a`$ and $`a^{n1}`$ depend explicitly on $`x^\mu `$ and are therefore present only in case I but not in case II, except when $`_{(I}P_{J)}(C)=0`$ (then $`d^nxa`$ reproduces the first line of (13.20). When one multiplies $`a`$ by $`n4`$, one gets solutions for all $`n`$. For $`n=4`$, they become solutions of the form $`[K_\mathrm{\Delta }+j_\mathrm{\Delta }A^I_I]P^\mathrm{\Delta }(C)`$ with gauge invariant Noether currents $`j_\mathrm{\Delta }`$ involving explicitly the $`x^\mu `$. One may now proceed along the previous lines. However, two questions remain open in case I: Does (13.10) impose restrictions on the $`P_I(C)`$ when $`n=4`$ ? Are there characteristic classes $`P(F)`$ with form-degree $`n`$ which are trivial in the equivariant characteristic cohomology $`H_{\mathrm{char}}(d,)`$ with $``$ as in (13.8) ? (See also footnote 17.)
### 13.3 Uniqueness of Yang-Mills cubic vertex
We now use the above results to discuss the consistent deformations of the action (13.1). Requiring that the interactions be Poincaré invariant, the relevant results are those of case II.
As shown in , the consistent deformations of an action are given, to first order in the deformation parameter, by the elements of $`H^{0,n}(s|d)`$, i.e., here, from (13.20),
$`\omega ^{0,n}=[C^I_I+(A^I)A^J_J_I+\frac{1}{2}(F^I)A^JA^K_K_J_I]P(C)`$
$`+B^n+I^n+[K_\mathrm{\Delta }+j_\mathrm{\Delta }A^I_I]P^\mathrm{\Delta }(C)`$ (13.22)
where $`P(C)`$ has ghost number $`3`$ and $`P^\mathrm{\Delta }(C)`$ ghost number one ($`P_\alpha (C)`$ in (13.20) has ghost number zero and thus is a constant; this has been taken into account in (13.22)).
The term $`B^n`$ is the familiar Chern-Simons term , and exists only in odd dimensions. It belongs to the small algebra and is of the form $`AF\mathrm{}F`$. The term $`I^n`$ is strictly gauge-invariant and thus involves the abelian field strengths and their derivatives. Born-Infeld or Euler-Heisenberg deformations are of this type. Since these terms are well understood and do not affect the gauge symmetry, we shall drop them from now on and focus on the other two terms, which are,
$$[C^I_I+(A^I)A^J_J_I+\frac{1}{2}(F^I)A^JA^K_K_J_I]P(C)$$
(13.23)
and
$$[K_\mathrm{\Delta }+j_\mathrm{\Delta }A^I_I]P^\mathrm{\Delta }(C)$$
(13.24)
Expression (13.23) involves the antifields conjugate to the ghosts, while (13.24) involves only the antifields conjugate to $`A^I`$.
Now, it has also been shown in (see for further details) that deformations involving nontrivially the antifields do deform the gauge symmetries. Those that involve the antifields conjugate to the ghosts deform not only the gauge transformations but also their algebra; while those that involve only $`A_I^{}`$ modify the gauge transformations but leave the gauge algebra unchanged (at least to first order in the deformation parameter).
Writing $`P(C)=(1/3!)f_{IJK}C^IC^JC^K`$ (with $`f_{IJK}`$ completely antisymmetric), one gets from (13.23) that the deformations of the theory that deform the gauge algebra are given by
$`\frac{1}{2}C^If_{IJK}C^JC^K+(A^I)A^Jf_{IJK}C^K+\frac{1}{2}(F^I)A^JA^Kf_{IJK}`$
$`=d^nx(\frac{1}{2}f_{IJK}C^KC^JC^I+f_{IJK}A_\mu ^JC^KA^{\mu I}+\frac{1}{2}f_{IJK}F^{\mu \nu I}A_\mu ^JA_\nu ^K).`$ (13.25)
The term independent of the antifields is the first order deformation of the action, and one recognizes the standard Yang-Mills cubic vertex – except that the $`f_{IJK}`$ are not subject to the Jacobi identity at this stage. This condition arises, however, when one investigates consistency of the deformation to second order: the deformation is obstructed at second order by a non trivial element of $`H^{1,n}(s|d)`$ unless $`_K(f_{IJK}f_{KLM}+f_{JLK}f_{KIM}+f_{LIK}f_{KJM})=0`$ . The obstruction is precisely of the type (13.23), with $`P(C)=(1/36)_Kf_{IJK}f_{KLM}C^IC^JC^LC^M`$. The field monomials in front of the antifields in (13.25) give the deformations of the BRST transformations of the ghosts and gauge fields respectively (up to a minus sign) and provide thus the nonabelian extension of the abelian gauge transformations and their algebra. Thus, one recovers the known fact that the Yang-Mills construction provides the only deformation of the action for a set of free abelian gauge fields that deforms the algebra of the gauge transformations at first order in the deformation parameter. Any consistent interaction which deforms nontrivially the gauge algebra at first order contains therefore the Yang-Mills vertex. Furthermore, this deformation automatically incorporates the Lie algebra structure underlying the Yang-Mills theory, without having to postulate it a priori. This result has been derived recently in , along different lines und under stronger assumptions on the form of the new gauge symmetries. These extra assumptions are in fact not necessary as the cohomological derivation shows.
Having dealt with the deformations (13.23), we can turn to the deformations (13.24), which do not deform the (abelian) gauge algebra at first order although they do deform the gauge transformations. These involve Lorentz covariant and gauge invariant conserved currents $`j_\mathrm{\Delta }^\mu `$. An example of a deformation of this type is given by the Freedman-Townsend vertex in three dimensions . In four dimensions, however, the results of indicate that there is no (non trivial) candidate for $`j_\mathrm{\Delta }^\mu `$. There is an infinite number of conservation laws because the theory is free, but these do not involve gauge invariant Lorentz vectors. Thus, there is no Poincaré invariant deformation of the type (13.24) in four dimensions. This strengthens the above result on the uniqueness of the Yang-Mills cubic vertex, which is the only vertex deforming the gauge transformations in four dimensions. Accordingly, in four dimensions, the most general deformation of the action for a set of free abelian gauge fields is given, at first order, by the Yang-Mills cubic vertex and by strictly gauge invariant deformations. We do not know whether the results of generalize to higher dimensions, leaving the (unlikely in our opinion) possibility of the existence of interactions of the type (13.24) in $`n>4`$ dimensions, which would deform the gauge transformations without modifying the gauge algebra at first order in the deformation parameter.
## 14 Three-dimensional Chern-Simons theory
### 14.1 Introduction – $`H(s)`$
We shall now describe the local BRST cohomology in 3-dimensional pure Chern-Simons theory with general gauge group $`G`$, i.e., $`G`$ may be abelian, semisimple, or the direct product of an abelian and a semisimple part. Since pure Chern-Simons theory is of the Yang-Mills type, theorem 11.1 applies to it when the gauge group is semisimple. So, the Chern-Simons case is not really special from this point of view. However, the results are particularly simple in this case because the theory is topological. As we shall make it explicit below, there is no non-trivial local, gauge-invariant function and the BRST cohomology reduces to the Lie algebra cohomology with coefficients in the trivial representation.
It is because of this, and because of the physical interest of the Chern-Simons theory, that we devote a special section to it.
The Chern-Simons action is
$$S_{\mathrm{CS}}=g_{IJ}[\frac{1}{2}A^IdA^J+\frac{1}{6}ef_{KL}^{}{}_{}{}^{I}A^JA^KA^L].$$
(14.1)
where $`g_{IJ}=\delta _{IJ}`$ for the abelian part of $`G`$ and $`g_{IJ}=f_{IK}^{}{}_{}{}^{L}f_{JL}^{}{}_{}{}^{K}`$ for the nonabelian part. This yields explicitly
$`sA_I^\mu `$ $`=`$ $`\frac{1}{2}g_{IJ}ϵ^{\mu \nu \rho }F_{\nu \rho }^J+eC^Jf_{JI}^{}{}_{}{}^{K}A_K^\mu ,`$
$`sC_I^{}`$ $`=`$ $`D_\mu A_I^\mu +eC^Jf_{JI}^{}{}_{}{}^{K}C_K^{}.`$ (14.2)
Again we shall determine $`H(s,\mathrm{\Omega })`$ both in the space of all local forms (case I) and in the space of Poincaré-invariant local forms (case II).
We first specify $`H(s,\mathrm{\Omega })`$ in these cases, using the results of Section 8. Since in pure Chern-Simons theory the field strengths vanish weakly and do not contribute to $`H(s,\mathrm{\Omega })`$ at all, we find from this section that in case I $`H(s,\mathrm{\Omega })`$ is represented by polynomials in the $`\theta _r(C)`$ which can also depend explicitly on the spacetime coordinates $`x^\mu `$ and the differentials $`dx^\mu `$,
$$\text{Case I:}s\omega =0\omega =P(\theta (C),x,dx)+s\eta .$$
(14.3)
In case II, a similar result holds, but now no $`x^\mu `$ can occur and Lorentz invariance enforces that the differentials can contribute nontrivially only via the volume form $`d^3x`$,
$$\text{Case II:}s\omega =0\omega =Q(\theta (C))+d^3xP(\theta (C))+s\eta .$$
(14.4)
(14.3) and (14.4) provide the solutions of the consistency condition with a trivial descent. They also yield the bottom forms which can appear in nontrivial descents. To find all solutions with a nontrivial descent, we investigate how far these bottom forms can be lifted to solutions with higher form-degree.
### 14.2 BRST cohomology in the case of $`x`$-dependent forms
In order to lift a bottom form $`P(\theta (C),x,dx)`$ once, it is necessary and sufficient that $`d_xP(\theta (C),x,dx)=0`$ where $`d_x=dx^\mu /x^\mu `$ (this is nothing but Eq. (11.14), specified to $`P(\theta (C),x,dx)=I^\alpha (x,dx)\mathrm{\Theta }_\alpha `$). $`d_xP(\theta (C),x,dx)=0`$ implies $`P(\theta (C),x,dx)=Q(\theta (C))+d_xP^{}(\theta (C),x,dx)`$ by the ordinary Poincaré lemma, and thus $`P(\theta (C),x,dx)=Q(\theta (C))+dP^{}(\theta (C),x,dx)+s[A^I_IP^{}(\theta (C),x,dx)]`$ where we used once again Eq. (11.13) and the notation
$$_I=\frac{}{C^I}.$$
The pieces $`dP^{}(\mathrm{})+s[A^I_IP^{}(\mathrm{})]`$ are trivial and can thus be neglected without loss of generality. Hence, all bottom forms which can be lifted once can be assumed to be of the form $`Q(\theta (C))`$. Furthermore there are no obstructions to lift these bottom forms to higher form-degrees. An elegant way to see this and to construct the corresponding solutions at higher form-degree is the following. We introduce
$$𝒞^I=C^I+A^I+A^I+C^I,$$
(14.5)
where
$$A^I=\frac{1}{2}dx^\mu dx^\nu ϵ_{\mu \nu \rho }g^{IJ}A_J^\rho ,C^I=d^3xg^{IJ}C_J^{},$$
(see also for instance ). Using (14.2), one verifies
$$(s+d)𝒞^I=\frac{1}{2}ef_{JK}^{}{}_{}{}^{I}𝒞^K𝒞^J.$$
(14.6)
Hence, $`(s+d)`$ acts on the $`𝒞^I`$ exactly as $`s`$ acts on the $`C^I`$. This implies $`(s+d)\theta _r(𝒞)=0`$ (which is analogous to $`s\theta _r(C)=0`$) and thus
$$(s+d)Q(\theta (𝒞))=0.$$
(14.7)
This equation decomposes into the descent equations $`s[Q]^3+d[Q]^2=0`$, $`s[Q]^2+d[Q]^1=0`$, $`s[Q]^1+d[Q]^0=0`$, $`s[Q]^0=0`$ where $`[Q]^p`$ is the $`p`$-form contained in $`Q(\theta (𝒞))`$,
$`Q(\theta (𝒞))`$ $`=`$ $`{\displaystyle \underset{p=0}{\overset{3}{}}}[Q]^p`$
$`[Q]^0`$ $`=`$ $`Q(\theta (C))`$
$`[Q]^1`$ $`=`$ $`A^I_IQ(\theta (C))`$
$`[Q]^2`$ $`=`$ $`[\frac{1}{2}A^IA^J_J_I+A^I_I]Q(\theta (C))`$
$`[Q]^3`$ $`=`$ $`[\frac{1}{6}A^IA^JA^K_K_J_I+A^IA^J_J_I+C^I_I]Q(\theta (C)).`$ (14.8)
We conclude that the general solution of the consistency condition is at the various form-degrees given by
$`\text{Case I:}H^{,0}(s|d,\mathrm{\Omega }):`$ $`\omega ^0P(\theta (C),x)`$
$`H^{,1}(s|d,\mathrm{\Omega }):`$ $`\omega ^1[Q]^1+dx^\mu P_\mu (\theta (C),x)`$
$`H^{,2}(s|d,\mathrm{\Omega }):`$ $`\omega ^2[Q]^2+dx^\mu dx^\nu P_{\mu \nu }(\theta (C),x)`$
$`H^{,3}(s|d,\mathrm{\Omega }):`$ $`\omega ^3[Q]^3`$ (14.9)
where one can assume that $`dx^\mu P_\mu (\theta (C),x)`$ and $`dx^\mu dx^\nu P_{\mu \nu }(\theta (C),x)`$ are not $`d_x`$-closed because otherwise they are trivial by the arguments given above. For the same reason every contribution $`d^3xP(\theta (C),x)`$ to $`\omega ^3`$ is trivial (it is a volume form and thus automatically $`d_x`$-closed) and has therefore not been written in (14.9).
### 14.3 BRST cohomology in the case of Poincaré invariant forms
This case is easy. By Eq. (14.4), all nontrivial bottom forms that can appear in case II are either 0-forms $`Q(\theta (C))`$ or volume-forms $`d^3xP(\theta (C))`$. We know already that the former can be lifted to the above Poincaré-invariant solutions $`[Q]^1`$, $`[Q]^2`$ and $`[Q]^3`$. Hence, we only need to discuss the volume-forms $`d^3xP(\theta (C))`$. They are nontrivial in case II, in contrast to case I, except for the banal case that $`P(\theta )`$ vanishes identically. Indeed, assume that $`d^3xP(\theta (C))`$ is trivial, i.e., that $`d^3xP(\theta (C))=s\eta _3+d\eta _2`$ for some Poincaré-invariant local forms $`\eta _3`$ and $`\eta _2`$. The latter equation has to hold identically in all the fields, antifields and their derivatives. In particular, it must therefore be fulfilled when we set all fields, antifields and their derivatives equal to zero except for the undifferentiated ghosts. This yields an equation $`P(\theta (C))=sh(C)`$ since $`\eta _2`$ does not involve $`x^\mu `$ in case II (in contrast to case I). By the Lie algebra cohomology, $`P(\theta (C))=sh(C)`$ implies $`P(\theta (C))=0`$, see Section 8. Hence, in the space of Poincaré-invariant local forms, no nonvanishing volume-form $`d^3xP(\theta (C))`$ is trivial and the general solution of the consistency condition reads
$`\text{Case II:}H^{,0}(s|d,\mathrm{\Omega }):`$ $`\omega ^0[Q]^0`$
$`H^{,1}(s|d,\mathrm{\Omega }):`$ $`\omega ^1[Q]^1`$
$`H^{,2}(s|d,\mathrm{\Omega }):`$ $`\omega ^2[Q]^2`$
$`H^{,3}(s|d,\mathrm{\Omega }):`$ $`\omega ^3[Q]^3+d^3xP(\theta (C)).`$ (14.10)
##### Antifield dependence.
$`[Q]^2`$ and $`[Q]^3`$ contain antifields. This antifield dependence can actually be removed by the addition of trivial solutions, except when $`Q(\theta (C))`$ contains abelian ghosts. For instance, consider a $`\theta _r(C)=\frac{1}{3}e\mathrm{Tr}(C^3)`$ with $`m(r)=2`$. We know that this $`\theta _r(C)`$ can be completed to $`q_r(\stackrel{~}{C},F)=\mathrm{Tr}[\stackrel{~}{C}F\frac{1}{3}e\stackrel{~}{C}^3]`$, $`\stackrel{~}{C}=C+A`$, which satisfies $`(s+d)q_r(\stackrel{~}{C},F)=0`$ in 3 dimensions, see subsection 10.6. The solutions arising from $`q_r(\stackrel{~}{C},F)`$ do not involve antifields and are indeed equivalent to those obtained from $`\theta _r(𝒞)=\frac{1}{3}e\mathrm{Tr}(𝒞^3)`$. Namely one has
$$\mathrm{Tr}(\stackrel{~}{C}F\frac{1}{3}e\stackrel{~}{C}^3)=\frac{1}{3}e\mathrm{Tr}(𝒞^3)+(s+d)\mathrm{Tr}(\stackrel{~}{C}A^{}+\stackrel{~}{C}C^{})$$
where $`A^{}=A^IT_I`$, $`C^{}=C^IT_I`$. Analogous statements apply to all $`\theta _r(C)`$ with $`m(r)>1`$ and thus to all polynomials thereof. In particular, when the gauge group is semisimple, all $`[Q]^p`$ can be replaced by antifield independent representatives arising from polynomials $`Q(q(\stackrel{~}{C},F))`$. It is then obvious that (14.9) and (14.10) reproduce theorem 11.1 when the gauge group is semisimple: in the notation of theorem 11.1, one gets solutions $`B^1`$, $`B^2`$ and $`B^3`$ given by the 1-, 2- and 3-forms in $`Q(q(\stackrel{~}{C},F))`$, and solutions $`I^\alpha (x,dx)\mathrm{\Theta }_\alpha `$ given by $`P(\theta (C),x)`$, $`dx^\mu P_\mu (\theta (C),x)`$ and $`dx^\mu dx^\nu P_{\mu \nu }(\theta (C),x)`$ in case I, and by $`[Q]^0`$ and $`d^3xP(\theta (C))`$ in case II respectively. In particular there are no solutions $`V_{\mathrm{\Delta }\alpha }`$ when the gauge group is semisimple case because there are no nontrivial Noether currents at all in that case. The latter statement follows directly from the results, because $`H^{1,3}(s|d,\mathrm{\Omega })`$ is empty when the gauge group is semisimple.
In contrast, the antifield dependence cannot be completely removed when $`Q(\theta _r(C))`$ contains abelian ghosts (recall that the abelian ghosts coincide with those $`\theta _r(C)`$ which $`m(r)=1`$). For instance, the abelian $`𝒞^I`$ satisfy $`(s+d)𝒞^I=0`$ and provide thus solutions to the descent equations by themselves. The 3-form solution in an abelian $`𝒞^I`$ is $`d^3xC_I^{}`$. This is a nontrivial solution because it is also a nontrivial representative of $`H_2^3(\delta |d,\mathrm{\Omega })`$, see Section 6. Since it is a nontrivial solution with negative ghost number, it is impossible to make it antifield independent.
### 14.4 Examples
Let us finally spell out the results for $`H^{g,3}(s|d,\mathrm{\Omega })`$, $`g1`$, when the gauge group is either simple and compact, or purely abelian. The generalization to a general gauge group (product of abelian factors times a semi-simple group) is straightforward.
##### Simple compact gauge group.
In this case $`H^{g,3}(s|d,\mathrm{\Omega })`$ vanishes for $`g<0`$ and $`g=1`$, while $`H^{0,3}(s|d,\mathrm{\Omega })`$ is one-dimensional and represented by the 3-form $`\omega ^{0,3}`$ contained in $`\frac{1}{3}e\mathrm{Tr}(𝒞^3)`$ (both in case I and II),
$`H^{g,3}(s|d,\mathrm{\Omega })=0\text{for }g<0\text{ and }g=1,`$ (14.11)
$`\omega ^{0,3}=e\mathrm{Tr}[\frac{1}{3}A^3+(CA+AC)A^{}+C^2C^{}].`$ (14.12)
##### Purely abelian gauge group.
In this case we have $`Q(\theta (C))f_{I_1\mathrm{}I_k}C^{I_1}\mathrm{}C^{I_k}`$ where $`f_{I_1\mathrm{}I_k}`$ are arbitrary constant antisymmetric coefficients. In case I, this yields the following nontrivial representatives of $`H^{g,3}(s|d,\mathrm{\Omega })`$ for $`g=2,\mathrm{},1`$:
$`\omega ^{2,3}`$ $`=`$ $`f_IC^I`$
$`\omega ^{1,3}`$ $`=`$ $`2f_{IJ}(A^IA^J+C^IC^J)`$
$`\omega ^{0,3}`$ $`=`$ $`f_{IJK}(A^IA^JA^K+6C^IA^JA^K+3C^IC^JC^K)`$
$`\omega ^{1,3}`$ $`=`$ $`f_{IJKL}C^I(4A^JA^KA^L+12C^JA^KA^L+4C^JC^KC^L).`$ (14.13)
In case II one gets in addition representatives of $`H^{0,3}(s|d,\mathrm{\Omega })`$ and $`H^{1,3}(s|d,\mathrm{\Omega })`$ given by the volume element $`d^3x`$ and by $`d^3xa_IC^I`$ respectively (with $`a_I`$ arbitrary constant coefficients).
## 15 References for other gauge theories
In this section, we give references to works where the previous algebraic techniques have been used to find the general solution of the consistency condition with antifields included (and for the BRST differential associated with gauge symmetries) in other field theoretical contexts.
Some aspects of local BRST cohomology for the Stueckelberg model are investigated in . The general solution of the Wess-Zumino consistency condition for gauged non-linear $`\sigma `$-models is discussed in .
Cohomological techniques (Poincaré lemma, BRST cohomology) have been recently analyzed on the lattice in .
Algebraic aspects of gravitational anomalies are discussed in . The general solution of the consistency condition without antifields is given in ; this work is extended to include the antifields in spacetime dimensions strictly greater than two in , where again, the cohomology of the Koszul-Tate differential is found to play a crucial rôle.
In $`2`$ spacetime dimensions, these groups have been analyzed in the context of the (bosonic) string world sheet action coupled to backgrounds in . Complete results, with the antifields included, are derived in .
Algebraic results on the Weyl anomaly may be found in .
Algebraic aspects of the consistency condition for $`N=1`$ supergravity in $`4`$ dimensions have been discussed in . The complete treatment, with antifields included, is given in .
For $`p`$-form gauge theories, the local BRST cohomology groups without antifields have been investigated in and more recently, with antifields, in .
## Acknowledgements
We acknowledge discussions with many colleagues in various stages of development of this work, especially José de Azcárraga, Norbert Dragon, Michel Dubois-Violette, Jean Fisch, Maximilian Kreuzer, Sergei Kuzenko, Christiane Schomblond, Jim Stasheff, Raymond Stora, Michel Talon, Claudio Teitelboim, Jan-Willem van Holten, Claude Viallet and Steven Weinberg. G.B. is Scientific Research Worker of the “Fonds National Belge de la Recherche Scientific”. He also acknowledges the hospitality of the Department of Theoretical Physics of the University of Valencia. The work of G.B. and M.H. is supported in part by the “Actions de Recherche Concertées” of the “Direction de la Recherche Scientifique - Communauté Française de Belgique”, by IISN - Belgium (convention 4.4505.86) and by Proyectos FONDECYT 1970151 and 7960001 (Chile). The work of F.B. was supported by the Deutsche Forschungsgemeinschaft through a Habilitation grant.
|
warning/0002/cond-mat0002068.html
|
ar5iv
|
text
|
# Untitled Document
Centre de Recherches sur les Très Basses Températures
Grenoble
Pierre MONCEAU
| |
| --- |
| Editors of Physical Review B |
Grenoble, January 12, 2000
Dear Editors,
We would like to submit our manuscript entitled “Dielectric response of charge induced correlated state in quasi-one-dimensional conductor (TMTTF)<sub>2</sub>PF<sub>6</sub> for publication in Phys. Rev. We previously submitted it to Phys. Rev. Lett. (code number LH7335).
The delay for resubmission has been caused by difficulties in communication between co-authors, one at Moscow, another visiting Japan during several weeks. We would like, however, the original received date to be retained. The manuscript has been amended according to the referee comments as described below :
Referee A
1) We modified our description of the extended Mott-Hubbard model for half-filled bands (bottom of page 2) taking into account both the on-site interaction (U) and the near-neighbor interaction (V).
2) The $`2k_F`$ lattice fluctuations coupled to the lattice are observed at much lower temperature (below 60 K) than the large increase of the dielectric permittivity. We have rewritten the beginning of the discussion part, page 6, to make this point more clear.
Referee B
1) These organic conductors are very fragile and cracks may reflect significant strain when the cooling rate is too fast and/or the sample holding too tight. We used all the care possible for avoiding cracks in our crystals and thus measuring bulk properties.
2) There is no evidence in (TMTTF)<sub>2</sub>PF<sub>6</sub> of a structural phase transition affecting the main Bragg reflections. The tentative explanation of our data is the occurrence of a superstructure ($`4k_F`$) transition. Such a transition could be very likely detected from NMR experiments.
Yours sincerely,
P. MONCEAU
|
warning/0002/hep-th0002084.html
|
ar5iv
|
text
|
# 1 Introduction.
## 1 Introduction.
Deformations of mathematical structures have been used at different moments in physics. When Galilean transformations between inertial systems were seen not to describe adequately the physical world, a deformation of the group law arose as the solution to this paradox. The Lorenz group is a deformation of the Galilei group in terms of the parameter $`\frac{1}{c}`$. From the mathematical point of view it is not difficult to imagine the deformation of a group inside the category of groups, but form the physical point of view it has enormous consequences. In this deformation scheme, the old structure is seen as a limit or contraction when the parameter takes a preferred value.
The mathematical structure of quantum mechanics has also an ingredient of deformation with respect to classical mechanics. The first star product or formal deformation of the commutative algebra of classical observables was written in Ref.. The star product is a product in the space of formal series in $`\mathrm{}`$ whose coefficients are functions on the phase space. It is homomorphic to the product of operators in quantum mechanics. In this case the deformation occurs inside the category of algebras, although giving up the commutativity. More complicated features arise in the mathematical framework of quantum mechanics, and the first thing one can realize is that this deformation in the parameter $`\mathrm{}`$ is a formal deformation (that is, the series in $`\mathrm{}`$ of a product of two functions is not convergent), unless strong restrictions are made on the functions. Nevertheless, Bayen, Flato, Fronsdal, Lichnerowicz and Sternheimer realized in their seminal papers that a first approach to a quantum system could be studying the formal deformations of the classical one, leaving aside problems of convergence and of construction of the Hilbert space. The existence and uniqueness of this deformation (up to gauge transformations), finally showed in for any Poisson manifold, supports the belief that the formal deformation encloses the essential information of the quantum system.
Much more recently, non commutative geometry has entered in physics in different contexts. One context is string theory and M-theory. In their pioneering paper, Connes, Douglas and Schwarz introduced non commutative spaces (tori) as possible compactification manifolds of space-time. Non commutative geometry arises as a possible scenario for short-distance behaviour of physical theories. Since non commutative geometry generalises standard geometry in using a non commutative algebra of “functions”, it is naturally related to the simpler context of deformation theory.
Matrix theory on non commutative tori is related to 11 dimensional supergravity toroidal compactifications with 3-form backgrounds. In this framework, T-duality arises as Morita equivalence in non commutative geometry . This lead to subsequent developments of Yang-Mills theories on non commutative tori , as the study of their BPS states and a reformulation of T and U dualities of Born-Infeld actions on non commutative tori . Non commutative geometry also appeared in the framework of open string theory .
More recently, Seiberg and Witten identified limits in which the entire string dynamics, in presence of a $`B`$-field, is described by a deformed gauge theory in terms of a Weyl-Moyal star product on space-time. In particular, they showed that the pure quadratic gauge theory with deformed abelian gauge symmetry is related through a change of variables to a non polynomial gauge theory with undeformed gauge group. This brings a connection between the Born-Infeld action and gauge theories in non commutative spaces. In view of the fact that the supersymmetric Born-Infeld action naturally arises as the Goldstone action of $`N=2`$ supersymmetry, partially broken to $`N=1`$ , it must be the case that this interpretation should have its counterpart in the framework of deformed gauge theory. This connection will be clarified in this paper. Subsequently, this deformation of space-time was used for ordinary four dimensional field theories with a $`B`$-field of maximal rank in $`𝐑^4`$ space-time. It was shown that the deformed theories enjoy unsuspected renormalization properties as well as UV/IR connection reminiscent of string theory .
Other approaches connecting deformation theory to theories of gravity have also appeared in the literature. Among others, we can mention the deformation quantization of M-theory , quantum anti de Sitter space-time , q-gravity and gauge theories of quantum groups .
In this paper some issues related to theories formulated in deformations of superspace are investigated. Aspects of non commutative supergeometry and noncommutative supersymmetric field theories were recently considered in the literature. Section 2 is a self explanatory account of the Moyal-Weyl deformation generalised to superspace. In particular we show that the deformation of a Grassmann algebra obtained with a Weyl ordering rule is a Clifford algebra once we specify a value for the parameter (non formal deformation). In section 3 we consider the compatibility of this deformation with the supertranslation group. In section 4 we present the simplest example of a deformed supersymmetric field theory, the Wess-Zumino model and we give its explicit expression in terms of field components. In section 5 we consider deformed gauge groups in superspace and derive a non commutative version of rank 1 Yang-Mills theory which is then coupled to chiral superfields. We show in particular, to first order in the deformation parameter, that the change of variables of Seiberg and Witten to convert the rank 1 non commutative quadratic gauge theory to a commutative higher derivative theory is consistent with supertranslations. Rank 1 $`N`$-extended super Yang-Mills theories are invariant under shifts of the $`N`$ gauginos by $`N`$ constant, anticommuting parameters, so they can be regarded as Goldstone actions of $`2N`$ supersymmetries spontaneously broken to $`N`$ supersymmetries. Finally in section 6 we derive the $`\alpha ^{}0`$ limit of supersymmetric Born-Infeld actions, which are the starting point for comparisons with super Yang-Mills theories in non commutative superspaces.
## 2 Star product in superspace.
### 2.1 Super Poisson bracket.
A super vector space over $`𝐑`$ or $`𝐂`$ is a $`𝐙_2`$-graded vector space $`V=V_0V_1`$ with grading $`p=0,1`$. On $`V`$ we can give a associative operation $`:VVV`$ respecting the grading, that is
$$p(ab)=p(a)+p(b).$$
Then $`V`$ is a super algebra. A super Lie algebra is a super vector space $`V`$ with a bracket $`[,]:VVV`$ satisfying
$$[X,Y]=(1)^{p_Xp_Y}[Y,X],$$
(1)
$$[X,[Y,Z]]+(1)^{p_Z(p_X+p_Y)}[Z,[X,Y]]+(1)^{p_X(p_Y+p_Z)}[Y,[Z,X]].$$
(2)
The super space of dimension (p,q) is the affine space $`𝐑^p`$ together with a super algebra $`𝒮^{p,q}=C^{\mathrm{}}(𝐑^p)\mathrm{\Lambda }(𝐑^q)`$ (the algebra of “functions” on superspace), where $`\mathrm{\Lambda }(𝐑^q)=_{i=0}^q\mathrm{\Lambda }^i(𝐑^q)`$ is the exterior algebra of $`q`$ symbols $`\theta ^1,\theta ^2,\mathrm{}\theta ^q`$. We assign grade one to the symbols $`\theta ^i`$. It is clear what are the even and odd subspaces.
An element of this superalgebra can be written as
$$a(x,\theta )=a_0(x)+a_i(x)\theta ^i+a_{i_1i_2}\theta ^{i_1}\theta ^{i_2}+\mathrm{}+a_{i_1i_2\mathrm{}i_q}\theta ^{i_1}\theta ^{i_2}\mathrm{}\theta ^{i_q},$$
where $`a_{i_1i_2\mathrm{}i_j}`$ is antisymmetric in all its indices. It is a commutative superalgebra, that is,
$$ab=(1)^{p_ap_b}ba$$
for homogeneous elements of degrees $`p_a`$ and $`p_b`$.
A left derivation of degree $`m=0,1`$ of a super algebra is a linear map $`^L:VV`$ such that
$$^L(ab)=^L(a)b+(1)^{mp_a}a^L(b).$$
Graded left derivations form a $`𝐙_2`$ graded vector space. Any linear map $`L`$ can be decomposed as the sum $`L=L_0+L_1`$, where $`L_0`$ and $`L_1`$ are maps of degree 0 (they preserve the degree) and 1 (they change the degree) respectively. If the superalgebra is commutative, an even derivation has degree 0 as a linear map and an odd derivation has degree 1 as a linear map.
In the same way right derivations are defined,
$$^R(ab)=(1)^{mp_b}^R(a)b+a^R(b).$$
Notice that derivations of degree zero are both, right and left.
A super Poisson structure on a commutative (this condition could be relaxed, in particular, to introduce matrix valued superfields) super algebra is a super Lie algebra structure $`\{,\}`$ on it which is also a bi-derivation with respect to the commutative super algebra product. More specifically, it satisfies the following derivation property on homogeneous elements,
$$\{a,bc\}=\{a,b\}c+(1)^{p_ap_b}a\{b,c\},$$
which together with the antisymmetry property (1) implies
$$\{bc,a\}=b\{c,a\}+(1)^{p_ap_c}\{b,a\}c.$$
So, for example, if $`a`$ is even, $`\{a,\}`$ is a derivation of degree zero, and if it is odd it is a left derivation of degree 1.
#### Example.
Consider the superalgebra $`𝒮^{p,2}`$, with elements
$$\mathrm{\Phi }(x,\theta )=\mathrm{\Phi }_0(x)+\mathrm{\Phi }_\alpha (x)\theta _\alpha +\mathrm{\Phi }_{\alpha \beta }(x)\theta _\alpha \theta _\beta .$$
The derivations $`_i`$ defined as
$$_i\mathrm{\Phi }(x,\theta )=_i\mathrm{\Phi }_0(x)+_i\mathrm{\Phi }_\alpha (x)\theta _\alpha +_i\mathrm{\Phi }_{\alpha \beta }(x)\theta _\alpha \theta _\beta $$
are even derivations. The left derivations $`_\alpha ^L`$ defined as
$$_\alpha ^L\mathrm{\Phi }(x,\theta )=\mathrm{\Phi }_\alpha (x)+2\mathrm{\Phi }_{\alpha \beta }(x)\theta _\beta $$
are odd. We can also define right derivations,
$$_\alpha ^R\mathrm{\Phi }(x,\theta )=\mathrm{\Phi }_\alpha (x)+2\mathrm{\Phi }_{\beta \alpha }(x)\theta _\beta $$
Notice that $`_\alpha ^R=_\alpha ^L`$ on odd elements and $`_\alpha ^R=_\alpha ^L`$ on even elements. This implies that $`[_\alpha ^R,_\beta ^L]_{}=0`$
These definitions can easily be extended to algebras with bigger odd dimension in an obvious manner.
As an example, consider the following super Poisson bracket
$$\{\mathrm{\Phi },\mathrm{\Psi }\}=P^{ab}_a\mathrm{\Phi }_b\mathrm{\Psi }+P^{\alpha \beta }_\alpha ^R\mathrm{\Phi }_\beta ^L\mathrm{\Psi }=P^{AB}_A^R\mathrm{\Phi }_B^L\mathrm{\Psi }.$$
(3)
where $`P`$ is a constant matrix and satisfies the symmetry properties
$$P^{ab}=P^{ba},P^{\alpha \beta }=P^{\beta \alpha }.$$
It is easy to see that it satisfies the Jacobi identity (2).
### 2.2 Super star product.
A generalisation of the Moyal-Weyl deformation of $`C^{\mathrm{}}(𝐑^n)`$ to $`𝒮^{p,q}`$ exists. This algebra structure corresponds to the quantization of systems with both, bosonic and fermionic degrees of freedom, and it was studied by Berezin as early as in . There, the quantization was studied in terms of products of Weyl symbols of operators, very much in the same spirit than . In a language closer to ours, it appeared in and in .
We remind that a deformation of the commutative algebra $`C^{\mathrm{}}(𝐑^p)`$ is an associative product on the space of formal series on a parameter $`h`$ with coefficients in $`C^{\mathrm{}}(𝐑^n)`$, that is $`C^{\mathrm{}}(𝐑^n)[[h]]=𝐑[[h]]C^{\mathrm{}}(𝐑^n)`$. The term of first order in $`h`$, antisymmetrized, is always a Poisson bracket.
We denote by $`P(fg)=\{f,g\}`$ a super Poisson bracket like (3), in a space of arbitrary odd dimension. A deformation of the commutative superalgebra $`𝒮^{p,q}`$ is then given by
$`:𝒮^{p,q}[[h]]𝒮^{p,q}[[h]]`$ $``$ $`𝒮^{p,q}[[h]]`$
$`fg`$ $``$ $`e^{hP}(fg)`$ (4)
where we have denoted
$$e^{hP}=\underset{n=0}{\overset{\mathrm{}}{}}\frac{h^n}{n!}P^n$$
with
$$P^n(fg)=P^{A_1B_1}P^{A_2B_2}\mathrm{}P^{A_nB_n}(_{A_1}^R_{A_2}^R\mathrm{}_{A_n}^R)f(_{B_1}^L_{B_2}^L\mathrm{}_{B_n}^Lg).$$
(We remind here that $`P^{AB}`$ is a constant matrix). For $`q=0`$ this is the standard Moyal-Weyl deformation. Notice that the first order term is exactly the super Poisson bracket. The proof of the associativity of this product is parallel to the one developed in for the bosonic case.
### 2.3 Non formal deformation.
We consider the associative algebra over $`𝐑[[h]]`$, $`𝒜^{p,q}`$, generated by the symbols $`X^1,\mathrm{},X^p`$, $`\mathrm{\Theta }^1,\mathrm{}\mathrm{\Theta }^q`$ and the relations given by the super Poisson bracket (3).
$`[X^a,X^b]_{}=hP^{ab},`$ (5)
$`[\mathrm{\Theta }^\alpha ,\mathrm{\Theta }^\beta ]_+=hP^{\alpha \beta }.`$ (6)
where $`h`$ is a formal parameter. Since $`X`$’s and $`\mathrm{\Theta }`$’s commute, it is clear that $`𝒜^{p,q}U_h^p\mathrm{\Lambda }_h^q`$, where $`U_h^p`$ is the associative algebra over $`𝐑[[h]]`$ generated by the symbols $`X`$’s and relations (5) and $`\mathrm{\Lambda }_h^q`$ is the associative algebra over $`𝐑[[h]]`$ generated by the symbols $`\mathrm{\Theta }`$’s and relations (6). $`𝒜^{p,q}`$ is isomorphic to $`(\text{Pol}(𝐑^p)\mathrm{\Lambda }(𝐑^q)[[h]],)`$, (polynomials are closed under the $``$ operation). To prove this, we take a basis in $`\text{Pol}(𝐑^p)[[h]]`$,
$$x^{i_1}x^{i_2}\mathrm{}x^{i_n},i_1i_2\mathrm{}i_n.$$
(7)
There is a $`𝐑[[h]]`$-module isomorphism $`\text{Sym}:\text{Pol}(𝐑^p)[[h]]U_h^p`$ mapping the elements of the basis (7) in the following way
$`\text{Sym}(x^{i_1}x^{i_2}\mathrm{}x^{i_n})={\displaystyle \frac{1}{n}}{\displaystyle \underset{\sigma S_n}{}}X^{\sigma (i_1)}X^{\sigma (i_2)}\mathrm{}X^{\sigma (i_n)}=`$
$`\text{exp}(X^i_i)(x^{i_1}x^{i_2}\mathrm{}x^{i_n})|_{x^{i_k}=0},`$
which is the usual Weyl or symmetric ordering. The product in $`\text{Pol}(𝐑^p)[[h]]`$ defined by
$$\text{Sym}^1(\text{Sym}(f)\text{Sym}(g))$$
(8)
is equal to $``$ in (4) restricted to polynomials. The proof of this fact is given in (where indeed, the argument is extended to $`C^{\mathrm{}}`$ functions).
Consider the basis in $`\mathrm{\Lambda }(𝐑^q)[[h]]`$
$$\theta _{i_1}\theta _{i_2}\mathrm{}\theta _{i_n},i_1i_2\mathrm{}i_n.$$
(9)
(dim($`\mathrm{\Lambda }(𝐑^q)`$)=$`2^q`$). We define the isomorphism $`\text{Sym}:\mathrm{\Lambda }(𝐑^q)[[h]]\mathrm{\Lambda }_h^q`$ as
$$\text{Sym}(\theta _{i_1}\theta _{i_2}\mathrm{}\theta _{i_n})=\mathrm{\Theta }_{i_1}\mathrm{\Theta }_{i_2}\mathrm{}\mathrm{\Theta }_{i_n}$$
for the elements of the basis (9). This is the equivalent of the Weyl ordering for odd generators. One can see directly by inspection that the product defined on $`\mathrm{\Lambda }(𝐑^q)[[h]]`$ by this isomorphism is the same than $``$ in (4) restricted to the exterior algebra. So we can conclude that the algebra generated by $`X,\mathrm{\Theta }`$ and relations (5) and (6) is isomorphic to the $``$-product algebra. Given any $`𝐑[[h]]`$-module isomorphism among $`\text{Pol}(𝐑^p)[[h]]`$ and $`U_h^p`$, one can construct a star product as in (8). The resulting (isomorphic) star products are called equivalent.
For polynomials, the formal parameter $`h`$ can be specialized to any real value and one obtains a convergent star product. We want to look closer to this algebra over $`𝐑`$.
By a linear change of coordinates $`PA^TPA`$, we can always bring the matrices $`P^{ab}`$ and $`P^{\alpha \beta }`$ into a canonical form , that is
$$P^{ab}=\left(\begin{array}{ccc}0& I& 0\\ I& 0& 0\\ 0& 0& 0\end{array}\right),P^{\alpha \beta }=\left(\begin{array}{cccc}\eta _1& 0& \mathrm{}& 0\\ 0& \eta _2& \mathrm{}& 0\\ 4\\ 0& 0& \mathrm{}& \eta _q\end{array}\right)$$
(10)
where $`\eta _\alpha =\pm 1`$ for $`\alpha =1,\mathrm{}q^{}`$ and $`\eta _\alpha =0`$ for $`\alpha =q^{}+1,\mathrm{}q`$.
Denote $`q^{}=m+n`$, where $`\eta _\alpha =1`$ for $`\alpha =1,\mathrm{}m`$ and $`\eta _\alpha =+1`$ for $`\alpha =m+1,\mathrm{}m+n`$. It is obvious that $`\mathrm{\Lambda }_h^q`$, evaluated for a real value of $`h`$ is isomorphic to the Clifford algebra $`𝒞(m,n)`$ tensor product with the exterior algebra on the remaining $`qq^{}`$ generators, which doesn’t get deformed. (The isomorphism is given by $`\gamma _\alpha =\sqrt{2h}\mathrm{\Theta }_\alpha `$). This relation with Clifford algebras was noticed in .
If $`m=n`$, we can make again a linear change of variables that brings $`P^{\alpha \beta }`$ to the form
$$P^{\alpha \beta }=\left(\begin{array}{cccccc}0& 1& 0& 0& \mathrm{}& 0\\ 1& 0& 0& 0& \mathrm{}& 0\\ 0& 0& 0& 1& \mathrm{}& 0\\ 0& 0& 1& 0& \mathrm{}& 0\\ 6\\ 0& 0& 0& 0& \mathrm{}& 0\end{array}\right).$$
We have then $`n`$ pairs of canonically conjugate fermionic variables. The Poisson bracket for fermionic variables was first written in .
## 3 Formal deformations of rigid supersymmetry.
We are interested in describing physical theories defined on a deformation of superspace. Superfields are used as basic objects of such theories. Mathematically, they are a generalisation of the superalgebra $`𝒮^{p,q}`$. Consider the trivial bundle over $`𝐑^p`$ with fibre the Grassmann or exterior algebra $`\mathrm{\Lambda }(ϵ_1,\mathrm{},ϵ_n)`$. Consider the algebra of sections on that bundle, $`\mathrm{\Gamma }^q(𝐑^p)`$ and the tensor product $`\mathrm{\Phi }_n^{p,q}=\mathrm{\Gamma }^n(𝐑^p)\mathrm{\Lambda }^n`$.
$`\mathrm{\Phi }_n^{p,q}`$ is a commutative superalgebra with the product defined as usual
$$(a\mathrm{\Psi }_1)(b\mathrm{\Psi }_2)=(1)^{p_1p_b}ab\mathrm{\Psi }_1\mathrm{\Psi }_2a,b\mathrm{\Gamma }^n(𝐑^p),f,g\mathrm{\Lambda }^n,$$
and the left and right $`\mathrm{\Gamma }^n(𝐑^p)`$-module structures are given by
$$b(a\mathrm{\Psi })=(ba\mathrm{\Psi }),(a\mathrm{\Psi })b=(1)^{p_bp_\mathrm{\Psi }}(ab\mathrm{\Psi }).$$
The rank ($`n`$) of the trivial bundle can be chosen $`n=\text{dim}\mathrm{\Lambda }^q=2^q`$. Then scalar superfields are an even subalgebra of $`\mathrm{\Phi }_n^{p,q}`$, generated by elements of the form.
$$\mathrm{\Phi }(x,\theta )=\mathrm{\Phi }_0(x)+\theta _i\mathrm{\Phi }_i(x)+\theta _j\theta _j\mathrm{\Phi }_{ij}+\mathrm{}$$
where $`\mathrm{\Phi }_{i_1i_2\mathrm{}i_k}`$ are independent global sections in $`\mathrm{\Gamma }^q(𝐑^p)`$, antisymmetric in the indices $`i_1i_2\mathrm{}i_k`$.
One can extend (3) and (4) to $`\mathrm{\Phi }_n^{p,q}𝐑[[h]]`$ by linearity. It follows that $`(\mathrm{\Phi }_n^{p,q}[[h]],)`$ is a non commutative superalgebra.
In what follows, we will restrict ourselves to four dimensional space-time, although all considerations could be easily extended to other dimensions. We consider the superspace associated with the four dimensional N-extended Poincaré supersymmetry, with coordinates (or generators) $`\{x^\mu ,\theta ^{\alpha i},\overline{\theta }_i^{\dot{\alpha }}\}`$, where $`\{x^\mu \}`$ are the coordinates of ordinary four dimensional Minkowski space $``$, and $`\{\theta ^{\alpha i},\overline{\theta }_i^{\dot{\alpha }}\}`$ are Weyl spinors under the Lorentz group. We are interested in deformations of this superspace such that they have an action of the supertranslation group. The odd supertranslations with parameters $`ϵ^{\alpha i},\overline{ϵ}_i^{\dot{\alpha }}`$, act on the generators of superspace as
$`x^\mu `$ $``$ $`x_{}^{}{}_{}{}^{\mu }=x^\mu +i(\theta ^{\alpha i}(\sigma ^\mu )_{\alpha \dot{\alpha }}\overline{ϵ}_i^{\dot{\alpha }}ϵ^{\alpha i}(\sigma ^\mu )_{\alpha \dot{\alpha }}\overline{\theta }_i^{\dot{\alpha }})`$
$`\theta ^{\alpha i}`$ $``$ $`\theta _{}^{}{}_{}{}^{\alpha i}=\theta ^{\alpha i}+ϵ^{\alpha i}`$
$`\overline{\theta }_i^{\dot{\alpha }}`$ $``$ $`\overline{\theta ^{}}_i^{\dot{\alpha }}=\overline{\theta }_i^{\dot{\alpha }}+\overline{ϵ}_i^{\dot{\alpha }}.`$ (11)
By convention, we write a scalar superfield as
$$\mathrm{\Phi }(x^\mu ,\theta ^{\alpha i},\overline{\theta }_i^{\dot{\alpha }})=\mathrm{\Phi }_0(x)+\theta ^{\alpha i}\mathrm{\Psi }_\alpha ^i+\overline{\theta }_i^{\dot{\alpha }}\overline{\mathrm{\Sigma }}_{\dot{\alpha }i}+\theta ^{\alpha i}\theta ^{\beta j}\mathrm{\Psi }_{\alpha \beta }^{ij}+\mathrm{},$$
where we have dropped the symbols “$``$” and “$``$”. Let $`g(ϵ)`$ be a super translation as in (11). The action of $`g`$ on superfields is given by $`(g^1\mathrm{\Phi })(x,\theta ,\overline{\theta })=\mathrm{\Phi }(x^{},\theta ^{},\overline{\theta }^{})`$. We require that the super translation group acts as a group of automorphisms on the deformed algebra, that is
$$g(\mathrm{\Phi }_1\mathrm{\Phi }_2)=(g\mathrm{\Phi }_1)(g\mathrm{\Phi }_2).$$
It is convenient to introduce the right and left odd derivations called super covariant derivatives,
$`D_{}^{R,L}{}_{\alpha i}{}^{}`$ $`=`$ $`_{}^{R,L}{}_{\alpha i}{}^{}+(i\sigma _{\alpha \dot{\alpha }}^\mu \overline{\theta }_i^{\dot{\alpha }})_\mu )^{R,L},`$
$`\overline{D}_{}^{R,Li}{}_{\dot{\alpha }}{}^{}`$ $`=`$ $`\overline{}_{}^{R,Li}{}_{\dot{\alpha }}{}^{}(i\theta ^{\alpha i}\sigma _{\alpha \dot{\alpha }}^\mu _\mu )^{R,L}.`$
They have the property that
$$D_{}^{R,L}{}_{\alpha i}{}^{}(g\mathrm{\Phi })=g(D_{}^{R,L}{}_{\alpha i}{}^{}\mathrm{\Phi }).$$
and the same for $`\overline{D}_{}^{R,Li}{}_{\dot{\alpha }}{}^{}`$. We define a Poisson bracket
$$\{\mathrm{\Phi },\mathrm{\Psi }\}=P^{\mu \nu }_\mu \mathrm{\Phi }_\nu \mathrm{\Psi }+P^{\alpha i\beta j}D_{}^{R}{}_{\alpha i}{}^{}\mathrm{\Phi }D_{}^{L}{}_{\beta j}{}^{}\mathrm{\Psi }.$$
(12)
The crucial properties are that $`[D_{}^{R}{}_{\alpha i}{}^{},D_{}^{L}{}_{\beta j}{}^{}]_{}=0`$ and that $`D_{\alpha i}^{R,L}P^{AB}=0`$ ($`A=\mu ,\{\alpha i\}`$), so one can again construct a Weyl-Moyal star product as in (4), which will also be covariant under the supertranslation group. One could also extend (12) by using $`Nk`$ $`D`$’s and $`k`$ $`\overline{D}`$’s $`k=1,\mathrm{}N`$, which can be taken to anticommute. Notice that this Poisson structure is degenerate in the space of odd variables.
A chiral field satisfies the constraint $`\overline{D}_{}^{R,Li}{}_{\dot{\alpha }}{}^{}\mathrm{\Phi }=0`$. The solution of this equation can be written (after a change of variables) as $`\mathrm{\Phi }(x,\theta )`$. Chiral superfields are a subalgebra under ordinary multiplication, but they are not closed under the star product defined with (12) unless $`P^{\alpha i\beta j}=0`$.
When considering extended supersymmetry this notion of chirality can be generalised. The R-symmetry group U($`N`$) acts by automorphisms on the super Poincaré algebra, leaving invariant the even generators. For $`N>1`$, one can take the direct product of the Minkowski space with the flag manifold SU($`N`$)/U(1)<sup>N-1</sup>, and consider a supermanifold structure of odd dimension $`4N`$ on it. This is constructed by taking the quotient $`(\text{SU}(N))_s𝒮𝒯/\text{U(1)}^{N1}`$, where $``$ is the Lorentz group and $`𝒮𝒯`$ is the supertranslation group. It it is called harmonic superspace . The algebra of global sections (or functions) on the resulting supermanifold is isomorphic to
$$C^{\mathrm{}}(\times \text{SU}(N)/U(1)^{N1})\mathrm{\Lambda }^{4N}.$$
(13)
This isomorphism is not canonical, since it is not preserved by supersymmetry transformations, but it is preserved by the action of the R-symmetry group.
Let us denote the coordinates in an open set as $`\{x^\mu ,u,\theta ^{\alpha i},\overline{\theta }_i^{\dot{\alpha }}\}`$ where $`u`$ is a unitary matrix or coset representative of SU($`N`$)/U(1)<sup>N-1</sup>. With the coset representatives one can define rotated covariant derivatives
$$𝒟_{\alpha I}=u_I^iD_{\alpha i},\overline{𝒟}_{\dot{\alpha }}^I=u_I^i\overline{D}_{\dot{\alpha }}^i.$$
The advantage of such formulation is that the notion of chiral superfield can be generalised by imposing the following R-symmetry covariant constraints on the superfields $`\mathrm{\Phi }(x,u,\theta ,\overline{\theta })`$,
$$𝒟_{\alpha 1}\mathrm{\Phi }=\mathrm{}=𝒟_{\alpha k}\mathrm{\Phi }=0=\overline{𝒟}_{\dot{\alpha }}^{k+1}\mathrm{\Phi }=\mathrm{}=\overline{𝒟}_\alpha ^N\mathrm{\Phi }.$$
(no superscript will mean that we are considering a left derivative). The solution of these constraints can be expressed as superfields that do not depend on $`k`$ $`\theta `$’s and $`Nk`$ $`\overline{\theta }`$’s ($`k=0,N`$ being the usual chiral and antichiral superfields). These fields have been called “Grassmann analytic” in the literature and, as chiral superfields, they form a subalgebra.
One can consider deformations of this supermanifold for a given super Poisson structure. In particular one can consider a deformation affecting only the first factor in (13). Any deformation of this form will have the supersymmetry algebra as an algebra of derivations. As the simplest case, let us take a non trivial Poisson bracket only in the directions of $``$,
$$iP=iP^{\mu \nu }\frac{}{x^\mu }\frac{}{x^\nu }$$
that is, the Poisson bracket of two (complex) superfields is
$$\{\mathrm{\Phi }_1,\mathrm{\Phi }_2\}=iP^{\mu \nu }\frac{\mathrm{\Phi }_1}{x^\mu }\frac{\mathrm{\Phi }_2}{x^\nu }$$
where $`P^{\mu \nu }`$ is an arbitrary constant antisymmetric matrix and $`\mathrm{\Phi }_i(x,u,\theta )`$ arbitrary superfields. Then, the Weyl-Moyal star product on the algebra of superfields is given by
$$\mathrm{\Phi }_1\mathrm{\Phi }_2=\mathrm{exp}(iP)(\mathrm{\Phi }_1\mathrm{\Phi }_2).$$
(14)
It is clear from this expression that Grassmann analytic superfields are closed under the star product (14), as chiral superfields are.
## 4 Non commutative Wess-Zumino model.
The simplest example of an N=1 supersymmetric field theory is the Wess-Zumino model, whose action is
$$d^4xd^\theta d^2\overline{\theta }\mathrm{\Phi }\overline{\mathrm{\Phi }}+d^4x(d^2\theta (\frac{m}{2}\mathrm{\Phi }^2+\frac{g}{3}\mathrm{\Phi }^3)+\text{c. c.}),$$
where he chiral superfield $`\mathrm{\Phi }`$ has the expansion
$$\mathrm{\Phi }=A(y)+\sqrt{2}\theta \psi (y)+\theta \theta F(y),$$
where $`y=x+i\theta \sigma \overline{\theta }`$. A formal deformation of this action can be written using the star product defined above (14),
$$d^4xd^2\theta d^2\overline{\theta }\mathrm{\Phi }\overline{\mathrm{\Phi }}+d^4x(d^2\theta (\frac{m}{2}\mathrm{\Phi }^2+\frac{g}{3}\mathrm{\Phi }^3)+\text{c. c.}).$$
(15)
where $`\mathrm{\Phi }^n=\mathrm{\Phi }\mathrm{\Phi }\mathrm{}(n)\mathrm{}\mathrm{\Phi }`$. This model was also considered in . This Lagrangian is unique (for every star product) as a consequence of the fact that
$$d^4xAB=d^4xAB=d^4xBA$$
(16)
and
$$d^4xA_1\mathrm{}A_n=d^4xA_{\sigma (1)}\mathrm{}A_{\sigma (n)}$$
(17)
where $`\sigma `$ is a cyclic permutation of $`(1,\mathrm{},n)`$. Notice also that the above action is real, since
$$\overline{AB}=\overline{B}\overline{A}.$$
As a consequence of (16) and (17), the auxiliary field $`F`$ satisfies pure algebraic equations
$$F=m\overline{A}g\overline{A}\overline{A}$$
so the component form of the Lagrangian is
$`i_\mu \overline{\psi }\overline{\sigma }^\mu \psi +\overline{A}_\mu ^\mu A{\displaystyle \frac{1}{2}}m(\psi \psi +\overline{\psi }\overline{\psi })m^2\overline{A}Ag(A(\psi \psi )+`$
$`\overline{A}(\overline{\psi }\overline{\psi }))mg(A(\overline{A}\overline{A})+\overline{A}(AA))g^2(AA)(\overline{A}\overline{A}).`$ (18)
The non deformed Wess-Zumino model is a renormalizable field theory which only requires a (logarithmically divergent) wave function renormalization . This is due to supersymmetric non renormalization theorems of chiral terms .
The deformed Wess-Zumino model is the supersymmetric extension of the $`\varphi ^4`$ theory considered in where the model was proven to be “renormalizable” in some extended sense. Consequently, we expect that its supersymmetric extension is also “renormalizable”. Moreover, since the interactions are purely chiral, no quadratic divergences appear and then the UV/IR connection induced by extra poles in the propagator does not appear in this model . Additional aspects of the UV/IR connection in non commutative supersymmetric models are discussed in .
The non deformed Lagrangian has a quartic interaction that is invariant under a local U(1) symmetry. This invariance is inherited from the superconformal symmetry present in the model when $`m=0`$. The deformed Lagrangian only preserves the global U(1) invariance. It is interesting to observe that there is another possible quartic term
$$(\overline{A}A)^2,$$
(19)
which is invariant under a non commutative local U(1)-symmetry . Supersymmetry picks the first choice without any contradiction because the R-symmetry is not deformed.
Incidentally, it was also shown that the pure bosonic theory of a complex scalar field $`A`$ with the quartic interaction as given in Lagrangian (18) is not renormalizable unlike the theory with the quadratic invariant (19). This is not a contradiction because in the Wess-Zumino model the additional interactions due to supersymmetry are responsible for the cancellation of dangerous divergences (in particular quadratic divergences).
As a side remark, we note that while the interaction $`\overline{A}\overline{A}AA`$ is typical of an $`F`$-term, the other possibility $`\overline{A}A\overline{A}A`$ is typical of a $`D`$-term, so we expect the latter to occur in the deformed version of supersymmetric Q.E.D. Even more, both quartic terms occur and in fact are related one to another when $`N`$-extended supersymmetry is present. This will be the case in the deformed version of $`N=2,4`$ super Yang-Mills theories, which in addition require a deformation of the gauge symmetry.
## 5 Non commutative rank 1 gauge theory in superspace.
In this section we introduce a deformation of an abelian gauge theory in superspace . The gauge group is a group of formal series in a parameter with coefficients which are chiral superfields ($`U`$, with $`\overline{D}_{\dot{\alpha }}U=0`$). The multiplication law is given by the star product in (14)
$$U_1U_2=U_3,$$
which preserves chirality. We will denote by $`U^1`$ the inverse with respect to the star product,
$$UU^1=U^1U=1.$$
Notice that for $`\theta =\overline{\theta }=0`$, the gauge parameter is a complex function, so the gauge group is the complexification of U(1). We can write an element $`U`$ as
$$U=e^{i\mathrm{\Lambda }}=\underset{n=0}{\overset{\mathrm{}}{}}\frac{1}{n!}(i\mathrm{\Lambda })^n,$$
and then
$$U^1=e^{i\mathrm{\Lambda }},U^{}=e^{i\overline{\mathrm{\Lambda }}},U_{}^{}{}_{}{}^{1}=e^{i\overline{\mathrm{\Lambda }}}.$$
We introduce a connection superfield $`V`$ which transforms under the gauge group as
$`e^VU^{}e^VU`$
$`e^VU^1e^VU_{}^{}{}_{}{}^{1}.`$
The non commutative field strength
$$W_\alpha =\overline{D}^2(e^VD_\alpha e^V),\overline{D}_{\dot{\alpha }}W_\alpha =0,$$
transforms as
$$W_\alpha U^1W_\alpha U.$$
The action
$$𝒮_{NCYM}=d^4x(d^2\theta W_\alpha W^\alpha +\text{c.c.})$$
(20)
defines the non commutative rank 1 gauge theory. It is gauge invariant as a consequence of (16). If we set the gaugino $`\lambda `$ and the auxiliary field $`D`$ to zero, this action reduces to the bosonic non commutative action considered in . Note also that the equation of motion of the auxiliary field is $`D=0`$.
The infinitesimal gauge transformation of the connection superfield is an infinite power series with terms of the type $`V^n`$. To first order in $`V`$ it is
$$\delta V=i(\mathrm{\Lambda }\overline{\mathrm{\Lambda }})\frac{1}{2}i[(\mathrm{\Lambda }+\overline{\mathrm{\Lambda }})VV(\mathrm{\Lambda }+\overline{\mathrm{\Lambda }})].$$
This is actually the transformation in the Wess-Zumino gauge ($`V^3=0`$) . In this gauge the field strength becomes
$$W_\alpha =D_\alpha V\frac{1}{2}(VD_\alpha VD_\alpha VV).$$
Since the Wess-Zumino gauge depends on the deformation parameter $`P`$, the modified supersymmetry transformations which preserve this gauge will also depend on $`P`$. Indeed, the gaugino transformation contains the two-form field strength $`F=dA+iAA`$. Also, the supersymmetry transformation of the auxiliary field $`D`$ contains the covariant derivative of the gaugino $`\lambda =d\lambda +i(A\lambda \lambda A)`$.
It is worth noticing that the action in (20) is invariant under a non linearly realised supersymmetry transformation
$$\delta W_\alpha =\eta _\alpha $$
where $`\eta _\alpha `$ is a constant, anticommuting spinor. This leads to the interpretation of a non commutative Yang-Mills theory as a Goldstone action of partial breaking of supersymmetry.
We may now couple this action to matter chiral multiplets $`S_i`$. This can be done in two different ways, depending whether we introduce adjoint matter $`SU^1SU`$ (which is neutral in the commutative limit), or charged matter $`SSU`$. In the first case the non commutative gauge invariant coupling is
$$d^4xd^2\theta d^2\overline{\theta }Se^V\overline{S}e^V.$$
(21)
Note that we can add to the action any chiral interaction such as (15), which will be automatically gauge invariant.
If we now consider a single chiral multiplet and a vector multiplet, then the sum of the two actions (20) and (21) is known to have in the commutative limit $`N=2`$ supersymmetry. Therefore, following the discussion in section 3, the deformed theory is the first example of deformed theory with $`N=2`$ supersymmetry. This theory could in fact be reformulated using harmonic superspace which is the natural set up for $`N=2`$ Yang-mills theories .
If we introduce three chiral adjoint multiplets $`S_i,i=1,\mathrm{}3`$ with an additional self coupling
$$d^2\theta ϵ^{ijk}S_iS_jS_k+\text{c. c. },$$
(which vanishes in the commutative limit) we obtain a deformation of $`N=4`$, rank 1 supersymmetric Yang-Mills theory, which could also be reformulated in harmonic superspace . The Yang-Mills field is then a Grassmann analytic function and it is important that the star product (14) preserves Grassmann analyticity.
$`N`$-extended non commutative rank 1 gauge theories are invariant under $`N`$ non linearly realized supersymmetries, corresponding to constant shifts of the gauginos by anticommuting parameters. This is due to the fact that the cubic interactions involving gauginos, under such transformation, vary into a Moyal bracket (antisymmetrized star product), which is a total space-time derivative. This brings more evidence to the fact that such theories are closely connected to world volume brane theories which, as a microscopic description of 1/2 BPS states, have $`2N`$ supersymmetries, with half of them non linearly realized in the spontaneously broken phase .
For a charged matter field, the gauge invariant non commutative action is
$$d^4xd^2\theta d^2\overline{\theta }Se^V\overline{S}.$$
It was shown in that the phase space of a non commutative Yang-Mills theory can be mapped to the phase space of an ordinary Yang-Mills theory by a “change of variables” realised in the following way: the gauge potential of the ordinary gauge group $`A`$ is mapped into the gauge potential $`\widehat{A}(A)`$ of the non commutative (deformed) gauge group , while the gauge group parameter $`\lambda `$ is mapped into the noncommutative gauge group parameter $`\widehat{\lambda }(\lambda ,A)`$ in such a way that the respective gauge transformations satisfy
$$\widehat{A}(A)+\widehat{\delta }_{\widehat{\lambda }}\widehat{A}(A)=\widehat{A}(A+\delta _\lambda A).$$
For rank one and to first order in $`P`$, the solution to this differential equation is
$`\widehat{A}_\mu (A)=A_\mu {\displaystyle \frac{1}{2}}P^{\rho \sigma }A_\rho (_\sigma A_\mu +F_{\sigma \mu })`$
$`\widehat{\lambda }(\lambda ,A)=\lambda +{\displaystyle \frac{1}{4}}P^{\mu \nu }_\mu \lambda A_\nu .`$ (22)
These transformations can be supersymmetrized as follows. The non commutative gauge connection and gauge parameter superfields will be now denoted by $`\widehat{V}`$ and $`\widehat{\mathrm{\Lambda }}`$, while we will reserve the notation $`V`$ and $`\mathrm{\Lambda }`$ for their ordinary counterparts. The transformations then read,
$`\widehat{V}(V)=V+aP^{\mu \nu }_\mu V_\nu V+(bP^{\alpha \beta }D_\alpha VW_\beta +\text{c. c.})+`$
$`(cP^{\alpha \beta }VD_\alpha W_\beta +\text{c. c.}),`$
$`\widehat{\mathrm{\Lambda }}(\mathrm{\Lambda },V)=\mathrm{\Lambda }+d\overline{D}^2(P^{\alpha \beta }D_\alpha \mathrm{\Lambda }D_\beta V),`$ (23)
where
$`P^{\alpha \beta }`$ $`=`$ $`(\sigma _{\mu \nu })^{\alpha \beta }P^{\mu \nu },(\text{symmetric in }(\alpha ,\beta )),`$
$`W_\alpha `$ $`=`$ $`\overline{D}^2D_\alpha V,`$
$`_\nu `$ $`=`$ $`(\sigma _\nu )^{\alpha \dot{\alpha }}[D_\alpha ,D_{\dot{\alpha }}].`$
$`a,b,c,d`$ are numerical coefficients, which are uniquely fixed in order to reproduce (22). We also want to note that $`\widehat{\mathrm{\Lambda }}=\mathrm{\Lambda }`$ for $`D_\alpha \mathrm{\Lambda }=0`$, which is required by consistency. Analogously, $`\widehat{V}=V`$ for constant $`V`$.
The above results can be easily generalised to non commutative super Yang-Mills theories of arbitrary rank.
## 6 The $`\alpha ^{}0`$ limit of supersymmetric Born-Infeld action and deformed U(1) gauge theory.
In this section we will present the supersymmetric version of the $`\alpha ^{}0`$ limit of the Born-Infeld action when a $`B`$-field is turned on. This action is supposed to describe the deformed version of supersymmetric U(1) gauge theory (in the limit $`\alpha ^{}=𝒪(ϵ^{\frac{1}{2}})0`$ and slowly varying fields) , where the constant field $`B^{\mu \nu }`$ is related to the Poisson bivector by $`B^{\mu \nu }=P_{}^{1}{}_{}{}^{\mu \nu }`$ as in (14).
Let us first remind the expression obtained in the bosonic case. The Lagrangian is given by
$$_{BI}=\sqrt{det(ϵ^{\frac{1}{2}}+F)}=\sqrt{ϵ^2+\frac{ϵ}{2}F^2+\frac{1}{16}(F\stackrel{~}{F})^2}.$$
(24)
To order $`ϵ`$, the Lagrangian is readily seen to be
$$\frac{1}{4}|F\stackrel{~}{F}|+ϵ\frac{F^2}{|F\stackrel{~}{F}|}.$$
To obtain the supersymmetric Born-Infeld action, we will use the following identity ,
$$\sqrt{X^2Y}=X+Y\frac{\sqrt{X^2Y}X}{Y}=X\frac{Y}{\sqrt{X^2Y}+X},$$
where
$$X=ϵ+\frac{1}{4}F^2,Y=\frac{1}{16}\left((F^2)^2(F\stackrel{~}{F})^2\right).$$
Denoting by
$$F_\pm =\frac{1}{2}(F\pm \stackrel{~}{F})$$
the self dual and anti self dual combinations of $`F`$ in an Euclidean metric, we also have
$$F_\pm ^2=\frac{1}{2}(F^2\pm F\stackrel{~}{F}),F_+^2F_{}^2=\frac{1}{4}\left((F^2)^2(F\stackrel{~}{F})^2\right).$$
We consider a chiral spinor superfield $`W_\alpha `$ $`(\overline{D}_{\dot{\alpha }}W_\alpha =0)`$, and the chiral scalar superfield $`T`$
$$T=\overline{D}\overline{D}\overline{W}^2=\frac{1}{2}F_{}^2+\mathrm{}$$
We promote $`X`$ and $`Y`$ to superfields
$$X=ϵ\frac{1}{2}(T+\overline{T}),Y=T\overline{T}.$$
The supersymmetric Born-Infeld action is
$`_{SBI}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle d^2\theta W^2}{\displaystyle \frac{1}{2}}{\displaystyle d^2\overline{\theta }\overline{W}^2}{\displaystyle d^2\theta d^2\overline{\theta }\frac{\overline{W}^2W^2}{\sqrt{X^2Y}+X}}.`$ (25)
One has that
$$\sqrt{X^2Y}=\sqrt{ϵ^2ϵ(T+\overline{T})+\frac{1}{4}(T\overline{T})^2}.$$
The order 0 in $`ϵ`$ in the above expression is the square root of a square, so it should be understood as
$$\sqrt{\frac{1}{4}(T\overline{T})^2}=\pm \frac{1}{2}(T\overline{T})$$
(26)
depending if
$$\frac{1}{2}(T\overline{T})|_{\theta =0}=\frac{1}{4}F\stackrel{~}{F}$$
is grater or less than zero. So the $`ϵ=0`$ term of $`_{SBI}`$ is
$$\frac{1}{2}(d^2\theta W^2d^2\overline{\theta }\overline{W}^2)=\frac{1}{4}|F\stackrel{~}{F}|+\mathrm{}$$
For the order $`ϵ`$ term one gets (in the case with + sign in (26))
$$2ϵd^2\theta d^2\overline{\theta }\frac{W^2\overline{W}^2}{D^2W^2(D^2W^2\overline{D}^2\overline{W}^2)}.$$
(for the other case in (26) we exchange $`W`$ by $`\overline{W}`$ and $`D`$ by $`\overline{D}`$).
Finally we get for the $`𝒪(ϵ)`$ in $`_{SBI}`$
$`\pm ϵ{\displaystyle d^2\theta d^2\overline{\theta }W^2\overline{W}^2(\frac{1}{D^2W}+\frac{1}{\overline{D}^2\overline{W}^2})\frac{1}{(D^2W^2\overline{D}^2\overline{W}^2)}}`$
$`ϵ{\displaystyle d^2\theta d^2\overline{\theta }\frac{W^2\overline{W}^2}{(D^2W^2\overline{D}^2\overline{W}^2)}}=ϵ({\displaystyle \frac{F^2}{|F\stackrel{~}{F}|}}1)+\mathrm{}.`$ (27)
Note that the last term in (27) corresponds to a shift by $`ϵ`$ in the original Born-Infeld action (24),
$$=\sqrt{det(ϵ^{\frac{1}{2}}+F)}ϵ.$$
When the $`B`$ field is turned on ($`FF+B`$ in the bosonic action), the superfield strength $`W_\alpha =\overline{D}\overline{D}D_\alpha V`$ ($`V`$ is the gauge superfield) is shifted into $`W_\alpha L_\alpha `$ , where $`L_\alpha `$ is the spinor chiral superfield containing the $`B`$ field in its $`\theta `$-component,
$$L_\alpha =\theta ^\beta (\sigma _{\alpha \beta }^{\mu \nu }B_{\mu \nu }+ϵ_{\alpha \beta }\varphi )+\theta ^2\chi _\alpha ,$$
where we used the fact that the combination $`W_\alpha L_\alpha `$ is invariant under the superspace gauge transformation
$`V`$ $``$ $`V+U`$
$`L_\alpha `$ $``$ $`L_\alpha +\overline{D}^2D_\alpha U.`$
where $`U`$ is an arbitrary real scalar superfield. If we want now to compute the supersymmetric Born-Infeld action in the $`ϵ0`$ limit with a constant $`B`$ field, it is then sufficient to set $`\varphi =\chi _\alpha =0`$, replace $`W_\alpha `$ by $`W_\alpha \theta _\beta \sigma _{\alpha \beta }^{\mu \nu }B_{\mu \nu }`$ and then use (27).
The $`𝒪(ϵ)`$ supersymmetric version of the Born-Infeld bosonic Lagrangian,
$$\frac{F^2}{|F\stackrel{~}{F}|}$$
has a straightforward generalisation to the case of extended supersymmetry as a full superspace integral <sup>1</sup><sup>1</sup>1We observe that such generalizations are not unique unless we impose additional requirements on the Born-Infeld action such as electromagnetic duality invariance for its equations of motion .. For $`N=2`$ theories we have
$$_{SBI}(N=2)=d^4\theta d^4\overline{\theta }\frac{W^2\overline{W}^2}{D^4W^2\overline{D}^4\overline{W}^2}(\frac{1}{D^4W^2}+\frac{1}{\overline{D}^4\overline{W}^2})$$
where $`W`$ is the $`N=2`$ chiral superfield strength. This is in fact the $`\alpha ^{}0`$ limit of the $`N=2`$ supersymmetric Born-Infeld action .
For $`N=4`$ we may write an on-shell superspace action
$$_{SBI}(N=4)=d^8\theta d^8\overline{\theta }\frac{W^{4(0,4,0)}W^{4(0,4,0)}|_{\text{singlet}}}{F_+^2F_{}^2}(\frac{1}{F_+^4F_{}^2}+\frac{1}{F_{}^4F_+^2}),$$
where the $`N=4`$ superfield strength $`W^{ij}=W^{ji}`$ satisfies the following constraints,
$`W^{ij}={\displaystyle \frac{1}{2}}ϵ^{ijkl}\overline{W}^{kl}`$
$`\overline{D}_{i\dot{\alpha }}W^{jk}={\displaystyle \frac{1}{3}}(\delta _i^jW^{lk}\delta _i^k\overline{D}_{l\dot{\alpha }}W^{lj}),`$
$`D_\alpha ^iW^{jk}+D_\alpha ^jW^{ik}=0,`$
and
$$F_+^2=D_{0,2,0}^4W^{2(0,2,0)}|_{\text{singlet}},F_{}^2=\overline{D}_{0,2,0}^4W^{2(0,2,0)}|_{\text{singlet}}.$$
The indices $`(a,b,c)`$ refer to the SU(4) Dynkin labels, and “singlet” means a projection on SU(4) invariant combinations.
It is a challenging problem to show that the above actions should reproduce a deformation of the supersymmetric U(1) gauge theory.
Acknowledgments
We would like to thank Yaron Oz for a useful discussion. M. A. Ll. is thankful to the Theory Division of CERN for its kind hospitality. The work of S. F. has been supported in part by the European Commission TMR program ERBFMRX-CT96-0045 (Laboratori Nazionali di Frascati, INFN) and by DOE grant DE- FG03-91ER40662.
|
warning/0002/astro-ph0002296.html
|
ar5iv
|
text
|
# Primordial Nucleosynthesis For The New Millennium
## 1. Introduction
Among the quantitative, “hard” sciences, astronomy has traditionally been scorned, with particular disdain reserved for cosmology. No more. In the decade of the nineties the combination of an avalanche of high quality observational data and theoretical advances driven by enhaced computer (and brain) power, have succeeded in transforming cosmology to a precise science. In this introductory lecture to IAU Symposium 198 on The Light Elements and Their Evolution it is my intent to describe primordial nucleosynthesis in this precision era of cosmology and to highlight the challenges, along with some goals, for the new millennium. After a brief review of the important physics during the era of primordial nucleosynthesis in the standard, hot big bang cosmological model (SBBN), I will present an overview of the predicted primordial abundances, emphasizing the generally very small theoretical uncertainties. These will then be compared to the present best estimates (including their uncertainties) of the primordial abundances inferred from current observational data. After assessing the consistency of SBBN, I will explore what SBBN has to offer to Cosmology and to Particle Physics and, what Cosmology may teach us about SBBN. I will conclude with a summary of the key issues/problems confronting SBBN and with a wish list of topics I hope will be addressed during this meeting – and beyond.
## 2. An Early Universe Chronology
Our story begins when the Universe is a few tenths of a second old and the temperature of the cosmic background radiation has dropped to a few MeV as the Universe expanded and cooled from its denser, hotter infancy. At this time (and earlier) the density and average energy of colliding particles is so high that even the weak interactions occur sufficiently rapidly to establish equilibrium. In particular, at this stage all flavors of neutrinos ($`e,\mu ,\tau `$) are in thermal equilibrium with the cosmic background radiation (CBR) photons and with the copius electron-position pairs present ($`\nu _i+\overline{\nu }_ie^++e^{}\gamma +\gamma `$). However, as the Universe ages beyond a few tenths of a second and the temperature drops below a few MeV, these weak interactions become too slow to keep pace with the rapid expansion of the Universe and the neutrinos decouple from the CBR. The electron-type neutrinos continue to play a role in transforming neutrons into protons and, vice-versa ($`p+e^{}n+\nu _e`$, $`n+e^+p+\overline{\nu }_e`$, $`np+e^{}+\overline{\nu }_e`$). As the temperature continues to drop, less massive protons are favored over the more massive neutrons and the $`n/p`$ ratio falls (roughly as $`e^{\mathrm{\Delta }m/kT}`$, where $`\mathrm{\Delta }m`$ is the neutron – proton mass difference $`1.3`$ MeV). After the temperature drops below 800 keV or so, when the Universe is a few seconds old, even these weak interactions become too slow to keep pace with the expansion and the neutron-to-proton ratio “freezes out” (in fact, the ratio continues to decrease, albeit very slowly). All the while, neutrons and protons have been colliding, occasionally forming deuterons ($`p+nD+\gamma `$). However, the deuterons find themselves bathed in a high density background of energetic CBR photons which quickly photodissociate them ($`D+\gamma p+n`$) before they can find a proton or neutron and form the more tightly bound, less fragile, <sup>3</sup>H or <sup>3</sup>He nuclei. Since, as we shall see, there are roughly nine to ten orders of magnitude more CBR photons than nucleons in the Universe, the deuteron “stepping-stone” to further nucleosynthesis is absent until the temperature drops sufficiently low so that even in the high-energy tail of the black-body spectrum there are too few photons to prevent the deuteron from acting as a catalyst for primordial nucleosynthesis. This critical temperature, which is weakly (logarithmically) dependent on the nucleon abundance (the nucleon-to-photon ratio $`\eta `$), is roughly 80 keV. Now, at last, when the Universe is a few minutes old, Big Bang Nucleosynthesis finally commences. However, the Universe was a fatally flawed nuclear reactor, cooling and diluting rapidly as it aged. When the Universe is some 10 – 20 minutes old ($`1000`$ sec) and the temperature has dropped below 30 keV or so, the coulomb barriers preventing nuclear reactions between charged nuclei and protons and among charged nuclei become insurmountable (in the short amount of time available) and primordial nucleosynthesis comes to an abrupt end. In this all too brief but shining era there has been time to synthesize (in abundances comparable to those observed or observable) only the lightest nuclides: D, <sup>3</sup>He, <sup>4</sup>He, and <sup>7</sup>Li. In “standard” (a homogeneous Universe, expanding isotropically with the particle content of the standard model of particle physics in which there are three flavors of light ($`m`$ MeV) or massless neutrinos) big bang nucleosynthesis (SBBN) the abundances (relative to protons $``$ hydrogen) of these four nuclides are determined by only one free parameter, the present epoch nucleon-to-photon ratio $`\eta `$ ($`\eta (n_\mathrm{N}/n_\gamma )_0,\eta _{10}10^{10}\eta `$).
## 3. SBBN-Predicted Primordial Abundances
Once the deuterium photodissociation bottleneck is breached primordial nucleosynthesis begins in earnest, quickly burning D to <sup>3</sup>H, <sup>3</sup>He and <sup>4</sup>He. The higher the nucleon density, the faster D is destroyed. The same is true of <sup>3</sup>H (which, if it survives will decay to <sup>3</sup>He) and <sup>3</sup>He. Thus, the primordial abundances of D and <sup>3</sup>He are determined by the competition between the nuclear reaction rates and the universal expansion rate. The former rate depends on the overall density of the reactants – the nucleon density. Since all densities decrease as the Universe expands, it is convenient to quantify the nucleon density by specifying the ratio of the nucleon density to the photon density (measured after $`e^+e^{}`$ annihilation which enhances the Universe’s photon budget) $`\eta `$. Since observations of the cosmic background radiation (CBR) temperature (T = 2.73 K) determine the present density of CBR photons, a knowledge of $`\eta `$ is equivalent to a determination of the present mass density in nucleons (“baryons” $``$ B). In terms of the density parameter $`\mathrm{\Omega }_\mathrm{B}`$ (the ratio of the mass density to the critical mass density) and the present value of the Hubble parameter (H$`{}_{0}{}^{}100h`$ km/s/Mpc), $`\eta _{10}=273\mathrm{\Omega }_\mathrm{B}h^2`$. As $`\eta `$ increases the surviving abundances of D and <sup>3</sup>He decrease; since the <sup>3</sup>He nucleus is more tightly bound than the deuteron, the decrease of the <sup>3</sup>He/H ratio with $`\eta `$ is less rapid than that of D/H.
In contrast to D and <sup>3</sup>He, the primordial abundance of <sup>4</sup>He is not reaction rate limited since the nuclear reactions building helium-4 are so rapid that virtually all neutrons available when BBN commences are incorporated into <sup>4</sup>He. As a result the <sup>4</sup>He abundance, conventionally presented as the mass fraction of all nucleons which are in <sup>4</sup>He, Y<sub>P</sub>, is neutron limited. Since the neutron-to-proton ratio is determined by the competition between the (charged-current) weak interactions which mediate the transformation of neutrons into protons (and, vice-versa) and the universal expansion rate, Y<sub>P</sub> is sensitive to the universal expansion rate at the time the $`n/p`$ ratio “freezes” and when the deuterium photodissociation barrier disappears. Since the universal expansion rate is controlled by the total energy density, Y<sub>P</sub> provides an important test of cosmology and of particle physics in the early Universe (Steigman, Schramm & Gunn 1977). It should be noted that Y<sub>P</sub> is not entirely insensitive to the nucleon density since the higher $`\eta `$, the earlier the photodissociation barrier is overcome. At earlier times when the temperature is higher, fewer neutrons have been transformed into protons and are available for incorporation into <sup>4</sup>He. As a result, Y<sub>P</sub> increases logarithmically with $`\eta `$.
There is no stable nucleus at mass-5 and this presents a gap in the road to the synthesis of nuclei heavier than <sup>4</sup>He. In order to bridge the gap nuclear reactions must occur among nuclei with two or more nucleons. But, the abundances of D, <sup>3</sup>H, and <sup>3</sup>He are small and the coulomb barriers (especially between <sup>3</sup>He and <sup>4</sup>He and between <sup>4</sup>He and <sup>4</sup>He) suppress these reactions as the Universe expands and cools. As a result, there is very little “leakage” to nuclei beyond mass-4; as a corollary, virtually all the <sup>4</sup>He formed, survives. The only heavier nucleus produced primordially in an abundance comparable to that observed (or, even, observable with current technology) is <sup>7</sup>Li, whose BBN abundance is some 4 – 5 orders of magnitude smaller than that of D and <sup>3</sup>He. The absence of a stable nucleus at mass-8 provides another gap preventing the production of astrophysically interesting abundances of any heavier nuclei.
As will become clear in our subsequent discussion, the “interesting” range of $`\eta `$ is $`\eta _{10}=110`$ ($`\mathrm{\Omega }_\mathrm{B}h^2=0.0040.037`$), so we focus our discussion here on values of $`\eta `$ in this range. In the current precision era of BBN most of the nuclear reactions relevant to the synthesis of the light elements have been measured to reasonable accuracy at energies directly comparable to the thermal energies at the time of primordial nucleosynthesis (e.g., see Nollett, this volume). As a result, the theoretical uncertainties in the BBN-predicted abundances are generally quite small. For $`\eta `$ in the above range, the 1$`\sigma `$ uncertainties in D/H and <sup>3</sup>He/H vary from 8 – 10%. Since <sup>4</sup>He is most sensitive to the very well measured weak interaction rates, the error in SBBN-predicted Y<sub>P</sub> is very small (0.2 – 0.5% or, $`\sigma _\mathrm{Y}`$ = 0.0005 – 0.0011). In contrast, larger uncertainties, of order 12 – 21%, afflict the predicted primordial abundance of <sup>7</sup>Li.
Since this Symposium devoted much discussion to <sup>7</sup>Li, and space-limitations here prevent me from discussing all the light elements in detail, I will concentrate in the following on the two key light elements, deuterium and helium-4. In Figure 1 is shown the relation between the BBN-predicted abundances of D and <sup>4</sup>He. The band going from upper left to lower right represents the $`\pm 2\sigma `$ range of uncertainties in the primordial abundances ((D/H)<sub>P</sub> and Y<sub>P</sub>). Low D/H (high $`\eta `$) corresponds to high Y<sub>P</sub> and high D/H (low $`\eta `$) corresponds to low Y<sub>P</sub>. This anticorrelation will be very important when we confront the predictions of SBBN with the observational data.
## 4. Precise (Accurate?) Primordial Abundances
To test SBBN and fully exploit the opportunities it offers to constrain cosmology (e.g., the baryon density) and particle physics (e.g., new particles with weak or weaker-than-weak interactions) requires that observational data be used to pin down the primordial abundances of the light elements to precisions as good as (or, better than) those of the SBBN predictions. As we approach the new millennium there is good news along with some bad news. The good news is that new detectors on ever larger telescopes which cover the spectrum from radio to x-ray energies and beyond are providing very high quality data, leading to inferred abundances of high statistical accuracy. Furthermore, the abundances of the light elements are determined from observations which differ from element to element in the telescopes and techniques employed as well as in the astrophysical sites explored. As a result, insidious correlated errors between and among the various element abundances are unlikely to be a problem. The good news is also responsible for the bad news. Since the statistical errors have become so small, systematic errors now tend to dominate the uncertainties in the derived primordial abundances. As Bob Rood has said during this Symposium, estimating systematic errors is an oxymoron. When a potential source of systematic error is identified, observations can (and should) be designed to eliminate or bound its contribution to the error budget. It is a pointless and potentially misleading exercise to “estimate” the magnitude of unidentified systematic errors. In part to remind us that our precise abundance determinations may not be accurate, and in part to challenge our observational colleagues who have done such a magnificent job of reducing the statistical errors, I will try to focus on the potential sources of systematic uncertainty (when I can identify them) in the following overview of the current observational status.
### 4.1. Deuterium
As J. Linsky (this volume) has reminded us, the deuterium abundance in the local interstellar medium (the local interstellar cloud: LIC) is known very accurately: (D/H)$`{}_{\mathrm{LIC}}{}^{}=1.5\pm 0.1\times 10^5`$ (Linsky 1998). Since deuterium is only destroyed during the evolution of the Galaxy (Epstein, Lattimer & Schramm 1976), the LIC abundance provides a lower bound to its primordial (pre-Galactic) value. This bound is strong enough to bound the nucleon density from above ($`\eta _{10}<10`$; $`\mathrm{\Omega }_\mathrm{B}h^2<0.04`$), ensuring that baryons cannot “close” the Universe ($`\mathrm{\Omega }_\mathrm{B}1`$), nor even dominate its present mass density ($`\mathrm{\Omega }_\mathrm{B}\mathrm{\Omega }_\mathrm{M}0.30.4`$). Thus, local observations of deuterium, combined with the assumption of the correctness of SBBN (which we must test), already reaps great rewards: the mass-energy density of the Universe must be dominated by unseen (“dark”) non-baryonic matter. To go beyond (in the quest for the primordial deuterium abundance) we must look for observing targets which are less evolved than the LIC. The presolar nebula is one such site. From solar system observations of <sup>3</sup>He reported by G. Gloeckler (this volume), it is possible to infer the presolar deuterium abundance (Geiss & Reeves 1972; Geiss & Gloecker 1998): (D/H)$`{}_{}{}^{}=1.9\pm 0.5\times 10^5`$. Although marginally higher than the LIC abundance, the larger errors prevent us from using this determination to improve on our previous bounds from the LIC. What this result does indicate is that there has been very little (if any) evolution in the D-abundance in the solar vicinity of the Galaxy in the last 4.5 Gyr. This is consistent with a large class of Galactic chemical evolution models discussed by M. Tosi (this volume) which point to only a modest overall destruction of primordial deuterium by a factor of 2 – 3 (Tosi et al. 1998). If this theoretical estimate is combined with the LIC abundance, we may estimate the primordial abundance: (D/H)$`{}_{\mathrm{P}}{}^{}2.65.1\times 10^5`$ ($`2\sigma `$). Although possibly model dependent, this estimate is in remarkable agreement with the 2 – 3 determinations of D/H in high-redshift, low-metallicity (hence very nearly primordial) Ly-$`\alpha `$ absorbers illuminated by background QSOs described by D. Tytler and S. Levshakov (this volume). The data and analysis of Burles & Tytler (1998a,b: BT) suggests that (D/H)$`{}_{\mathrm{P}}{}^{}=2.94.0\times 10^5`$ ($`2\sigma `$). Notice that the 1$`\sigma `$ uncertainty in the observationally determined primordial abundance, $`8`$%, is impedance-matched to the $`8`$% SBBN theoretical uncertainty cited earlier. However, lest we risk dislocating a shoulder while patting ourselves on the back at the triumph of such wonderful data, we should not ignore the claim (Webb et al. 1997; Tytler et al. 1999) that the deuterium abundance in at least one Ly-$`\alpha `$ absorption system may be much higher. This is a reminder that while any determination of the deuterium abundance anywhere in the Universe (LIC, solar system, Ly-$`\alpha `$ absorbers, etc.) provides a lower bound to primordial deuterium, finding an upper bound is more problematic. Indeed, in some absorbing systems it may be impossible to distinguish D-absorption from that due to hydrogen in an interloping, low column density, “wrong-velocity” system. Thus, the deuterium abundance inferred from absorption-line data may only provide an upper bound to the true deuterium abundance. Since the low-Z, high-z QSO absorbing systems hold the greatest promise of revealing for us nearly unevolved, nearly primordial material, we look forward to the time when we can use the distribution of D/H values from more than a handful of such systems to eliminate – statistically – the uninvited contribution to the inferred primordial deuterium abundance from such interlopers. Keeping this in mind, in the following I will, nevertheless, use the BT determination when confronting theory with data.
### 4.2. Helium-4
In contrast to deuterium whose primordial abundance only decreases as pristine gas is incorporated into stars, stars burn hydrogen to helium. As a result, the <sup>4</sup>He observed anywhere in the Universe is an unknown mixture of primordial and stellar-produced helium. It has long been appreciated that to minimize the uncertain correction due to the debris of stellar evolution, it is best to concentrate on <sup>4</sup>He abundance determinations in the lowest-metallicity regions available. These are the low-Z, extragalactic H $`\mathrm{II}`$ regions which have been discussed by K. Olive, T. Thuan, and S. M. Viegas at this Symposium (this volume). The reader is urged to consult their papers for details; here I will merely summarize my view of the current status of the determination of the primordial <sup>4</sup>He mass fraction Y<sub>P</sub>. Several years ago Olive & Steigman (1995: OS) gathered together the data from the literature (dominated by the data assembled by Pagel et al. 1992). More recently Olive, Skillman & Steigman (1997: OSS) supplemented this with newer data (some of it, unfortunately, still unpublished). Using a variety of approaches such as the regression of Y on the oxygen and/or nitrogen abundances and the weighted means of Y in the lowest metal-abundance H $`\mathrm{II}`$ regions, OSS concluded that Y<sub>P</sub> = 0.234 $`\pm `$ 0.003 (note that, in contrast to the published (OSS) result, this value is obtained when the NW region of IZw18, suspected of being contaminated by underlying stellar absorption, is excluded from the fit, and the newer data of Izotov, Thuan and collaborators is not included). Izotov, Thuan and their collaborators (Izotov, Thuan, & Lipovetsky 1994, 1997; Izotov & Thuan 1998(IT); Thuan, this volume) have been systematically observing a mostly independent set of H $`\mathrm{II}`$ regions. Although, as with the data employed in the OS and OSS studies, they ignore the ionization correction ($`icf1`$), they take special care with the correction for collisional excitation. IT (also Thuan, this volume) find Y$`{}_{\mathrm{P}}{}^{}(\mathrm{IT})=0.244\pm `$ 0.002. Comparing the IT and OSS estimates of Y<sub>P</sub> we find that difference between the two Y<sub>P</sub> estimates far exceeds the statistical errors, suggesting systematic differences in the acquisition and/or analysis of the data samples. In a recent discussion which attempted to account for these unidentified systematic differences, Olive, Steigman & Walker (1999: OSW) combined the 2$`\sigma `$ ranges for each determination to conclude: Y<sub>P</sub> = 0.238 $`\pm `$ 0.005; at the 2$`\sigma `$ level, Y<sub>P</sub> $`0.248`$. Note, that this is also the 2$`\sigma `$ upper bound to the IT data alone. Since, as we shall see shortly, it is the upper bound which is crucial to testing the consistency of SBBN, in the following we shall adopt the IT value (and error estimate) for the primordial abundance of <sup>4</sup>He.
Recently, Viegas, Gruenwald & Steigman (1999: VGS; see Viegas & Gruenwald, this volume) have emphasized the importance of the ionization correction which has heretofore been ignored. VGS suggest that the IT helium abundance (Y<sub>P</sub>) should be reduced by 0.003 to account for unseen neutral hydrogen in regions where the helium is still ionized in H $`\mathrm{II}`$ regions ionized by young, hot, metal-poor stars. In subsequent comparisons I shall explore the implications of adopting Y$`{}_{\mathrm{P}}{}^{}(\mathrm{VGS})=0.241\pm `$ 0.002.
### 4.3. Helium-3 and Lithium-7
The cosmic history of the two other light nuclides produced in astrophysically interesting abundances during SBBN, <sup>3</sup>He and <sup>7</sup>Li, is considerably more complex than that of D or <sup>4</sup>He, which limits their utility as probes of the consistency of SBBN. <sup>3</sup>He is destroyed in the hotter interiors of all stars, but some <sup>3</sup>He does survive in the cooler, outer layers. For lower mass stars this <sup>3</sup>He survival layer increases and, indeed, newly synthesized <sup>3</sup>He is produced by incomplete hydrogen burning. The competition between destruction, survival, and synthesis complicates the Galactic history of the <sup>3</sup>He abundance. Nonetheless, since any deuterium incorporated into stars is first burned to <sup>3</sup>He, the apparent lack of enhanced <sup>3</sup>He (see Bania & Rood, this volume) argues against a very large pre-Galactic abundance of deuterium (Steigman & Tosi 1995). For further discussion of the evolution of <sup>3</sup>He see Bania & Rood (this volume).
As with <sup>3</sup>He, any <sup>7</sup>Li incorporated into stars is quickly burned away. However, fusion and spallation reactions between cosmic ray nuclei and those in the interstellar medium are a potent source of <sup>7</sup>Li (as well as of <sup>6</sup>Li, <sup>7</sup>Be, <sup>10</sup>B, and <sup>11</sup>B). It is also likely that there are stellar sources of <sup>7</sup>Li as indicated by the sample of lithium-rich red giants (V. Smith, this volume). Since the abundance of lithium in the solar system and in the interstellar medium (“here and now”) greatly exceeds that in the very metal-poor halo stars (T. Beers & S. Ryan, this volume), the latter likely provide the closest approach to a nearly primordial sample. Since a significant fraction of this Symposium is devoted to lithium, I will defer here to those other discussions except to comment that, within the theoretical and observational uncertainties, the primordial abundances inferred from the observational data are consistent with SBBN constrained by the confrontation with D and <sup>4</sup>He.
## 5. Confrontation Of SBBN With Data
Although SBBN does lead to the prediction of the abundances of D, <sup>3</sup>He, <sup>4</sup>He, and <sup>7</sup>Li, the currently best-constrained primordial abundances are those of deuterium and helium-4 which we are concentrating on in this status report. For each value of $`\eta `$, SBBN predicts a pair of (D/H)<sub>P</sub> and Y<sub>P</sub> values. Therefore, in SBBN there is a unique connection between (D/H)<sub>P</sub> and Y<sub>P</sub> which, allowing for the theoretical uncertainties discussed above, is shown as the band (solid lines) in Figure 1 going from the upper left to the lower right (2$`\sigma `$ uncertainties). Note that high-helium correlates with low-deuterium and, vice-versa. Also shown as the dotted ellipse in Figure 1 is the contour of the (independent) 2$`\sigma `$ uncertainties in the BT deuterium abundance and the IT helium-4 mass fraction.
Although the overlap between theory and data is not complete, Figure 1 shows that, at the $`2\sigma `$ level, the predictions of SBBN are consistent with current observational data. This is a dramatic success for the standard hot, big bang cosmological model. Of course it is not at all surprising that some value of $`\eta `$ may be found to provide consistency with the inferred primordial deuterium abundance. But there was no guarantee at all that the helium-4 abundance corresponding to this choice would bear any relation to its inferred primordial value. Consistency with the BT D-abundance limits the nucleon abundance to the range (2$`\sigma `$) $`\eta _{10}=4.45.9`$ or, $`\mathrm{\Omega }_\mathrm{B}h^2=0.0160.022`$. For $`\eta `$ in this range there is consistency, within the theoretical and observational uncertainties, between the SBBN-predicted and observationally inferred primordial abundances of <sup>3</sup>He and <sup>7</sup>Li as well. Four for the price of one! There is, of course, one more test – and opportunity – offered by this result. This SBBN-inferred nucleon abundance must also be consistent with present epoch estimates of the baryon density. Indeed, the SBBN-determined value of $`\mathrm{\Omega }_\mathrm{B}`$ is larger than estimates (Persic & Salucci 1992) of the “luminous” matter in the Universe suggesting that the majority of baryons are “dark”. This is good ($`\mathrm{\Omega }_\mathrm{B}>\mathrm{\Omega }_{\mathrm{LUM}}`$); the opposite would have been a disaster. This early-Universe estimate of the baryon density is in good agreement with that inferred from the X-ray cluster baryon fraction (Steigman, Hata & Felten 1999) and with the independent estimate from the Ly-$`\alpha `$ forest (Weinberg et al. 1997) discussed below.
### 5.1. What BBN May Do For Cosmology
X-ray clusters likely provide a “fair” sample of the universal baryon fraction $`f_\mathrm{B}`$ (White et al. 1993; Steigman & Felten 1995; Evrard, Metzler, & Navarro 1996) which, when combined with the SBBN-inferred baryon density $`\mathrm{\Omega }_\mathrm{B}`$, leads to a “clean” prediction, independent of detailed cosmological models, of the overall matter density $`\mathrm{\Omega }_\mathrm{M}`$. If the results presented here are combined with the determination of $`f_\mathrm{B}`$ from Evrard (1997), and with a Hubble parameter $`h=0.70\pm 0.07`$ (Mould et al. 1999), we predict $`\mathrm{\Omega }_\mathrm{M}=0.35\pm 0.08`$, in excellent agreement with several other recent, independent determinations. For example, a lower bound to the cosmic baryon density follows from the requirement that the high-redshift intergalactic medium contain enough neutral hydrogen to produce the Ly-$`\alpha `$ absorption observed in quasar spectra. According to Weinberg et al. (1997), depending on estimates of the quasar UV background intensity, this lower bound corresponds to $`\eta _{10}>3.44.9`$, in excellent agreement with the SBBN prediction based on the BT deuterium determination. Note that this lower bound from the Ly-$`\alpha `$ absorption forbids (in the context of SBBN) the primordial deuterium abundance to be any larger than $`8\times 10^5`$, largely excluding the one surviving claim of high D (Webb et al. 1997).
Indeed, if the SBBN results are combined with the magnitude-redshift data from surveys of high-redshift Type Ia supernovae (Garnavich et al. 1998; Perlmutter et al. 1999) which bound a linear combination of $`\mathrm{\Omega }_\mathrm{M}`$ and the cosmological constant $`\mathrm{\Omega }_\mathrm{\Lambda }\mathrm{\Lambda }/3`$H$`{}_{}{}^{2}{}_{0}{}^{}`$, we may also constrain the cosmological constant ($`\mathrm{\Omega }_\mathrm{\Lambda }=0.80\pm 0.20`$), the curvature ($`\mathrm{\Omega }_k1(\mathrm{\Omega }_\mathrm{M}+\mathrm{\Omega }_\mathrm{\Lambda })=0.15\pm 0.25`$), and the deceleration parameter ($`q_0=\mathrm{\Omega }_\mathrm{M}/2\mathrm{\Omega }_\mathrm{\Lambda }=0.62\pm 0.18`$).
### 5.2. What Cosmology May Do For BBN
As we have just seen, the SBBN-determined baryon density is consistent with that determined or constrained by observations of the Universe during its present or recent evolution. We may turn the argument around and ask what baryon density is suggested by non-BBN contraints, and then compare the light element abundances which correspond to that density with those inferred from the observational data. As an exercise of this sort, suppose (for reasons of “naturalness” or inflation) that the Universe is “flat”: $`\mathrm{\Omega }_\mathrm{M}+\mathrm{\Omega }_\mathrm{\Lambda }=1`$. When combined with the SN Ia magnitude-redshift data (Perlmutter et al. 1999), this suggests that $`\mathrm{\Omega }_\mathrm{M}=0.29\pm 0.07`$ (and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.71\pm 0.07`$). Now, if this mass density estimate ($`\mathrm{\Omega }_\mathrm{M}`$) is combined with with the X-ray determined cluster baryon fraction $`f_\mathrm{B}`$ (Evrard 1997; Steigman, Hata & Felten 1999), the resulting nucleon abundance is $`\eta _{10}=4.5\pm 1.5`$. Although the uncertainty is large, it is reassuring that this non-BBN estimate has significant overlap with our SBBN estimate. Indeed, for the baryon density in this range SBBN predicts: (D/H)$`{}_{\mathrm{P}}{}^{}=4.3\pm 2.3\times 10^5`$ and Y$`{}_{\mathrm{P}}{}^{}=0.245\pm 0.004`$.
### 5.3. What SBBN May Do For Particle Physics
The expansion rate of the early Universe is controlled by the density of the relativistic particles present. In the standard model at the time of BBN these are: photons, electron-positron pairs (when T $`>m_e`$) and three “flavors” of neutrinos ($`\nu _e`$, $`\nu _\mu `$, $`\nu _\tau `$) which, if “light” ($`m_\nu 1`$ MeV), are relativistic at BBN even if one or more of them may contribute to the present density of non-relativistic (“hot”) dark matter. If “new” particles were to contribute to the energy density at BBN, the increase in the density would result in an increase in the universal expansion rate, leaving less time for neutrons to transform into protons. The higher $`n/p`$ ratio at BBN would result in the production of more primordial <sup>4</sup>He (Steigman, Schramm & Gunn 1977). It is convenient (and conventional) to characterize such additional contributions to the energy density by comparing their effects to that of an additional “flavor” of (light) neutrino: $`\mathrm{\Delta }\rho \mathrm{\Delta }N_\nu \rho _\nu `$. For $`\mathrm{\Delta }N_\nu `$ small, $`\mathrm{\Delta }`$Y $`0.01\mathrm{\Delta }N_\nu `$. Notice in Figure 1 that the predicted <sup>4</sup>He abundance is a little high for perfect overlap with the observations. If $`\mathrm{\Delta }N_\nu `$ were $`<0`$, (N$`{}_{\nu }{}^{}2.8`$) the overlap would improve (e.g., Hata et al. 1995), while if $`\mathrm{\Delta }N_\nu >0`$, the overlap would be reduced until it disappeared. This is illustrated in Figure 2 which shows the Y versus D/H BBN band that would result if $`\mathrm{\Delta }N_\nu =0.2`$ (i.e., $`N_\nu =3.2`$, in contrast to the SBBN value of 3.0). Notice that due to the faster expansion, more deuterium survives being burnt away so that, for fixed $`\eta `$, the D-abundance also increases; however since D/H is a much more sensitive function of $`\eta `$, this has a much smaller effect on the Y versus D/H relation than does the increase in Y.
## 6. Conclusions And Outlook
The study of the early evolution of the Universe and, in particular primordial nucleosynthesis, has truly entered the precision era of cosmology. Precise abundances of the light nuclides are predicted and inferred from observations and the two are – apparently – in excellent agreement. As pleased as we may be at this success, it behooves us to avoid the temptation to rest on our laurels and to test this consistency ever more carefully. To this end, it doesn’t take much contemplation to identify several clouds looming on the horizon. What follows is my personal list of some problems/issues I would like to see addressed at this Symposium and beyond.
### 6.1. Problems/Issues
First consider deuterium. On the one hand, any determination of the D/H ratio anywhere, anytime provides a lower bound to the primordial abundance. On the other hand, since “wrong” velocity hydrogen can masquerade as deuterium, any observation of “deuterium” is really an upper bound to its true abundance. More data tracking the velocity structure of the absorbing features used to identify D and H and exploring variations in D/H in material with similar histories will be very valuable. More data at high-redshift and low-metallicity will be very valuable. After all, at present we are drawing profound conclusions on the basis of only two such systems.
Much remains to be done concerning the primordial abundance of <sup>4</sup>He. For the most part, the H $`\mathrm{II}`$ regions from which the helium abundance is inferred have been modelled as homogeneous spheres or plane-parallel slabs. A glance at the beautiful HST images of real H $`\mathrm{II}`$ regions reveals that they are anything but such idealizations. What are the effects of temperature and/or density inhomogeneities, and how large may they be? What of underlying stellar absorption which, if present but neglected, would lead to an underestimate of the helium abundance. And, what of the usually neglected ionization correction for neutral hydrogen and helium (Viegas, Gruenwald & Steigman 1999; see Viegas & Gruenwald, this volume)? Considering this latter work, where models of H $`\mathrm{II}`$ regions ionized by realistic spectra of young star clusters were used in a reanalysis of the IT data, a reduction in Y<sub>P</sub> of order 0.003 was suggested. A comparison with Figure 1 shows that if Y<sub>P</sub> were reduced by this amount, the overlap between theory and data would, in fact, disappear.
### 6.2. Wish List
Given the setting of this Symposium (Natal) and the proximity to the Christmas season, I’d like to conclude with my personal wish list. To avoid being greedy, I’ll only ask for two gifts.
A half-dozen or so observations of deuterium in high-z, low-Z systems along the lines-of-sight to distant quasars, with D/H determined in each (on average) to 10% or better. With such a gift, I could determine $`\eta `$ to better than 4%, predict Y<sub>P</sub> to $`<0.0007`$, and constrain $`\mathrm{\Delta }N_\nu `$ to an uncertainty less than $`\pm `$ 0.05. I’d be a very happy cosmologist indeed.
My second wish is for <sup>4</sup>He measured to 3% accuracy (or better) in each of about a dozen low-metallicity, extragalactic H $`\mathrm{II}`$ regions, with care taken to address the several problems outlined above. With such data, Y<sub>P</sub> could be fixed to better than the current level of $`\pm 0.002`$, permitting <sup>4</sup>He to be used as a baryometer ($`\mathrm{\Delta }\eta /\eta <20\%`$).
#### Acknowledgments.
Much of what I know about this subject I have learned from my collaborators and I would be remiss if I failed to thank them for their contributions. In particular, I wish to acknowledge R. Gruenwald, K. Olive, E. Skillman, M. Tosi, and S. M. Viegas and, of course, my late friend Dave Schramm. L. da Silva, M. Spite and the Scientific and Local Organizing committees deserve great credit for their efficient organization of a very enjoyable and successful meeting. In part, this work is supported at The Ohio State University by DOE grant DE–AC02–76ER–01545.
## References
Burles, S., & Tytler, D. 1998a, ApJ, 499, 699 (BT)
Burles, S., & Tytler, D. 1998b, ApJ, 507, 732 (BT)
Epstein, R. Lattimer, J., & Schramm, D. N. 1976, Nature, 263, 198
Evrard, A. E. 1997, MNRAS, 292, 289
Evrard, A. E., Metzler, C. A., & Navarro, J. F. 1996, ApJ, 469, 494
Garnavich, P. M. et al. 1998, ApJ, 509, 74
Geiss, J., & Reeves, H. 1972, A&A, 18, 126
Geiss, J., & Gloeckler, G. 1998, Space Sci.Rev., 84, 239
Hata, N., Scherrer, R. J., Steigman, G., Thomas, D., Walker, T. P., Bludman, S., & Langacker P. 1995, Phys.Rev.Lett, 75, 3977
Linsky, J. L. 1998, Space Sci.Rev., 84, 285
Izotov, Y. I., Thuan, T. X., & Lipovetsky, V. A. 1994, ApJ, 435, 647
Izotov, Y. I., Thuan, T. X., & Lipovetsky, V. A. 1997, ApJS, 108, 1
Izotov, Y. I., & Thuan, T. X. 1998, ApJ, 500, 188 (IT)
Mould, J. R. et al. 1999, ApJ, submitted (astro-ph/9909260)
Olive, K. A., Skillman, E., & Steigman, G. 1997, ApJ, 483, 788 (OSS)
Olive, K. A., &, Steigman, G. 1995, ApJS, 97, 49 (OS)
Olive, K. A., &, Steigman, G., & Walker, T. P. 1999, Physics Reports, in press (astro-ph/9905320) (OSW)
Pagel, B. E. J., Simonson, E. A., Terlevich, R. J. & Edmunds, M. 1992, MNRAS, 255, 325
Perlmutter, S. et al. 1999, ApJ, 517, 565
Persic, M., & Salucci, P. 1992, MNRAS, 258, 14P
Steigman, G., Schramm, D. N., & Gunn, J. E. 1977, Phys. Lett., B66, 202
Steigman, G., & Felten, J. E. 1995, Space Sci.Rev., 74, 245
Steigman, G., Hata, N., & Felten, J. E. 1999, ApJ, 510, 564
Steigman, G., & Tosi, M. 1995, ApJ, 453, 173
Tosi, M., Steigman, G., Matteucci, F., & Chiappini, C. 1998, ApJ, 498, 226
Tytler, D., Burles, S., Lu, L., Fan, X. M., Wolfe, A., & Savage, B. D. 1999, AJ, 117, 63
Viegas, S.M., Gruenwald, R., & Steigman, G. 1999, ApJ, 532 (in press, March 20, 2000; astro-ph/9909213)
Webb, J. K., Carswell, R. F., Lanzetta, K. M., Ferlet, R., Lemoine, M., Vidal-Madjar, A., & Bowen, D. V. 1997, Nature, 388, 250
Weinberg, D. H., Miralda-Escud$`\stackrel{´}{\mathrm{e}}`$, J., Hernquist, L., & Katz, N. 1997, ApJ490, 564
White, S. D. M., Navarro, J. F., Evrard, A. E., & Frenk, C. S. 1993, Nature, 366, 429
|
warning/0002/hep-th0002162.html
|
ar5iv
|
text
|
# Dilatonic formulation for conducting cosmic string models
## 1 The traditional formulation
Previous work on generic elastic string models has been based on a variational principle specified in terms of an action integral of the form
$$=𝑑𝒮,$$
(4)
in which the element $`d𝒮`$ represents the induced surface measure on the world sheet, i.e.
$$𝒮=\gamma ^{1/2}d\sigma ^_0d\sigma ^_1$$
(5)
where $`\sigma ^a`$ ($`a=0,1`$) are worldsheet coordinates and $`|\gamma |`$ is the determinant of the induced metric (2). The Lagrangian scalar $``$ here is a function that, in the usual formulation, depends only on the magnitude of the gauge covariant worldsheet gradient (3) of the scalar worldsheet potential $`\phi `$, i.e. it is a function only of the scalar
$$w=\kappa __0\phi _{|a}\phi ^{|a},$$
(6)
in which $`\kappa __0`$ is an adjustable constant that is used to obtain a standard normalisation as described below.
The particular kind of model originally proposed by Witten himself is the subcategory for which the dependence of $``$ on the scalar $`\phi _{|a}\phi ^{|a}`$ is linear:
$$=\frac{_1}{^2}wm^2,$$
(7)
where $`m`$ is a scale constant interpretable the Kibble mass characterising the isotropic Goto-Nambu type state of the string in the limit of vanishing current, i.e. the limit for which the scalar $`\phi `$ is uniform.
Witten’s simple linear model (7) still seems the best available for the ordinary fermionic case, for which it makes sense as the obvious first approximation in the weak current limit for which the fermions are sufficiently sparcely distributed for their effect on each other and on the string background to be negligible. Allowance for the corrections due to such effects would require a sophisticated analysis presumably based on second quantisation which has not yet been carried out.
The situation is different in the bosonic case, for which the background will be significantly affected by the presence of the condensate even in the zero current limit, a complication that is conveniently compensated by the possibility of dealing with it adequately on the basis just of a first quantised analysis. For the particular case of the simplified field theoretical model originally proposed by Witten, such an analysis has actually been carried out, and has shown that for essential purposes, such as calculation of characteristic propagation speeds of small perturbations, the linear model (7) is quite inadequate for describing the bosonic case. A satisfactory description can however be provided within the framework of the more general category of elastic string models, which were originally developed in terms of a variation principle for which the Lagrangian $``$ in (4) is given by an appropriate “equation of state” as a non linear function of the variable $`w`$ defined by (5).
Since only gradients, but not the absolute values, of $`\phi `$ are involved, such a Lagrangian function will determine a corresponding conserved particle current vector, $`z^a`$ say, in the worldsheet, according to the Noetherian prescription
$$z^a=\frac{}{\phi _{|a}},$$
(8)
which implies
$$𝒦z^a=\kappa __0\phi ^{|a},$$
(9)
(using the induced metric for internal index raising) where $`𝒦`$ is given as a function of $`w`$ by setting
$$\frac{1}{𝒦}=2\frac{d}{dw}.$$
(10)
This current $`z^a`$ in the worldheet can be represented by the corresponding tangential current vector $`z^\mu `$ on the worldsheet, where the latter is defined with respect to the background coordinates, $`x^\mu `$, by $`z^\mu =z^ax_{,a}^\mu `$.
The purpose of introducing the dimensionless scale constant $`\kappa __0`$ is to simplify macroscopic dynamical calculations by arranging for the variable coefficient $`𝒦`$ to tend to unity when $`w`$ tends to zero, i.e. in the limit for which the current is null. To obtain the desired simplification it is convenient not to work directly with the fundamental current vector $`z^\mu `$ that (in units such that the speed of light and the Dirac Planck constant $`\mathrm{}`$ are set to unity) will represent the quantized particle flux, but to work instead with a corresponding rescaled particle current $`c^\mu `$ that is got by setting
$$z^\mu =\sqrt{\kappa __0}c^\mu .$$
(11)
In terms of the squared magnitude $`\chi =c^\mu c_\mu `$ of this rescaled current $`c^\mu `$, the primary state variable $`w`$ defined by (6) will be given simply by $`w=𝒦^2\chi `$. It is to be remarked that in the gauge coupled case, i.e. if $`e`$ is non zero, there will be a corresponding electromagnetic current vector obtained by a prescription of the usual form $`j^\mu =/A_\mu `$ which simply gives $`j^\mu =ez^\mu `$ $`=e\sqrt{\kappa __0}c^\mu `$.
The complete system of dynamical equations can conveniently be expressed in terms of the surface stress momentum energy density tensor given by the formula
$$\overline{T}{}_{}{}^{\mu \nu }=2\frac{}{g_{\mu \nu }}+\eta ^{\mu \nu },$$
(12)
using the notation
$$\eta ^{\mu \nu }=\gamma ^{ab}x_{,a}^\mu x_{,b}^\nu $$
(13)
for what is interpretable as the (first) fundamental tensor of the worldsheet. Independently of the particular form of the Lagrangian, the equations of motion obtained from the action (4) will be expressible in the standard form
$$\overline{}_\mu \overline{T}{}_{}{}^{\mu }{}_{\nu }{}^{}=\overline{f}_\nu ,$$
(14)
in which $`\overline{f}_\mu `$ is the external force density acting on the worldsheet, and in which $`\overline{}_\mu `$ denotes the operator of surface projected covariant differentiation, as formally defined by
$$\overline{}{}_{}{}^{\mu }\eta ^{\mu \nu }_\nu x_{,a}^\mu \gamma ^{ab}_b$$
(15)
where $``$ is the usual operator of covariant differentiation with respect to the Riemannian background connection. When the effect of electromagnetic coupling is significant the corresponding force density $`f_\mu `$ will be given in terms of the field $`F_{\mu \nu }=A_{\nu ,\mu }A_{\mu ,\nu }`$ by $`\overline{f}_\mu =eF_{\mu \nu }z^\nu `$. The formula (12) for the string surface stress tensor $`\overline{T}^{\mu \nu }`$ (from which the surface energy density $`U`$ and the string tension $`T`$ are obtainable as the negatives of its non-vanishing eigenvalues) can be seen to give a result having the simple form
$$\overline{T}{}_{}{}^{\mu \nu }=\eta ^{\mu \nu }+𝒦c^\mu c^\nu .$$
(16)
Even if the force density $`f_\mu `$ is non zero, its contraction with the current vector $`z^\mu `$, or with the corresponding rescaled current vector $`c^\mu `$, will vanish, and hence it can be seen from the preceeding formulae that the equations of motion (14) automatically imply the surface current conservation law
$$\overline{}_\mu c^\mu =0.$$
(17)
The formulation presented above is the natural adaptation to strings of the Clebsch type variational formulation for relativistic fluid dynamics in which the requisite Lagrangian scalar is interpretable as the pressure function $`P`$ say. It is well known that the Clebsch formulation is related via a generalised Legendre transformation to a corresponding Taub type variational formulation, in which it is the flow world lines that are treated as free variables, and in which instead of the pressure function $`P`$ the relevant Lagrangian is interpretable as the relativistic mass-energy density function $`\rho `$ say. In an an analogous manner, in the elastic string case, the formulation presented above in terms of $``$ can be replaced by an equivalent dually related formulation for which, instead of the phase potential $`\phi `$, the independent variable is an approriately defined stream funcion $`\stackrel{~}{\phi }`$ say, and for which the dual Lagrangian, $`\mathrm{\Lambda }`$ say, is obtainable from the original Lagrangian function $``$ by the Legendre type transformation formula
$$\mathrm{\Lambda }=+\frac{w}{𝒦}.$$
(18)
In the timelike current range where $`w<0`$ the tension and energy density will be respectively given by $`T=`$, $`U=\mathrm{\Lambda }`$, whereas in the spacelike range where $`w>0`$ they will be given by $`T=\mathrm{\Lambda }`$, $`U=`$. In either case the extrinsic perturbation (“wiggle”) speed $`c__\mathrm{E}`$ and the longitudinal perturbation speed $`c__\mathrm{L}`$ will be given (relative to the preferred frame that exists in all except the “chiral” case) by
$$c__\mathrm{E}^{\mathrm{\hspace{0.17em}2}}=\frac{T}{U},c__\mathrm{L}^{\mathrm{\hspace{0.17em}2}}=\frac{dT}{dU}.$$
(19)
## 2 The dilatonic reformulation
The purpose of this article is to present a new formulation in terms of an alternative kind of action principle that has recently been developed in the context of ordinary relativistic fluid theory with a view to generalisation to the macroscopic treatment of superfluidity. In this new treatment, the action depends not just on the gradient of the phase variable $`\phi `$ but also on an auxiliary “dilatonic” amplitude variable $`\mathrm{\Phi }`$. The introduction of this auxiliary variable allows the kinetic term in the Lagrangian to retain its traditional homogeneously quadratic dependence on the phase gradient, while the essential non-linearity of the model is encapsulated in a potential function $`V`$ that is specified as an appropriately non-linear function of $`\mathrm{\Phi }`$, in terms of which the total Lagrangian scalar takes the generic form (1).
For an ordinary perfect fluid, the most important examples are the homogeneously quadratic case, $`V\mathrm{\Phi }^2`$ which corresponds to the case of a “dust” type fluid for which the pressure $`P`$ is zero, and the homogeneously quartic case, $`V\mathrm{\Phi }^4`$, which corresponds to the “radiation gas” case for which the pressure is related to the mass energy density $`\rho `$ by the familiar relation $`3P=\rho `$. It is of course to be remarked that in the case of an ordinary perfect fluid a Lagrangian of the simple form (1) can describe only irrotational motion, so further technical complications are needed for a fully generic treatment. However no such difficulty arises in the string case with which we are concerned here because, as remarked above, there is simply no room for rotation in a 1+1 dimensional worldsheet.
In the case of the elastic string models dealt with here, it can be seen by analogy with the fluid case that the required transformation to the standard form (1) will be obtained by taking the dilatonic amplitude $`\mathrm{\Phi }`$ to be given, as a function of $`w`$, by
$$\mathrm{\Phi }^2=\frac{\mathrm{\Phi }__0^2}{𝒦},$$
(20)
where its zero current value is given by the normalisation factor
$$\mathrm{\Phi }__0^2=\kappa __0,$$
(21)
while the required potential will have the (manifestly self dual) form
$$V=\frac{+\mathrm{\Lambda }}{2}=w\frac{d}{dw},$$
(22)
whose derivative will be given by
$$\frac{dV}{d\mathrm{\Phi }}=\frac{w\mathrm{\Phi }}{\mathrm{\Phi }__0^2}.$$
(23)
It can thereby be seen from the form (1) of the Lagrangian that the original defining relation (6) for $`w`$ will be recovered as the condition for invariance of the action with respect to $`\mathrm{\Phi }`$. Subject to this “on shell” identification, the surface stress energy momentum tensor $`\overline{T}^{\mu \nu }`$ obtained from the reformulated Lagrangian (1) will be given by the same formula (16) as in the traditional version, so it follows that the ensuing dynamical equations will also be equivalent.
The reformulation can of course be inverted. Starting from the new formulation as characterised by a Lagrangian of the form (1) for some given function $`V`$ of $`\mathrm{\Phi }`$, the corresponding expression for $``$ as a function of $`\mathrm{\Phi }`$, and hence implicitly, via (23) of $`w`$, will be given by
$$=\frac{\mathrm{\Phi }}{2}\frac{dV}{d\mathrm{\Phi }}V,$$
(24)
while the corresponding formula for the dual Lagrangian $`\mathrm{\Lambda }`$ as a function of $`\mathrm{\Phi }`$ (and hence implicitly via (20) as a function of $`\chi `$) will be given by
$$\mathrm{\Lambda }=\frac{\mathrm{\Phi }}{2}\frac{dV}{d\mathrm{\Phi }}V.$$
(25)
It is the ratio of these two quantities that determines the extrinsic “wiggle” speed $`c__\mathrm{E}`$, while the longitudinal (sound type) perturbation speed $`c__\mathrm{L}`$ will be determined by a differential relation: acccording to (19) we obtain
$$c__\mathrm{E}^{\pm 2}=\frac{\mathrm{\Lambda }}{},c__\mathrm{L}^{\pm 2}=1+4\frac{wd\mathrm{\Phi }}{\mathrm{\Phi }dw},$$
(26)
where the sign is taken to be positive, $`\pm =+`$ in the spacelike current regime where $`w>0`$, and to be negative $`\pm =`$ in the timelike current regime where $`w<0`$. It is to be observed that in each of these regimes the causality restriction $`c__\mathrm{L}^{\mathrm{\hspace{0.17em}2}}1`$ (i.e. the restriction that the “sound” speed should not excede the speed of light) implies the monotonicity condition $`d\mathrm{\Phi }/dw0`$, so that as corollary it can be seen that we shall always have $`0<\mathrm{\Phi }\mathrm{\Phi }__0`$ in the spacelike current regime and $`\mathrm{\Phi }__0\mathrm{\Phi }`$ in the timelike current regime.
## 3 Noteworthy examples
### 3.1 Zeroth (chiral) model
For describing the effect of a current produced by the Witten mechanism in the bosonic case, five kinds of simplified but increasingly accurate kinds of elastic string model have been developed over the years. The first of these is the original Witten model given by (7), while the second, third, fourth, and fifth kinds can all be expressed in the form (1) with simple explicit expressions for the function $`V`$ that will be listed below.
Before proceeding to do so however, it is important to mention the degenerate special case of what can be appropriately listed as the zeroth model, namely the “chiral” case for which the potential function $`V`$ is just a constant,
$$V=m^2,$$
(27)
where, as in (7), $`m`$ is a constant having the dimensions of mass. (This Kibble type mass parameter $`m`$ can in practice be expected to be of the same order of magnitude as the mass of the Higgs field responsible for the vacuum degeneracy with respect to which the cosmic strings under consideration arise as vortex defects.) Invariance of the action with respect to free variations of the auxiliary field $`\mathrm{\Phi }`$ evidently entails that the current in such a model is restricted to satisfy the nullity condition
$$\phi _{|a}\phi ^{|a}=0.$$
(28)
It is this “zeroth” model that is appropriate for describing the special fermionic case mentionned above, in which the only occupied states are zero-modes with a unique (according to convention exclusively left moving or exclusively right moving) orientation. In this effectively self dual case, for which it can be seen from (18) that – since the current magnitude $`\chi `$ vanishes – there is no difference between $`\mathrm{\Lambda }`$ and $``$, the stream function $`\stackrel{~}{\phi }`$ will be identifiable (for suitable normalisation) with the phase potential $`\phi `$ of which it is the dual. (In much of the relevant literature the stream function $`\stackrel{~}{\phi }`$ is denoted for simplicity by the letter $`\psi `$, but since this usage is redundant in the self dual “chiral” case the symbol $`\psi `$ is available therein for other purposes, and has been used instead for the dilatonic amplitude that is denoted here by $`\mathrm{\Phi }`$.)
### 3.2 First (Witten) model
It is to be remarked that the Lagrangian (7) for the first model is obtainable from that of the zeroth (“chiral”) model with fixed potential (27), simply by replacing the variable amplitude $`\mathrm{\Phi }`$ by a constant value, i.e. imposing a restraint of the form
$$\mathrm{\Phi }=\mathrm{\Phi }__0,$$
(29)
thereby relaxing the nullity restriction. For the non-degenerate Witten model obtained in this way “sound” travels at the speed of light, i.e $`c__\mathrm{L}=1`$. (This is the string analogue of the Zeldovich model representing the “stiff” limit case for an ordinary fluid.)
In so far as the bosonic case is concerned, Peter’s analysis has shown, as remarked above, that the Witten mechanism leads to behaviour that is not correctly described by this Witten string model (7), since – unlike what occurs in this first string model – it turns out that $`c__\mathrm{L}`$ will really be not just less than $`1`$ but even less than the “wiggle” speed $`c__\mathrm{E}`$, at least when the current amplitude is small.
### 3.3 Second (transonic) model
The simplest case for which both kinds of perturbation are subluminal is the “transonic” case for which they are the same, $`c__\mathrm{L}=c__\mathrm{E}=1`$. A model of this second type has the convenient property that – unlike the first (Witten) kind, but like the zeroth (“chiral”) kind ) itsdynamical equations are exactly integrable in a flat spacetime background and it can be shown using this property or otherwise that this ‘transonic” model can provide a good description of the averaged effect of small wiggles in an underlying string of the simple Nambu Goto kind. In the traditional formulation, the Lagrangian for this “transonic” string model is expressible in the form
$$=m\sqrt{m^2+w}$$
(30)
in which $`m`$ is a fixed mass parameter as introduced above. In this case the transition to the dilatonic formulation is made by taking
$$\mathrm{\Phi }^2=\frac{m\mathrm{\Phi }__0^2}{\sqrt{m^2+w}}.$$
(31)
The result for the potential function $`V`$ is in this second (“transonic”) case is thereby obtained from (22) in the form
$$V=\frac{m^2}{2}\left(\frac{\mathrm{\Phi }^2}{\mathrm{\Phi }__0^2}+\frac{\mathrm{\Phi }__0^2}{\mathrm{\Phi }^2}\right).$$
(32)
### 3.4 Third (polynomial) model
For the purpose of representing the effect of a current arising from Witten’s “superconductivity” mechanism , a more accurate description, at least in the weak current limit, can be obtained by using a third kind of model having a Lagrangian function of the polynomial form
$$=m^2\frac{w}{2}\left(1+\frac{w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right),$$
(33)
in which as well as the original “Kibble” mass parameter $`m`$ an independent “Witten” constant mass parameter $`m_{}`$ is also involved. In a model of this third kind the transition to the dilatonic formulation is made by taking
$$\mathrm{\Phi }^2=\mathrm{\Phi }__0^2\left(1\frac{2w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right),$$
(34)
and the corresponding formula for the potential function is
$$V=m^2+\frac{m_{}^{\mathrm{\hspace{0.17em}2}}}{8}\left(\frac{\mathrm{\Phi }^2}{\mathrm{\Phi }__0^2}1\right)^2.$$
(35)
### 3.5 Fourth (rational) model
On the assumption that the $`m_{}`$ is relatively small compared with $`m`$ the preceding model will indeed be characterised by supersonic wiggle propagation, $`c__\mathrm{E}>c__\mathrm{L}`$, in the weak current ($`w0`$) limit, in accordance with results of detailed numerical, analysis of the Witten mechanism, but to obtain a model that gives a qualitatively realistic description for larger currents, at least in the spacelike case, $`w>0`$, a formula of the non polynomial but still rational form
$$=m^2\frac{w}{2}\left(1+\frac{w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right)^1,$$
(36)
was found to be more satisfactory. In a model of this fourth kind the transition to the dilatonic formulation is made by taking
$$\mathrm{\Phi }^2=\mathrm{\Phi }__0^2\left(1+\frac{w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right)^2,$$
(37)
and the corresponding formula for the potential function is
$$V=m^2+\frac{m_{}^{\mathrm{\hspace{0.17em}2}}}{2}\left(\frac{\mathrm{\Phi }}{\mathrm{\Phi }__0}1\right)^2.$$
(38)
It is to be remarked that although the Lagrangian (36) is more complicated than its polynomial predecessor (33) the ensuing merely quadratic formula (38) for the potential is actually simpler than its quartic predecessor (35).
### 3.6 Fifth (logarithmic) model
To obtain a qualitatively satisfactory description not only in the spacelike but also the timelike current regime (where $`w<0`$) it was found to be necessary resort to the use of a non-rational equation of state of the form
$$=m^2m_{}^{\mathrm{\hspace{0.17em}2}}\mathrm{ln}\left\{1+\frac{w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right\}$$
(39)
where $`m_{}`$ is another fixed mass parameter. (In order for this to agree in the weak field limit with the preceeding third and fourth kinds of model, this new mass parameter would have to be related to that of these preceding examples by $`2m_{}^{\mathrm{\hspace{0.17em}2}}=m_{}^{\mathrm{\hspace{0.17em}2}}`$.) In a model of this fifth – and, for the treatment of the Witten mechanism, most realistic – kind, the transition to the dilatonic formulation is made by taking
$$\mathrm{\Phi }^2=\mathrm{\Phi }__0^2\left(1+\frac{w}{m_{}^{\mathrm{\hspace{0.17em}2}}}\right)^1,$$
(40)
and the corresponding formula for the potential function is
$$V=m^2+\frac{m_{}^{\mathrm{\hspace{0.17em}2}}}{2}\left(\frac{\mathrm{\Phi }^2}{\mathrm{\Phi }__0^2}1\mathrm{ln}\left\{\frac{\mathrm{\Phi }^2}{\mathrm{\Phi }__0^2}\right\}\right).$$
(41)
|
warning/0002/hep-lat0002031.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Lattice QCD continues to maintain an important role in the search for the physics of color confinement. The lattice regulator maintains gauge invariance at all costs. Dynamical variables are group elements rather than elements of a Lie algebra. As a consequence many of the topological features that are prominent candidates for elucidating the physics of confinement have natural lattice definitions. These include U(1), Z(N), SU(N)/Z(N) monopole loops, Dirac sheets, Z(N) and SU(N) vortex sheets etc. These objects are often abundant in U(1) and SU(N) lattice gauge theories. They become singular only as one approaches the continuum limit.
Consider the case of SU(N). Further consider a multiply connected region in which all links are gauge equivalent to $`1`$ on any simply connected patch of the region, i.e. all neighboring plaquettes $`=1`$. Then the value of a Wilson loop lying in this region would take the value of a center element of SU(N) $`=e^{2\pi in/N}`$, which for $`n0`$ indicates the trapping of a vortex. The occurance and absence of vortices gives a fluctuating value that can disorder the Wilson loop and lead to an area law.
Yaffe, Tomboulis and Kovacs and Tomboulis have developed a formulation of SU(N) gauge theory that is manifestly SU(N)/Z(N) invariant. In this formulation, center elements, Z(N), multiplying each link leave the action and measure invariant. New Z(N) variables, defined on plaquettes, $`\sigma (p)`$ , carry the Z(N) degrees of freedom. This formalism, equivalent to the standard SU(N) form, allows an elegant topological classification of the SU(N)/Z(N) and Z(N) vortex configurations occuring on the lattice. See also references \[4-7\].
In this paper, we address the issues of simulating the Tomboulis formulation on a periodic lattice. Many results have followed from this formulation without doing simulations in these variables. As a first calculation, we tag Wilson loop measurements by the occurance of vortices linking the loop using a number of different vortex counters. We restrict our attention to SU(2) in this paper.
We also measure the P (projection) vortex counter in the original SU(2) formulation following \[9-17\] for comparison. Projection vortices arise in a Z(2) gauge theory derived from the original SU(2) theory by going to the maximal center gauge and then replacing links by $`\text{sgn}(\text{tr}(U_{x,\mu }))`$. The projected theory has “thin” Z(2) vortices defined on one lattice spacing. They have been found to be well correlated with center vortices and therefore a measurement of P vortices is a predictor of them.
The thrust of this paper is to study the configuration space in the Tomboulis variables on the torus. These variables are subject to constraints in a rather indirect way. We propose a constructive update algorithm which is straightforward to implement and we prove that it reaches all configurations, i.e. it is ergodic.
In section 2 we rederive the Tomboulis form of the partition function. There are configurations on the torus that give zero weight which we exhibit in section 3. In section 4 we elucidate two alternative definitions of the configuration space for the Z(2) variables $`\sigma (p)`$, the indirect definition and constructive definition. Appendix B gives the proof of their equivalence. In section 4, we discuss various vortex counters. In section 5, we measure these vortex counters for Wilson loops. This allows us to tag Wilson loops to study the disordering mechanism. In Appendix C we consider the case of anti-periodic boundary conditions.
## 2 Derivation of $`Z_{Z(2)\times SU(2)/Z(2)}`$ on a torus
Our derivation here is due to Tomboulis , and reviewed recently by Kovacs and Tomboulis for free boundary conditions. The latter paper gives a thorough pedagogical review of the formulation and the topological features. In addition they have paid close attention to visualization of vortices, both as surfaces and as curves in 3-d slices.
Consider the partition function for the Wilson action,
$`Z={\displaystyle \left[dU(b)\right]\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}\text{tr}[U(p)]\right)}.`$ (1)
Define a Z(2) variable on the links (bonds), $`\gamma (b)`$, and insert a constant into the partition function,
$`Z={\displaystyle \underset{\gamma (b)}{}}{\displaystyle \left[dU(b)\gamma (b)\right]\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}\text{tr}[U(p)]\right)}.`$ (2)
(See Appendix A for Z(2) notation, algebra, characters, delta functions, etc.) Apply the Haar invariant transformation
$`U(b)U(b)\gamma (b);Z={\displaystyle \underset{\gamma (b)}{}}{\displaystyle \left[dU(b)\right]\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}\text{tr}[U(p)]\gamma (p)\right)}.`$
Isolate the sign of the plaquette, $`\eta (p)`$, $`\text{tr}[U(p)]=|\text{tr}[U(p)]|\times \eta (p)`$, writing
$`Z={\displaystyle \underset{\gamma (b)}{}}{\displaystyle \left[dU(b)\right]\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}|\text{tr}[U(p)]|\gamma (p)\eta (p)\right)}.`$
Next introduce a new Z(2) variable, $`\sigma (p)`$, defined on plaquettes,
$`1={\displaystyle \underset{\sigma (p)}{}}\delta (\sigma (p)\times \gamma (p)\eta (p)),`$
$`Z={\displaystyle \underset{\gamma (b)}{}}{\displaystyle \left[dU(b)\right]\left\{\underset{\sigma (p)}{}\delta (\sigma (p)\times \gamma (p)\eta (p))\right\}\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}|\text{tr}[U(p)]|\sigma (p)\right)}.`$
Expand the delta function in Z(2) characters
$`\delta (\sigma (p)\eta (p)\times \gamma (p))`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\tau (p)}{}}\left\{\chi _{\tau (p)}\left(\sigma (p)\eta (p)\right)\chi _{\tau (p)}\left(\gamma (p)\right)\right\},`$
$`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\tau (p)}{}}\left\{\chi _{\tau (p)}\left(\sigma (p)\eta (p)\right)\chi _{\gamma (p)}\left(\tau (p)\right)\right\}.`$
This gives
$`Z`$ $`=`$ $`{\displaystyle \underset{\gamma (b)}{}}{\displaystyle \left[dU(b)\right]\underset{\sigma (p)}{}\underset{\tau (p)}{}}`$
$`\left\{{\displaystyle \underset{p}{}}{\displaystyle \frac{1}{2}}\chi _{\tau (p)}\left(\sigma (p)\eta (p)\right)\right\}\left\{{\displaystyle \underset{p}{}}\chi _{\gamma (p)}\left(\tau (p)\right)\right\}\mathrm{exp}\left(\beta {\displaystyle \underset{p}{}}{\displaystyle \frac{1}{2}}|\text{tr}[U(p)]|\sigma (p)\right).`$
In the second character, we can rearrange the product over plaquettes into a product over links,
$`{\displaystyle \underset{p}{}}\chi _{\gamma (p)}\left(\tau (p)\right)={\displaystyle \underset{b}{}}\chi _{\gamma (b)}\left(\tau (\widehat{}b)\right).`$
Now do the $`\gamma (b)`$ summation
$`Z`$ $`=`$ $`{\displaystyle \left[dU(b)\right]\underset{\sigma (p)}{}}`$ (3)
$`\left[{\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}{\displaystyle \frac{1}{2}}\chi _{\tau (p)}\left(\sigma (p)\eta (p)\right){\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right)\right]\mathrm{exp}\left(\beta {\displaystyle \underset{p}{}}{\displaystyle \frac{1}{2}}|\text{tr}[U(p)]|\sigma (p)\right).`$
The item in square brackets is the starting point for much of the analysis in this paper. The dynamical variables are $`\{U(b),\sigma (p)\}`$. The $`\tau (p)`$ sum can be done.
$`C(\sigma (p)\eta (p))`$ $``$ $`\left[{\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}\left(\sigma (p)\eta (p)\right){\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right)\right],`$ (4)
$`=`$ $`{\displaystyle \underset{c}{}}\delta \left(\sigma (c)\eta (c)\right)\times \{\begin{array}{c}1\\ 0\end{array}.\}\times \text{constant}`$ (7)
The constraint $`\delta \left(\tau (\widehat{}b)\right)`$ means that there must be an even number of $`\tau =1`$ plaquettes in the co-boundary of the links (i.e. the six plaquettes contiguous with the link).
The last equality needs further explanation. If a cube contains an odd number of faces with $`\eta =1`$ then by definition it contains an SO(3) monopole. Similarly if a cube contains an odd number of faces with $`\sigma =1`$ then it contains a Z(2) monopole. The delta function on the cube requires that any SO(3) monopole be paired with a Z(2) monopole at the same location. We show in Sec. 4 that for the integrand of the partition function to be different from zero it is necessary to have $`_c\delta \left(\sigma (c)\eta (c)\right)=1`$. However on the torus, this is not sufficient. There are configurations, $`\{U(b),\sigma (p)\}`$, for which the delta functions on the cube are unity yet the integral vanishes. We denote these configurations “ weight = 0.” In section 3 we give examples of such configurations which are co-closed vortex sheets that wrap around periodic boundary conditions. We further show that when the $`\tau `$ sum differs from zero, it is a constant in the variable
$`\alpha (p)\sigma (p)\eta (p).`$ (8)
We denote these “weight $`=1`$” configurations. In section 4 we exhibit an update algorithm that reaches all weight $`=1`$ configurations and respects all constraints.
Restricting $`\left\{U(b)\right\}`$ and $`\left\{\sigma (p)\right\}`$, to the weight $`=1`$ configurations we obtain
$`Z={\displaystyle \underset{\sigma (p)}{\overset{^{}}{}}}{\displaystyle \left[dU(b)\right]^{}\underset{c}{}\delta \left(\sigma (c)\eta (c)\right)\mathrm{exp}\left(\beta \underset{p}{}\frac{1}{2}|\text{tr}[U(p)]|\sigma (p)\right)}.`$ (9)
Note that this form is invariant under $`U(b)\gamma (b)U(b)`$. All configurations related by this transformation are SU(2) representatives of the invariance group SO(3). The $`Z(2)`$ part is explicit in the $`\sigma (p)`$ variables.
## 3 Zero weight configurations on the torus
In this section we find configurations on the torus which have zero weight as indicated in Eqn.(7). We take periodic boundary conditions in all directions. The zero weight reflects the fact that a vortex wrapped around the torus is topologically stable. It can not be reached from a non-vortex configuration. This property is inherent in the formalism.
Our starting point is Eqn.(4). Using Eqn.(8) this becomes
$`C(\alpha (p))={\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}\left(\alpha (p)\right){\displaystyle \underset{b}{}}\delta (\tau (\widehat{}b).`$ (10)
The $`\tau `$ delta function constraint requires that the plaquettes forming the co-boundary of any link must occur in even numbers. Clearly a closed surface made by tiling with $`\tau =1`$ plaquettes, with $`\tau =+1`$ elsewhere, will satisfy all these constraints.
Now lets turn to the $`\alpha `$ variables. Consider a configuration in which $`\alpha (p)=1`$ on all plaquettes $`p_{00k\mathrm{}}^{12}`$ and $`=+1`$ elsewhere. The upper indices denote the plaquette orientation, and the lower indices are the space-time coordinates, $`(i,j,k,\mathrm{})`$. This configuration is a co-closed vortex sheet wrapped around the torus in the $`3`$ and $`4`$ directions. If $`\sigma (p)=1`$ and $`\eta (p)=\pm 1`$ everywhere on this co-closed vortex, it is a $`\sigma `$/$`\eta `$ vortex. If these two cases occur on different patches of the vortex sheet, then it is a hybrid vortex with co-closed monopole loops at the boundaries between the patches.
The $`\alpha `$ cube delta function constraints, Eqn.(7), are satisfied because cubes will either have no vortex plaquettes in common, or will have two vortex plaquettes on opposite faces. In spite of this we now show that this $`\alpha `$ configuration has zero weight in the partition function.
Consider Eqn.(10) applied to an arbitrary function of $`F(\tau (p),\alpha (p))`$. Define the set $`𝒞`$
$`{\displaystyle \underset{\tau (p)}{}}F(\tau ;\alpha ){\displaystyle \underset{b}{}}\delta (\tau (\widehat{}b){\displaystyle \underset{𝒞}{}}F(\tau ;\alpha ),`$
i.e. it is the set of all $`\tau `$ configurations with the property that $`_b\delta \left(\tau (\widehat{}b)\right)=1`$.
The set $`𝒞`$ forms a group under the multiplication defined through
$`(\tau _1\tau _2)(p)=\tau _1(p)\tau _2(p).`$ (11)
Given that $`\tau _1`$ and $`\tau _2`$ are group elements then each has an even number of negative plaquettes in the co-boundary of any link. Clearly the product will have the same property. The identity element is $`\tau (p)=1`$ for all $`p`$, and the elements are their own inverses.
Using the invariance property of the group summation:
$`{\displaystyle \underset{\tau 𝒞}{}}F(\tau _0\tau ;\alpha )={\displaystyle \underset{\tau 𝒞}{}}F(\tau ;\alpha ),`$
where $`\tau _0`$ is any element of the group $`𝒞`$.
Substituting for $`F`$ our case reads:
$`C(\alpha (p))`$ $`=`$ $`{\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}\left(\alpha (p)\right){\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right),`$
$`=`$ $`{\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}\chi _{\tau _0(p)\tau (p)}\left(\alpha (p)\right){\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right),`$
$`=`$ $`{\displaystyle \underset{p}{}}\chi _{\tau _0(p)}\left(\alpha (p)\right)\times C(\alpha (p)),`$ (13)
for any $`\tau _0𝒞`$. Therefore if we can find a group element $`\tau _0`$ for which $`_p\chi _{\tau _0(p)}\left(\alpha (p)\right)1`$ then $`C(\alpha )=0`$.
Choose $`\tau _0(p)=1`$ on the subset of plaquettes: $`\{p_{ij00}^{12}\}`$ for all $`i,j`$ and +1 elsewhere. It is obviously a member of the set $`𝒞`$. This is a closed tiled surface of $`1,2`$ plaquettes wrapping around the $`1,2`$ directions for the $`3`$ and $`4`$ coordinates fixed to $`0`$. Using the fact that $`\chi _+(+)=\chi _+()=\chi _{}(+)=+1=\chi _{}()`$, we count the number of sites where both $`\alpha =1`$ and $`\tau _0=1`$.
The $`\alpha (p)`$ configuration under consideration is a co-closed set of $`1,2`$ plaquettes wrapped around the $`3,4`$ directions and the $`\tau (p)`$ configuration is a closed tiled set of $`1,2`$ plaquettes wrapped around the $`1,2`$ directions. They have only one negative plaquette in common at position $`(0,0,0,0)`$. Therefore $`_p\chi _{\tau _0(p)}\left(\alpha (p)\right)=1`$ implying $`C(\alpha )=0`$.
## 4 $`\sigma (p)`$ configuration space
We have seen examples in the last section of the interplay between the $`\tau `$ and $`\alpha `$ configurations in finding non-zero contributions to the partition function. We are interested in simulating in the variables $`\{U(b),\sigma (p)\}`$. Thus far the allowable $`\alpha (=\sigma \times \eta )`$ configurations are defined indirectly in terms in the allowable $`\tau `$ configurations which are also indirectly defined by constraints. The corresponding simulation would also be very indirect and perhaps difficult to implement.
We propose a constructive definition of allowable $`\alpha (p)`$ configurations by building them up from “star transformations”, i.e. correlated sign flips of the $`\sigma `$ plaquettes occurring in the co-boundary of each link. Since these are constrained updates of six plaquettes, it is not clear that we can reach all allowed $`\{\sigma \}`$ (or equivalently $`\{\alpha \}`$) configurations by this method. However we show that this definition of allowed $`\alpha (p)`$ configuration is identical to the above definition. The proof is relegated to Appendix B. In this section we give a summary of the result.
Before discussing the $`\sigma `$ configurations let us first describe the link updates. This is a straightforward generalization of the link updates for $`SU(2)`$. The proposed change in a link might change the sign of the $`\eta `$ plaquettes in the co-boundary of the link. If one of these changes sign, we need to flip the sign of the corresponding $`\sigma `$ plaquette so that the $`\alpha `$ configuration is unchanged. Then the Monte Carlo step is essentially the same as for the SU(2) update.
Next consider the above mentioned “star transformations”. Our proposed update is to flip the sign of the six $`\sigma `$ plaquettes forming the co-boundary of the links.
Assume we are starting from a weight $`=1`$ configuration, Eqn.(7). It is easy to see that both these update steps will preserve the cube constraints.
In the previous section we described vortex sheets that wrap around the torus. Consider the operator constructed out of $`\mu ,\nu `$ plaquettes
$`𝒩_{\mu ,\nu }={\displaystyle \underset{pS_{\mu ,\nu }}{}}\eta (p)\sigma (p)\eta (S_{\mu ,\nu })\sigma (S_{\mu ,\nu })=\pm 1,`$
where $`S_{\mu ,\nu }`$ is a whole tiled $`\mu ,\nu `$ plane. $`𝒩_{\mu ,\nu }=\pm `$ for an even/odd number of vortices of stacked $`\mu ,\nu `$ plaquettes wrapping around the orthogonal $`\xi ,\eta `$ directions of the torus.
We start the algorithm with this $`𝒩_{\mu ,\nu }=+1`$ in all 6 planes. It is easy to see that our update algorithm preserves $`𝒩_{\mu ,\nu }`$. Hence, starting with non-zero weight configurations, we do not generate the zero weight configurations described in the last section, since this would involve $`\mathrm{\Delta }𝒩_{\mu ,\nu }0`$.
Let us define the relevant sets of configurations more carefully. Consider Eqn.(3)
$`C(\alpha )`$ $``$ $`{\displaystyle \underset{\tau (p)}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}\left(\alpha (p)\right){\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right),`$ (14)
$`=`$ $`{\displaystyle \underset{\tau 𝒞}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}\left(\alpha (p)\right),`$
$``$ $`{\displaystyle \underset{\tau 𝒞}{}}\tau ,\alpha ={\displaystyle \underset{\tau 𝒞}{}}\alpha ,\tau .`$
The third line is a shorthand for the product over characters (see Appendix B). The bracket, $`\alpha ,\tau =\pm 1`$, is negative if and only if there are an odd number of plaquettes for which both $`\alpha (p)=1`$ and $`\tau (p)=1`$.
The second line is an alternative way to specify the sum, where:
$`𝒞=\{\tau 𝒜|{\displaystyle \underset{b}{}}\delta \left(\tau (\widehat{}b)\right)=1\}.`$
Configurations form a group, Eqn.(11). $`𝒞`$ is a subgroup of the group $`𝒜`$ of all configurations, $`|𝒜|=2^{6N}`$ in number, where $`N`$ is the number of lattice sites. Restating:
> $`𝒞`$ is the group of all configurations $`\{\tau \}`$ with an even number of $`\tau =1`$ plaquettes occurring in the co-boundary of every link, i.e. forming a closed tiled surface of negative plaquettes.
There is a second group of interest,
$`\overline{𝒞}=\{\alpha 𝒜|\alpha ,\tau =1\tau 𝒞\}.`$
Restating:
> $`\overline{𝒞}`$ is the group of all configurations $`\{\alpha \}`$ for which $`C(\alpha )`$ may be different from zero. We will show that on this set we have indeed $`C(\alpha )0`$. (See appendix B.2, Proposition 2) Recall from Eqn.(10) that if we can find a single configuration $`\tau _0`$ for which $`\alpha ,\tau _01`$ then $`C(\alpha )=0`$.
Therefore $`\overline{𝒞}`$ is the group of $`\alpha `$ configurations which have weight $`=1`$ in the sense of Eqn.(7), i.e. the configurations that contribute to the partition function. Further, $`𝒞`$ is the group of $`\tau `$ configurations that form closed tiled surfaces as required by the explicit constraints in Eqn.(14). In the previous section, we found a zero weight configuration $`\alpha `$ by finding a configuration $`\tau _0`$ for which $`\alpha ,\tau _0=1`$.
The group $`𝒞`$ has only an implicit definition here. The group $`\overline{𝒞}`$ has an implicit definition in terms of this $`𝒞`$. Therefore its definition is even more indirect. Even without an explicit definition, we have been able to specify precisely those configurations $`\{\alpha \}`$ that contribute non-zero weight to the partition function.
There is a third group of interest,
$`𝒟=\{\alpha 𝒜|\alpha ={\displaystyle \alpha _{\widehat{}b_i}}\}.`$
where $`\alpha _{\widehat{}b_i}`$ refers to an individual star transformation on the $`i`$’th link $`b_i`$, and the product indicates all possible products of them.
Restating:
> $`𝒟`$ is the group of all configurations $`\{\alpha \}`$ which can be built out of products of “star transformations” starting from the identity configuration.
This is the constructive definition that is straightforward to implement in a simulation. The proof in Appendix B shows that the group $`𝒟`$ is identical to the group $`\overline{𝒞}`$. In this way we have shown that by our proposed algorithm is ergodic, reaching all configurations allowed in Eqn.(7).
Let us return to Eqn.(7) and the “cube constraints.” Using our definition, $`\alpha (p)=\sigma (p)\eta (p)`$, the constraints are written as $`_c\delta (\alpha (c))=1`$. The cube constraint simply asserts that for a configuration to give a non-vanishing contribution, every cube in the lattice must have an even number of faces with $`\alpha =1`$.
Let us suppose that a particular cube has an odd number of faces with $`\alpha =1`$. Then consider a configuration $`\tau _0`$ which takes values $`1`$ on all 6 faces of this particular cube. This is a closed surface and therefore satisfies the constraints imposed on $`\tau `$. For this case
$`{\displaystyle \underset{p}{}}\chi _{\tau _0(p)}\left(\alpha (p)\right)=1`$
and therefore by Eqn.(13) $`C(\alpha )=0`$.
The reason for signaling out this necessary constraint is that it is local. The zero weight configurations described in the last section necessarily wind around the torus.
## 5 Simulation of vortex counters
Vortices have long been considered as prime candidates for the essential dynamical variable to describe confinement. A simulation offers a tool that allows one to correlate the occurance of vortices with values of other dynamical variables. Hence as a first application we use this formalism to measure various vortex counters for Wilson loops.
Consider the SU(2) formalism with the standard Wilson action. Further consider a multiply connected region in which all links are gauge equivalent to $`1`$ on any simply connected patch of the region, i.e. all plaquettes $`=1`$. This could occur if the vortices are very dilute and have a cross section small compared to average separation.
Then the value of a Wilson loop lying in this region $`=\pm 1`$, the center of SU(2), corresponding to an even/odd number of SU(2) vortices linking the region. The occurance and absence of vortices gives a fluctuating value that can disorder the Wilson loop and lead to an area law.
In the $`SO(3)\times Z(2)`$ formulation the Wilson loop is given by
$`W[C]={\displaystyle \frac{1}{2}}\text{tr}[C]\eta _S\sigma _S_{C=S},`$
where $`\eta _S`$ and $`\sigma _S`$ are products of $`\eta `$ and $`\sigma `$ over any spanning surface.
Kovacs and Tomboulis define three vortex counters for thick vortex sheets, thin vortex sheets, and hybrid (patches of each on the sheet). Their definitions require measurements on all spanning surfaces. We measure here only the minimum spanning surface. Hence we must interpret our measurements as best we can in this limited simulation.
* Thin:
$`𝒩_{\text{thin}}\sigma _S.`$
If this value, $`=\pm 1`$, is independent of the spanning surface, then this counts thin vortices.
* Thick:
$`𝒩_{\text{thick}}\eta _S\text{sgn[tr}W(C)].`$
This object is counting something more elusive since unlike the above case, the vortex structure is spread over many lattice spacings. Nevertheless it is always possible to find a representative of SO(3) such that the $`\eta `$ vortex defines the topological linkage. The $`\eta `$ vortices can be deformed by a $`Z(2)`$ transformation of links giving different representatives of SO(3) without cost of action. One can move a linked $`\eta `$ vortex sheet so that it no longer links the Wilson loop and further even transform it away. However in this case the negative contribution will be transferred to one of the perimeter links of the Wilson loop, and it will not affect the value of $`𝒩_{\text{thick}}`$. Again if this is independent of the spanning surface, then this counts “thick” vortices.
* Hybrid:
$`𝒩_{\text{hybrid}}=𝒩_{\text{thin}}\times 𝒩_{\text{thick}}=\sigma _S\eta _S\text{sgn[tr}W(C)].`$
As one considers all spanning surfaces, the sign of $`𝒩_{\text{thin}}`$ might change. However if the sign of $`𝒩_{\text{thick}}`$ always compensates then this counts hybrid vortices.
We measure these three counters for Wilson loops, taking the minimal spanning surface. Then, for example, $`𝒩_{\text{thin}}=1`$ does not distinguish thin from hybrid, and similarly $`𝒩_{\text{thick}}=1`$ does not distinguish thick from hybrid. However they do measure the occurance of an object piercing the spanning surface responsible for sign fluctuations of the Wilson loop. Notice that $`𝒩_{\text{hybrid}}`$ is just the sign of the Wilson loop, Eqn.(5), and can be calculated in the original SU(2) formulation.
A fourth definition of vortex counter enters in the work of Del Debbio, Faber, Greensite and Olejnik. See also Refs.() Starting with the SU(2) formalism, go to the maximal center gauge which maximizes
$`{\displaystyle \underset{x,\mu }{}}|\text{tr(}U_{x,\mu })|.`$
Then consider a $`Z(2)`$ gauge theory in which the links are replaced by their $`Z(2)`$ values
$`\text{sgn}(\text{tr}(U_{x,\mu })).`$
Denote the plaquettes in this Z(2) gauge theory by $`\xi (p)`$. The negative plaquettes of this theory, $`\xi (p)=1`$ form thin vortices, residing on one lattice spacing similar to $`\sigma `$ vortices. They are denoted P (projection) vortices. The authors find evidence for a strong correlation between P vortices and thick objects, denoted center vortices, analogous to the thick vortices of Tomboulis. Their calculations proceed in the original SU(2) theory with the added observable of the P vortex counter to segregate contributions to Wilson loops.
* Projection:
$`𝒩_{\text{projection}}=\xi _S.`$
This object is independent of the spanning surface.
We also measure the P vortex counter for comparison, using the SU(2) formalism.
## 6 Numerical Results
Simulations were done on a $`12^4`$ lattice for $`\beta =2.30`$. Measurements were binned to 10 bins and jackknife errors calculated. \[thin\]: 200 ; \[thick\]: 400 ; \[hybrid\]: 200 ; \[projection\]: 1000 measurements. We monitor the coincidence of Z(2) and SO(3) monopoles which can slip due to round off error.
Let us emphasize again that in our application here these counters measure only the occurance of various objects piercing the minimal spanning surface, not the species of vortex. More general analysis will be given elsewhere.
Fig. 1 shows the fraction of Wilson loops which have an even number of vortices, $`x_e`$, as a function of the area. All the counters approach $`50\%`$ from above with similar behavior ($`x_o=1x_e`$). They are each counting different things and are not expected to be equal.
$`𝒩_{\text{hybrid}}`$ is a good reference curve since it measures the sign of the Wilson loop itself. The area law arises from a near cancelation of fluctuating values due to approximately equal occurance and absence of thin, thick or hybrid vortices.
The “thick” curve lies on top of the “hybrid” one. Hence the added factor of $`\sigma _S`$ in the hybrid counter has little effect here. We expect $`\sigma `$ vortices to be heavily suppressed for increasing $`\beta `$ since they cost action proportional to the vortex area. However at $`\beta =2.30`$ the density of Z(2) (or SO(3)) monopoles $`=0.2155(2)`$ (Random plaquette signs would give a density of $`0.5`$). Hence in spite of the near coincidence of these two curves, $`\sigma `$ vortices and $`\sigma `$ patches of hybrid vortices are important at this value of $`\beta `$. There is further evidence below.
Since Wilson loops are positive, one expects that the fraction of even loops $`x_e`$ should always dominate, as they do in all cases.
There are two special points on these curves. $`𝒩_{\text{thick}}=+1`$ by definition for a plaquette, i.e. a single “thick” vortex can not link a $`1\times 1`$ Wilson loop. Hence the fraction $`x_e=1`$. Further this gives the fraction $`x_e`$ equal for the two cases $`𝒩_{\text{thin}}`$, and $`𝒩_{\text{hybrid}}`$
The $`𝒩_{\text{projection}}`$ case gives the same result as reported in Ref .
Figs 2 - 6 shows Wilson loops, $`W`$, and the tagged Wilson loops corresponding to even: $`W_e`$ and odd: $`W_o`$ number of vortices as a function of loop area and their logarithmic derivatives.
* Fig. 2: \[Thick or thin segments piercing the minimal surface\] This vortex counter is just the sign of the Wilson loop itself. The positive and negative contributions to the Wilson loop are averaged separately. Since each component has approximately an equal and opposite asymptote, this illustrates the cancellations due to disordering and the difficulty of measuring large loops.
* Fig 3: \[Thick segment piercing the minimal surface\] By definition $`W_e`$ has an even number of thick segments piercing the minimal surface. Yet it still has an exponential fall off with area. See Fig. 6 which gives the logarithmic derivative showing that $`W_e`$ has about half the string tension of the $`W`$. The thin segments are still present and they account for the disordering.
* Fig 4: \[Thin segment piercing the minimal surface\] By definition $`W_e`$ has an even number of thin segments piercing the minimal surface. It nearly coincides with $`W`$ and hence cancellations due to thin segments not important and are lost in the noise. This curve also shows a breakdown of the correlation of the sign of $`W_o`$ and the sign of the vortex counter for all but the small loop sizes. This object is not a good predictor of the sign of the Wilson loop.
* Fig. 5: \[Projection vortices piercing the minimal surface\] For loops of area 9 and higher, the sign of the vortex counter correlates with $`W_o`$. The last four points coincide with those reported in Ref.. Comparing Fig. 3 \[thick\] and Fig.5b. \[projection\] the latter data are about a factor of 10 smaller.
* Fig. 6: \[Logarithmic derivatives\] $`W`$ and $`W_e`$ \[thick\] show a constant string tension for larger loops. We suspect that with better statistics, $`W_e`$ \[thin\] will also. $`W_e`$ \[hybrid\] shows the vanishing of the string tension if one removes the disordering mechanism completely. Larger loop areas are needed to decide if the logarithmic derivative of $`W_e`$ \[projection\] will also go to zero, or stablize which would indicate that a disordering mechanism remains.
## 7 Conclusion
We have proposed a simulation algorithm for the partition function in the $`Z(2)\times SO(3)`$ formulation. We show that the algorithm is ergodic, reaching all relevant configurations. We consider vortices which wrap around the torus. We find that these have zero weight in the partition function reflecting the fact that they are topologically stable.
Thick vortices are known to be an important factor in disordering the Wilson loop. As a first calculation, we measure various vortex counters, in order to see how they are correlated with other observables.
Acknowledgments
We are pleased to thank E. T. Tomboulis and S. Cheluvaraja for many helpful discussions. R. H. would like to thank Tom DeGrand and the Physics Department at the University of Colorado for their hospitality where part of this work was done. This work was supported in part by United States Department of Energy grant DE-FG05-91 ER 40617.
## Appendix A Appendix on Z(2) Algebra
We follow the definitions of Tomboulis and Kovacs We used b, p and c to denote the links, plaquettes and cubes respectively. Occasionally we use indices (e.g. $`p_{ijkl}^{\mu \nu }`$) to denote the objects location ($`ijkl`$) and orientation ($`\mu \nu `$). The $``$ and $`\widehat{}`$ operators have the usual meaning: the boundary and coboundary operators.
$`Z(2)`$ is the multiplicative group with two elements and we will denote it’s elements with Greek letters: $`\alpha `$, $`\beta `$ … The $`Z(2)`$ group admits two representations: $`M_+(\pm 1)=+1`$ and $`M_{}(\pm 1)=\pm 1`$. The characters of the representations are: $`\chi _{+1}(\pm 1)=+1`$ for $`M_+`$ and $`\chi _1(\pm 1)=\pm 1`$ for $`M_{}`$.
The $`Z(2)`$ delta function is defined as: $`\delta (+1)=1`$ and $`\delta (1)=0`$.
We list the basic of the properties of the characters:
$`\chi _\sigma (\tau )`$ $`=`$ $`\chi _\tau (\sigma ),`$
$`\chi _\tau (\alpha \beta )`$ $`=`$ $`\chi _\tau (\alpha )\chi _\tau (\beta ),`$
$`\delta (\alpha )`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{\tau }{}}\chi _\tau (\alpha ).`$
## Appendix B Appendix, Proof of ergodicity of the $`\sigma `$ update algorithm
In this appendix we prove that the two groups described in Sec. 4: $`𝒟`$ built up by “star transformations” on each link is identical to the group $`\overline{𝒞}`$ defined by constraints on each link. Sections B.1, B.2 and B.3 give preliminaries. Section B.4 proves the result by induction. We define intermediate groups $`𝒟_E`$ built out of star transformations on a subset of links and similarly $`\overline{𝒞}_E`$ restricted by constraints on the same subset of links. We then increase the set to $`E^{}`$ by an additional link and proceed by induction.
The partition function contains the factor, Eqns.(3, 4, 7, 10, 3, 14)
$`C[\alpha ]={\displaystyle \underset{\tau 𝒞}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}(\alpha (p))={\displaystyle \underset{\tau 𝒞}{}}\alpha ,\tau .`$
We will see that $`C[\alpha ]=0`$ for $`\alpha \overline{𝒞}`$ and $`C[\alpha ]=|𝒞|`$ $`=`$ number of elements in the set $`𝒞`$ for $`\alpha \overline{𝒞}`$, where
$`\overline{𝒞}=\{\alpha 𝒞|\alpha ,\tau =1,\tau 𝒞\},`$
and where $`𝒞`$ is defined to be:
$`𝒞=\{\tau 𝒜|{\displaystyle \underset{b}{}}\delta (\tau (\widehat{}b))=1\}𝒞=\{\tau 𝒜|\tau (\widehat{}b)=1,bB\}.`$
and where $`B`$ is the set of all links of the lattice.
Recall $`𝒞`$ is a closed tiled surface of negative $`\tau `$ plaquettes as required by the constraints in the partition function. The $`\tau `$ variables are summed, leaving the $`\alpha `$ variables. $`\overline{𝒞}`$ contains, e.g., a vortex of stacked negative $`\alpha `$ plaquettes which have non-vanishing weight in the partition function.
### B.1 Notation
$`P`$ denotes the set of all plaquettes of the lattice and $`B`$ the set of all links. A “configuration” is defined to be a function that associates an element of $`Z_2`$ to each plaquette. We will denote the configurations with Greek letters and write:
$`\alpha :PZ_2p\alpha (p)Z_2.`$
The set of all configurations is denoted by $`𝒜=\{\alpha :PZ_2\}`$. For every set $`K`$ we will denote with $`|K|`$ the number of elements in the set. $`|𝒜|=2^{6N}`$ where $`N`$ is the number of sites in the lattice.
Configurations form a group under multiplication: let $`\alpha `$ and $`\beta `$ be two configurations in $`𝒜`$. Then $`\alpha \beta `$ is defined to be:
$`\alpha \beta (p)=\alpha (p)\beta (p).`$
We denote the unit element $`1𝒜`$ which assigns $`+1`$ to all plaquettes. Elements are their own inverses.
The partition function, Eqn. 3, contains a summation over all $`\tau `$ configurations that satisfy the constraints: $`_b\delta \left(\tau (\widehat{}b)\right)=1`$. We denote this set by $`𝒞`$:
$`C[\alpha ]={\displaystyle \underset{\tau 𝒞}{}}{\displaystyle \underset{p}{}}\chi _{\tau (p)}(\alpha (p)),`$ (15)
where $`𝒞=\{\tau 𝒜|_b\delta \left(\tau (\widehat{}b)\right)=1\}`$.
Simplifying the notation in the summand we define:
$`\tau ,\alpha {\displaystyle \underset{p}{}}\chi _{\tau (p)}(\alpha (p)).`$
We will list here some properties of the bracket $`,`$ without proof. Lets take $`\alpha ,\beta ,\gamma 𝒜`$. Then:
$`\alpha ,\beta `$ $`=`$ $`\beta ,\alpha ,`$
$`\alpha \beta ,\gamma `$ $`=`$ $`\alpha ,\gamma \beta ,\gamma ,`$
$`\alpha ,\beta \gamma `$ $`=`$ $`\alpha ,\beta \alpha ,\gamma ,`$
$`\alpha ,1`$ $`=`$ $`1,\alpha =1.`$
### B.2 Some theorems regarding the subgroups of $`𝒜`$
Consider an arbitrary subgroup of $`𝒜`$ denoted $`𝒦`$, (of which $`𝒞`$ is an example).
Proposition 1: Let $`𝒦`$ be a subgroup of $`𝒜`$ and define:
$`K[\alpha ]={\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta ={\displaystyle \underset{\beta 𝒦}{}}\beta ,\alpha .`$
Then:
$`K[\alpha ]=\alpha ,\beta _0K[\alpha ]`$ (16)
for any $`\beta _0𝒦`$.
Proof: Since $`𝒦`$ is a group we have:
$`{\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta _0\beta ={\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta `$
where we used the property of the group sum that $`\{\beta _0\beta \}`$ is a rearrangement of the group elements $`\{\beta \}`$. Using the properties of the bracket we have:
$`K[\alpha ]`$ $`=`$ $`{\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta _0\beta ={\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta _0\alpha ,\beta ,`$
$`=`$ $`\alpha ,\beta _0{\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta =\alpha ,\beta _0K[\alpha ].`$
Definition: Let $`𝒦`$ be a subgroup of $`𝒜`$. We define:
$`\overline{𝒦}=\{\alpha 𝒜|\alpha ,\beta =1\beta 𝒦\}.`$
$`\overline{𝒦}`$ (e.g. $`\overline{𝒞}`$) has always at least one element, $`1`$, since $`\alpha ,1=1`$ for any $`\alpha 𝒜`$. Moreover it is easy to prove that $`\overline{𝒦}`$ is a group too. Now we note the following lemma:
Lemma 1: $`\overline{𝒜}=\{1\}`$.
Proof: Choose $`\alpha 𝒜`$ with $`\alpha 1`$. This means that there is at least one plaquette $`p_0P`$ for which $`\alpha (p_0)=1`$. Then if we take $`\beta (p)=1`$ for all $`pp_0`$ and $`\beta (p_0)=1`$ which is an element of $`𝒜`$ we see that $`\alpha `$ and $`\beta `$ have only one plaquette, $`p_0`$, on which both are $`1`$. Then
$`\alpha ,\beta ={\displaystyle \underset{p}{}}\chi _{\alpha (p)}(\beta (p))=1,`$
proving that $`\alpha \overline{𝒜}`$. Thus we proved that if $`\alpha 1`$ then $`\alpha \overline{𝒜}`$. Hence $`\overline{𝒜}=\{1\}`$.
Proposition 2: Let $`𝒦`$ be a subgroup of $`𝒜`$. If $`K[\alpha ]`$ is defined as in proposition 1 we have $`K[\alpha ]=0`$ for $`\alpha \overline{𝒦}`$ and $`K[\alpha ]=|K|`$ for $`\alpha \overline{𝒦}`$.
Proof: Using proposition 1 we have:
$`K[\alpha ]=\alpha ,\beta _0K[\alpha ]`$
for any $`\beta _0𝒦`$. Clearly if $`\alpha ,\beta _01`$ then $`K[\alpha ]=0`$. Thus $`K[\alpha ]=0`$ for all $`\alpha `$ that have at least one element $`\beta _0𝒦`$ for which $`\alpha ,\beta _01`$. This is equivalent with saying that if $`\alpha \overline{𝒦}`$ then $`K[\alpha ]=0`$, thus proving the first part.
Now consider $`\alpha \overline{𝒦}`$. This means that $`\alpha ,\beta =1`$ for all $`\beta 𝒦`$. Then:
$`K[\alpha ]={\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta ={\displaystyle \underset{\beta 𝒦}{}}1=|𝒦|.`$
This concludes the proof.
Theorem 1: Let $`𝒦`$ be a subgroup of $`𝒜`$. Then $`|𝒦||\overline{𝒦}|=|𝒜|`$.
Proof: Consider the following:
$`I={\displaystyle \underset{\alpha 𝒜}{}}K[\alpha ]=|𝒦|{\displaystyle \underset{\alpha \overline{𝒦}}{}}1=|𝒦||\overline{𝒦}|,`$
where we used proposition 2. Interchange the sums in $`I`$ and using the commutative property of the bracket:
$`I={\displaystyle \underset{\alpha 𝒜}{}}K[\alpha ]={\displaystyle \underset{\alpha 𝒜}{}}{\displaystyle \underset{\beta 𝒦}{}}\alpha ,\beta ={\displaystyle \underset{\beta 𝒦}{}}{\displaystyle \underset{\alpha 𝒜}{}}\beta ,\alpha ={\displaystyle \underset{\beta 𝒦}{}}A[\beta ].`$
where $`A[\beta ]`$ is defined exactly as $`K[\alpha ]`$ using the fact that the bracket is commutative. Now proposition 2 tells us that $`A[\alpha ]=|𝒜|`$ for $`\alpha \overline{𝒜}`$ and 0 otherwise. Using lemma 1 we have that $`A[1]=|𝒜|`$ and zero otherwise and thus we get $`I=|𝒜|`$. But since we already know that $`I=|𝒦||\overline{𝒦}|`$ we have:
$`|𝒦||\overline{𝒦}|=|𝒜|.`$
### B.3 Constructing $`𝒟`$
We already know that $`\overline{𝒞}`$ is a subgroup of $`𝒜`$. In this section we construct a subgroup of $`\overline{𝒞}`$, denoted $`𝒟`$, which, in the next section, will be shown to be the entire $`\overline{𝒞}`$.
Definition: We will call “a star configuration around link b” the following configuration:
$`\alpha _{\widehat{}b}(p)=\{\begin{array}{cc}+1& p\widehat{}b,\\ 1& p\widehat{}b.\end{array}`$
Lemma 2: For any link b we have $`\alpha _{\widehat{}b}\overline{𝒞}`$.
Proof: We can prove that $`\alpha _{\widehat{}b}`$ is a member of $`\overline{𝒞}`$ by checking that:
$`\alpha _{\widehat{}b},\tau =1,\tau 𝒞.`$
Consider any link b and any $`\tau 𝒞`$ and compute:
$`\alpha _{\widehat{}b},\tau ={\displaystyle \underset{p}{}}\chi _{\alpha _{\widehat{}b}(p)}(\tau (p))={\displaystyle \underset{p\widehat{}b}{}}\chi _1(\tau (p)){\displaystyle \underset{p\widehat{}b}{}}\chi _1(\tau (p))=\chi _1(\tau (\widehat{}b)).`$
Since $`\tau 𝒞`$ then $`\tau (\widehat{}b)=1`$ for any link b. Then:
$`\alpha _{\widehat{}b},\tau =\chi _1(\tau (\widehat{}b))=\chi _1(1)=+1.`$
for all $`\tau 𝒞`$. This proves that for any link b we have $`\alpha _{\widehat{}b}\overline{𝒞}`$.
If we take the set of all star configurations by taking all possible products between them we can generate a group, $`𝒟`$. Since all the star configurations are included in $`\overline{𝒞}`$, which is a group itself, the group $`𝒟`$ is a subgroup of $`\overline{𝒞}`$. Lets write $`𝒟`$ explicitely:
$`𝒟=\{\alpha 𝒜|\alpha ={\displaystyle \alpha _{\widehat{}b_i}}\}.`$
where the product is over all configurations reached by star transformations. In order to prove that $`𝒟`$ covers all $`\overline{𝒞}`$ we will show that $`𝒟`$ has the same number of elements as $`\overline{𝒞}`$. Since $`𝒟\overline{𝒞}`$ we have $`|𝒟||\overline{𝒞}|`$. All we have to prove now is that $`|𝒟||\overline{𝒞}|`$.
### B.4 Proof that $`\overline{𝒞}𝒟`$
We now prove that $`|𝒟||\overline{𝒞}|`$. Using theorem 1 we know that $`|𝒞||\overline{𝒞}|=|𝒜|`$. Thus we can prove that $`|𝒟||\overline{𝒞}|`$ by proving that $`|𝒞||𝒟||𝒜|`$. Lets recall the definitions of these sets:
$`𝒞`$ $`=`$ $`\{\tau 𝒜|\tau (\widehat{}b)=1,bB\},`$
$`𝒟`$ $`=`$ $`\{\alpha 𝒜|\alpha ={\displaystyle \alpha _{\widehat{}b_i}},b_iB\}.`$
As we see the set $`𝒞`$ is constructed by means of eliminating the configurations that do not obey a certain constraint whereas the set $`𝒟`$ is constructed by generating all possible combinations built from star configurations acting on the identity configuration. These sets admit a generalization as follows. Define:
$`𝒞_E`$ $`=`$ $`\{\tau 𝒜|\tau (\widehat{}b)=1,bE\},`$
$`𝒟_E`$ $`=`$ $`\{\alpha 𝒜|\alpha ={\displaystyle \alpha _{\widehat{}b_i}},b_iE\},`$
where $`EB`$. It is easy to check that both $`𝒞_E`$ and $`𝒟_E`$ are groups. In words, $`𝒞_E`$ is the set of all the configurations that obey the constraint only on the subset $`E`$ of all links and $`𝒟_E`$ is the group generated by star configurations associated only with links in $`E`$. It is obvious that $`𝒞_B=𝒞`$ and that $`𝒟_B=𝒟`$. If we prove that $`|𝒟_E||𝒞_E||𝒜|`$ we will implicitly prove that $`|𝒟||𝒞||𝒜|`$ and thus proving that $`|𝒟||\overline{𝒞}|`$.
Theorem 2: $`|𝒟_E||𝒞_E||𝒜|`$.
Proof: We will prove this using induction. The first step will be to prove this for $`E=\mathrm{}`$ and the second step will be to prove that the relation holds for $`E^{}=E\{b\}`$, where b is any link, assuming that the inequality holds for $`E`$.
#### B.4.1 Initializing
Let $`E=\mathrm{}`$. We have $`𝒞_{\mathrm{}}=𝒜`$ since there is no constraint that has to be obeyed and $`𝒟_{\mathrm{}}=\{1\}`$ since this is the group that has no star configuration in it. Then we have:
$`|𝒟_{\mathrm{}}||𝒞_{\mathrm{}}|=1\times |𝒜|=|𝒜|.`$
Therefore the inequality is verified for $`E=\mathrm{}`$.
#### B.4.2 Iterative inductive step
Assume that:
$`|𝒟_E||𝒞_E||𝒜|`$
for a certain $`E`$. Let us prove that this also holds for $`E^{}=E\{b\}`$ for any $`bE`$. We will see what happens with $`𝒞_E`$ and $`𝒟_E`$ when we increase $`E`$ by one element.
Iteration on $`𝒞_E`$
When we increase the number of elements in $`E`$, $`𝒞_E`$ grows smaller since there will be more constraints to obey. It may happen that the new constraint, namely $`\tau (\widehat{}b)=1`$, is superfluous i.e. all configuration in $`𝒞_E`$ already satisfy this constraint. In this case $`𝒞_E^{}=𝒞_E`$ and $`|𝒞_E^{}|=|𝒞_E|`$. Lets now see what happens if there is at least one element in $`𝒞_E`$ that doesn’t obey the new constraint. In this case we can break down $`𝒞_E`$ in two disjoint sets: $`𝒞_E^{}`$ the sets of all configuration $`\tau 𝒞_E`$ that obey the new constraint and $`R`$ the set of all configurations $`\tau 𝒞_E`$ that do not obey the constraint. In case that $`R\mathrm{}`$ lets take $`\tau _0R`$. Using this we can construct a one to one mapping between $`𝒞_E^{}`$ and $`R`$:
$`f:𝒞_E^{}R,f(\tau )=\tau _0\tau ,`$
$`f^1:R𝒞_E^{},f(\tau )=\tau _0\tau .`$
It is easy to check that $`f`$ is indeed one to one.
Now since there is a one to one mapping between $`𝒞_E^{}`$ and $`R`$ we have that $`|𝒞_E^{}|=|R|`$. But since they are disjoint sets of $`𝒞_E`$ and $`𝒞_E^{}R=𝒞_E`$ we have $`|𝒞_E^{}|+|R|=|𝒞_E|`$. Thus we have $`|𝒞_E^{}|=\frac{1}{2}|𝒞_E|`$.
Summing up we know that by adding a new constraint $`|𝒞_E^{}|`$ is either equal with $`|𝒞_E|`$ when the constraint is superfluous or is $`\frac{1}{2}|𝒞_E|`$ when we have at least one element in $`𝒞_E`$ that violates the constraint.
Iteration on $`𝒟_E`$
Now we will look at $`𝒟_E^{}`$. What happens when we pass from $`𝒟_E`$ to $`𝒟_E^{}`$? We add a new element in the group. Thus the group grows larger with one exception: it may happen that $`\alpha _{\widehat{}b}𝒟_E`$ although $`bE`$. Then the group will stay the same since every combination that doesn’t involve $`\alpha _{\widehat{}b}`$ is already in the group $`𝒟_E`$ and every combination that involves $`\alpha _{\widehat{}b}`$ can be generated by elements already in the group. Formally, if $`\alpha _{\widehat{}b}𝒟_E`$ then for any $`\alpha 𝒟_E`$ we have $`\alpha _{\widehat{}b}\alpha 𝒟_E`$ and thus $`𝒟_E^{}=𝒟_E`$. This proves that if $`\alpha _{\widehat{}b}𝒟_E`$ we have $`|𝒟_E^{}|=|𝒟_E|`$. It is not obvious that this case happens but we can easily construct one: take a site in the lattice and take the eight links that form its coboundary. Since the coboundary of a coboundary is nul (i.e. $`\widehat{}^2=0`$) it means that if we perform all the star transformations associated with the links in the coboundary of the site we get the identity. This means that the product of any seven of this star transformations is equal to the eighth star transformation.
Consider what happens when $`\alpha _{\widehat{}b}𝒟_E`$. Then for every element $`\alpha 𝒟_E`$ we have two elements in $`𝒟_E^{}`$: $`\alpha 𝒟_E^{}`$ and $`\alpha _{\widehat{}b}\alpha 𝒟_E^{}`$. We can actually construct two sets in $`𝒟_E^{}`$: $`𝒟_E`$ which is trivially included in $`𝒟_E^{}`$ and $`\alpha _{\widehat{}b}\times 𝒟_E=\{\alpha 𝒟_E^{}|\alpha =\alpha _{\widehat{}b}\beta ,\beta 𝒟_E\}`$. If $`\alpha _{\widehat{}b}𝒟_E`$ then these two sets do not overlap.
If they were to overlap then there is $`\alpha 𝒟_E\alpha _{\widehat{}b}\times 𝒟_E`$ with $`\alpha 𝒟_E`$ and $`\alpha =\alpha _{\widehat{}b}\beta `$ with $`\beta 𝒟_E`$. Then $`\alpha _{\widehat{}b}=\alpha \beta `$ and since both $`\alpha `$ and $`\beta `$ are members of $`𝒟_E`$ then $`\alpha _{\widehat{}b}𝒟_E`$ which contradicts our assumption.
It is easy to see that $`𝒟_E`$ and $`\alpha _{\widehat{}b}\times 𝒟_E`$ have the same number of elements and that $`𝒟_E\alpha _{\widehat{}b}\times 𝒟_E=𝒟_E^{}`$. Now since $`𝒟_E\alpha _{\widehat{}b}\times 𝒟_E=𝒟_E^{}`$ and $`𝒟_E\alpha _{\widehat{}b}\times 𝒟_E=\mathrm{}`$ we have $`|𝒟_E|+|\alpha _{\widehat{}b}\times 𝒟_E|=|𝒟_E^{}|`$. Moreover $`|𝒟_E|=|\alpha _{\widehat{}b}\times 𝒟_E|`$ and thus $`|𝒟_E^{}|=2|𝒟_E|`$.
Summing up if $`\alpha _{\widehat{}b}𝒟_E`$ then $`|𝒟_E^{}|=|𝒟_E|`$ and if $`\alpha _{\widehat{}b}𝒟_E`$ we have $`|𝒟_E^{}|=2|𝒟_E|`$.
Intermediate summary of possible cases
Thus we arrived at the conclusion that if we add another link $`b`$ to $`E`$ one of the four following things may happen:
* the new constraint is superfluous ($`|𝒞_E^{}|=|𝒞_E|`$) and $`\alpha _{\widehat{}b}𝒟_E`$ ($`|𝒟_E^{}|=2|𝒟_E|`$). Then $`|𝒞_E^{}||𝒟_E^{}|=2|𝒞_E||𝒟_E||𝒜|`$.
* the new constraint is superfluous ($`|𝒞_E^{}|=|𝒞_E|`$) and $`\alpha _{\widehat{}b}𝒟_E`$ ($`|𝒟_E^{}|=|𝒟_E|`$). Then $`|𝒞_E^{}||𝒟_E^{}|=|𝒞_E||𝒟_E||𝒜|`$.
* the new constraint is not superfluous ($`|𝒞_E^{}|=\frac{1}{2}|𝒞_E|`$) and $`\alpha _{\widehat{}b}𝒟_E`$ ($`|𝒟_E^{}|=2|𝒟_E|`$). Then $`|𝒞_E^{}||𝒟_E^{}|=|𝒞_E||𝒟_E||𝒜|`$.
* the new constraint is not superfluous ($`|𝒞_E^{}|=\frac{1}{2}|𝒞_E|`$) and $`\alpha _{\widehat{}b}𝒟_E`$ ($`|𝒟_E^{}|=|𝒟_E|`$). Then $`|𝒞_E^{}||𝒟_E^{}|=\frac{1}{2}|𝒞_E||𝒟_E|`$ and we don’t know if this is smaller or greater than $`|𝒜|`$.
If we look at the scheme above we see that if we can prove that the last case never happens than we proved our theorem. This is exactly what we prove in the following lemma:
Lemma 3: If $`\alpha _{\widehat{}b}𝒟_E`$ then the new constraint is superfluous.
Proof: If $`\alpha _{\widehat{}b}𝒟_E`$ then there is a subset $`E_0E`$ with the property $`\alpha _{\widehat{}b}=_{b^{}E_0}\alpha _{\widehat{}b^{}}`$ (this simply asserts that $`\alpha _{\widehat{}b}`$ can be written as a product of star configurations associated with a subset of links in $`E`$).
Lets denote with $`P_0`$ the set of all plaquettes that form the coboundary of $`E_0`$: $`P_0=\widehat{}E_0`$ All the plaquettes in $`P_0`$ have at least one neighboring link in $`E_0`$ and thus they are flipped at least once when you take the product $`_{b^{}E_0}\alpha _{\widehat{}b^{}}`$.
Since the whole product is equal to $`\alpha _{\widehat{}b}`$ the coboundary of $`b`$, $`\widehat{}b`$, has to be included in $`P_0`$. Otherwise we cannot flip the plaquettes around $`b`$ and we cannot form $`\alpha _{\widehat{}b}`$.
Further since the final state is $`\alpha _{\widehat{}b}`$ we know that all plaquettes $`pP_0\widehat{}b`$ are flipped an even number of times and all plaquettes $`p\widehat{}b`$ are flipped an odd number of times.
We define a function $`\kappa (p)`$ on plaquettes in $`P_0`$ that returns the number of times this plaquette is flipped. Using this it is easy to prove that $`\kappa (p)`$ is even for $`pP_0\widehat{}b`$ and odd for $`p\widehat{}b`$:
$`{\displaystyle \underset{b^{}E_0}{}}\alpha _{\widehat{}b^{}}(p)={\displaystyle \underset{b^{}E_0}{}}(1)^{ϵ_b^{}(p)}=(1)^{_{b^{}E_0}ϵ_b^{}(p)}=(1)^{\kappa (p)}`$
where $`ϵ_b^{}(p)`$ is 1 if $`p\widehat{}b^{}`$ and zero otherwise. Now since $`\left[_{b^{}E_0}\alpha _{\widehat{}b^{}}\right]=\alpha _{\widehat{}b}`$ and since $`\alpha _{\widehat{}b}=(1)^{ϵ_b(p)}`$ we have that $`\kappa (p)`$ and $`ϵ_p(b)`$ have to have the same parity proving that $`\kappa (p)`$ is even for $`pP_0\widehat{}b`$ and odd for $`p\widehat{}b`$.
Now lets take an element $`\tau 𝒞_E`$. Since $`\tau (\widehat{}b^{})=1`$ for all $`b^{}E`$ and $`E_0E`$ we have that:
$`{\displaystyle \underset{b^{}E_0}{}}\tau (\widehat{}b^{})=1`$
At the same time we can write:
$`{\displaystyle \underset{b^{}E_0}{}}\tau (\widehat{}b^{})={\displaystyle \underset{pP_0}{}}(\tau (p))^{\kappa (p)}=1`$
But since $`\kappa (p)`$ is even for $`pP_0\widehat{}b`$ and odd for $`p\widehat{}b`$ we have:
$`{\displaystyle \underset{pP_0}{}}(\tau (p))^{\kappa (p)}={\displaystyle \underset{p\widehat{}b}{}}\tau (p)=\tau (\widehat{}b)=1`$
Thus we proved that if $`\tau 𝒞_E`$ then $`\tau (\widehat{}b)=1`$ and thus the new constraint is superfluous. This proves the lemma.
Now using this lemma we see that the case where $`|𝒞_E^{}|=\frac{1}{2}|𝒞_E|`$ and $`|𝒟_E^{}|=|𝒟_E|`$ never happens and since in all the other cases our theorem holds then we proved our theorem.
### B.5 Summary
We proved using theorem 2 that $`|𝒟_E||𝒞_E||𝒜|`$ which means that $`|𝒟||𝒞||𝒜|`$. Using this and the result of theorem 1: $`|\overline{𝒞}||𝒞|=|𝒜|`$ we see that $`|𝒟||\overline{𝒞}|`$ but we already know that since $`𝒟\overline{𝒞}`$ we have that $`|𝒟||\overline{𝒞}|`$. The only possible solution to this is that $`|𝒟|=|\overline{𝒞}|`$ and since $`𝒟\overline{𝒞}`$ we have that $`𝒟=\overline{𝒞}`$.
This is the result that we were looking for. Now we can say that:
$`C[\alpha ]=\{\begin{array}{c}|𝒞|\alpha 𝒟\\ 0\alpha 𝒟\end{array}`$ (17)
where $`𝒟`$ is the group formed by taking all possible products between the star configurations.
## Appendix C Appendix, Antiperiodic boundary conditions
We have shown that vortices wrapped around the periodic boundary conditions have zero weight. However we point out how these configurations can instead be weight $`=1`$ and the above configurations weight $`=0`$ by a minor change in the formulation. Consider
$`Z={\displaystyle [dU]e^{\left(\beta _p\frac{1}{2}\text{tr}[U(p)]\right)}}{\displaystyle [dU]e^{\left(\beta _{pP}\frac{1}{2}\text{tr}[U(p)]\beta _{pP}\frac{1}{2}\text{tr}[U(p)]\right)}}.`$ (18)
where $`P`$ is a co-plane of sign flipped terms in the action. This is known as antiperiodic or twisted boundary conditions in the literature.
Antiperiodic boundary conditions amounts to nothing more than a change in the action. The derivation of the partition function in terms of $`SU(2)/Z_2`$ and $`Z_2`$ variables can be generalized. In our notation , Eqn.(18) becomes
$`Z`$ $`=`$ $`{\displaystyle [dU]\underset{\sigma \eta \overline{𝒞}}{}e^{\left(\beta _p\frac{1}{2}|\text{tr}[U(p)]|\sigma (p)\right)}}`$ (19)
$`{\displaystyle [dU]\underset{\sigma \eta \overline{𝒞}}{}e^{\left(\beta _{pP}\frac{1}{2}|\text{tr}[U(p)]|\sigma (p)\beta _{pP}\frac{1}{2}|\text{tr}[U(p)]|\sigma (p)\right)}}.`$
Define:
$`\sigma ^{}=\sigma _0\sigma \sigma =\sigma _0\sigma ^{}`$
where $`\sigma _0(p)=1`$ for $`pP`$ and $`+1`$ elsewhere.
We can simplify $`Z`$, Eqn.(19)
$`Z`$ $`=`$ $`{\displaystyle [dU]\underset{\sigma ^{}\sigma _0\eta \overline{𝒞}}{}e^{\left(\beta _p\frac{1}{2}|\text{tr}[U(p)]|\sigma ^{}(p)\right)}}={\displaystyle [dU]\underset{\sigma ^{}\eta \sigma _0\overline{𝒞}}{}e^{\left(\beta _p\frac{1}{2}|\text{tr}[U(p)]|\sigma ^{}(p)\right)}}.`$
This looks exactly like the partition function for the periodic boundary conditions except that $`\eta \sigma ^{}\sigma _0\overline{𝒞}`$ instead of $`\overline{𝒞}`$.
|
warning/0002/hep-th0002259.html
|
ar5iv
|
text
|
# One particular approach to the non-equilibrium quantum dynamics
## Introduction: Jaynes-Gibbs principle
The objective of this talk is to present a novel approach to a non-equilibrium dynamics of quantum fields jizba-tututi . This approach is based on the Jaynes-Gibbs maximum entropy principle jaynes , which, in contrast to other approaches in use calzetta-hu ; kadanoff-baym ; chinesse ; eboli-jackiw-pi , constructs a density matrix $`\rho `$ directly from the experimental/theoretical initial-time data (e.g. pressure, density of energy, magnetization, ionization rate, etc.). We illustrate our method on the $`\varphi ^4`$ theory with the $`O(N)`$ internal symmetry in the large $`N`$ limit, provided that the non-equilibrium medium in question is translationally invariant.
To start, we consider the following definition of expectation value of some dynamical operator $`A`$: $`A=\mathrm{Tr}(\rho A)`$, where the trace is taken with respect to an orthonormal basis of physical states describing the ensemble at some initial time $`t_i`$. Let us consider the information (or Shannon) entropy $`S[\rho ]=\mathrm{Tr}(\rho \mathrm{ln}\rho )`$ jaynes . According to the Jaynes-Gibbs principle, we have to maximize $`S[\rho ]`$ subject to constrains imposed by the expectation value of certain experimental/theoretical observables: $`P_i[\mathrm{\Phi },\mathrm{\Phi }](x_1,\mathrm{})=g_i(x_1,\mathrm{}),i=1,\mathrm{}n,`$ where the operators $`P_i[\mathrm{\Phi },\mathrm{\Phi }]`$, in contrast to thermal equilibrium, need not to be constants of the motion; space-time dependences are allowed. The maximum of the entropy determines the density matrix with the least informative content.
$`\rho ={\displaystyle \frac{1}{𝒵(\lambda _i)}}\mathrm{exp}\left({\displaystyle \underset{i=1}{\overset{n}{}}}{\displaystyle \underset{j}{}d^4x_j\lambda _i(x_1,\mathrm{})P_i[\mathrm{\Phi },\mathrm{\Phi }]}\right),`$ (1)
where $`\lambda _i`$ are the Lagrange multipliers to be determined. The ‘partition function’ $`𝒵`$ is $`𝒵(\lambda _i)=\mathrm{Tr}\left\{\mathrm{exp}\left(_{i=1}^n_jd^4x_j\lambda _i(x_1,\mathrm{})P_i[\mathrm{\Phi },\mathrm{\Phi }]\right)\right\}.`$ In the previous equations the time integration is not present at all (i.e. $`g_i(\mathrm{})`$ are especified at the initial time $`t_i`$). In case when the constraint functions $`g_i(\mathrm{})`$ are known over some gathering interval $`(\tau ,t_i)`$ the correspondent integration $`_\tau ^{t_i}𝑑t`$ should be present in $`\rho `$. The Lagrange multipliers $`\lambda _i`$ might be eliminated if one solves $`n`$ simultaneous equations: $`g_i=\delta \mathrm{ln}𝒵/\delta \lambda _i`$. The solution can be formally written as $`\lambda _i=\delta S_G[g_1,\mathrm{},g_n]_{max}/\delta g_i`$.
## Off-equilibrium dynamical equations
In this section we briefly introduce the off-equilibrium dynamical (or Dyson-Schwinger) equations. For simplicity we illustrate this on a single scalar field $`\mathrm{\Phi }`$ coupled to an external source $`J`$ described by the action $`S^{}[\mathrm{\Phi }]=S[\mathrm{\Phi }]+J\mathrm{\Phi }`$. Associated with this action we have the functional equation of motion jizba-tututi ; chinesse :
$$\frac{1}{Z[J]}\frac{\delta S}{\delta \mathrm{\Phi }}\left[\mathrm{\Phi }_\alpha =i\frac{\delta }{\delta J}\right]Z[J]=J_\alpha ,$$
(2)
with $`Z[J]=\mathrm{Tr}\{\rho T_C\mathrm{exp}(i_Cd^4xJ(x)\mathrm{\Phi }(x)\}`$ being the generating functional of Green’s functions. Here $`C`$ is the Keldysh-Schwinger contour which runs along the real axis from $`t_i`$ to $`t_f`$ ($`t_f>t_i`$, $`t_f`$ is arbitrary) and then back to $`t_i`$. In (2) we have associated with the upper branch of $`C`$ the index $`\mathrm{`}\mathrm{`}+\mathrm{"}`$ and with the lower one the index $`\mathrm{`}\mathrm{`}\mathrm{"}`$ (in the text we shall denote the indices $`+/`$ by Greek letters $`\alpha ,\beta `$).
Let us define the classical field $`\varphi _\alpha `$ as the expectation value of the field operator in the presence of $`J`$: i.e. $`\varphi _\alpha =\mathrm{\Phi }_\alpha `$. Defining the generating functional of the connected Green’s functions as $`Z[J]=\mathrm{exp}(iW[J])`$, the two-point Green’s function is $`G_{\alpha \beta }(x,y)=\frac{\delta ^2W}{\delta J_\alpha (x)\delta J_\beta (y)}=iT_C\mathrm{\Phi }(x)\mathrm{\Phi }(y)+i\mathrm{\Phi }(x)\mathrm{\Phi }(y)`$. Eq.(2) is the first one of an infinite hierarchy of equations for Green functions. Further equations can be obtained from (2) by taking successive variations with respect to $`J`$. True dynamical equations are then obtained if one substitutes the physical condition $`J=0`$ into equations obtained.
To reflect the effects of the density matrix in the Dyson-Schwinger equations it is necessary to construct the corresponding boundary conditions.<sup>1</sup><sup>1</sup>1 Let us remind that at equilibrium the corresponding boundary conditions are the Kubo-Martin-Schwinger (KMS) conditions. Using the cyclic property of the trace together with the Baker-Campbell-Hausdorff relation: $`e^ABe^A=_{n=0}^{\mathrm{}}\frac{1}{n!}C_n`$, (where $`C_0=B`$ and $`C_n=[A,C_{n1}]`$), and setting $`A=\mathrm{ln}(\rho )`$ and $`B=\mathrm{\Phi }(x_1)`$ with $`x_{10}=t_i`$ we obtain the generalized KMS conditions:
$`\mathrm{\Phi }(x_1)\mathrm{}\mathrm{\Phi }(x_n)=\mathrm{\Phi }(x_2)\mathrm{}\mathrm{\Phi }(x_n)\mathrm{\Phi }(x_1)+_{k=1}^{\mathrm{}}\frac{1}{k!}\mathrm{\Phi }(x_2)\mathrm{}\mathrm{\Phi }(x_n)C_k(x_1).`$ So namely for the two-point Green function we have $`G_+(x,y)=G_+(x,y)+_{k=1}^{\mathrm{}}\frac{1}{k!}\mathrm{Tr}\{\rho \mathrm{\Phi }(x)C_k(x)\}.`$ As an example of the latter relation we can choose the particular situation when $`\rho =\mathrm{exp}(\beta H)/𝒵`$, in which case we get the well known KMS condition: $`G_+(𝐱;t,𝐲;0)=G_+(𝐱;ti\beta ,𝐲;0)`$.
## Example: out-of-equilibrium pressure
In order to apply our previous results let us consider the $`\varphi ^4`$ theory with the $`O(N)`$ internal symmetry in the large $`N`$ limit (also the Hartree-Fock approximation). It is well known that, in this limit only two-point Green’s functions are relevant jizba-tututi ; eboli-jackiw-pi ; amelinocamelia-pi . The Dyson-Schwinger equations for $`G_{\alpha \beta }`$ are automatically truncated and reduce to the Kadanoff-Baym equations kadanoff-baym : $`\left(\text{ }\text{ }\text{ }\text{ }\text{ }+m_0^2+\frac{i\lambda _0}{2}G_{\alpha \alpha }(x,x)\right)G_{\alpha \beta }(x,y)=\delta (xy)(\sigma _3)_{\alpha \beta },`$ where $`\sigma _3`$ is the Pauli matrix; $`\lambda _0`$ and $`m_0`$ are, respectively, the bare coupling and the bare mass of the theory. If the system is translationally invariant the Fourier transform solves the Kadanoff-Baym equations and the correspondingfundamental solution reads: $`G_{\alpha \beta }(k)=\frac{(\sigma _3)_{\alpha \beta }}{k^2+^2+iϵ(\sigma _3)_{\alpha \beta }}2\pi i\delta (k^2+^2)f_{\alpha \beta }(k),`$ where the (finite) $``$ is $`^2=m_0^2+i\frac{\lambda _0}{2}G_{++}(0)`$. Function $`f_{\alpha \beta }(k)`$ must be determined through the generalized KMS conditions. Let us now choose the constraint to be used. Keeping in mind that we are interested in a system which is invariant under both spatial and temporal translations, we choose the constraint $`g(𝐤)=\stackrel{~}{}(𝐤)`$, where $`\stackrel{~}{}=\omega _ka^{}(𝐤)a(𝐤)`$, with $`\omega _k=\sqrt{𝐤^2+^2}`$ (notice that in the large $`N`$ limit the Hamiltonian is always quadratic in the fields). The corresponding density matrix then reads
$`\rho ={\displaystyle \frac{1}{𝒵(\beta )}}\mathrm{exp}\left({\displaystyle \frac{d^3𝐤}{(2\pi )^32\omega _k}\beta (𝐤)\stackrel{~}{}(𝐤)}\right),`$ (3)
with $`\frac{\beta (𝐤)}{(2\pi )^32\omega _k}`$ being the Lagrange multiplier to be determined. According to the maximum entropy principle we find that $`\beta (𝐤)`$ fulfils equation
$`g(𝐤)={\displaystyle \frac{V}{(2\pi )^3}}{\displaystyle \frac{\omega _k}{e^{\beta (𝐤)\omega _k}1}},`$ (4)
where $`V`$ denotes the volume of the system. Eq.(4) can be interpreted as the density of energy per mode. Similarly as in the case of thermal equilibrium, $`\beta (𝐤)`$ could be interpreted as “temperature” with the proviso that different modes have now different “temperatures”.
The generalised KMS conditions in this case are $`G_+(k)=e^{\beta (𝐤)k_0}G_+(k)`$, and so the corresponding $`f_{++}`$ reads: $`f_{++}=[\mathrm{exp}(\beta (𝐤)\omega _k)1]^1`$. Let us now consider a particular system in which $`g(𝐤)=\frac{V}{(2\pi )^3}\mathrm{exp}(\omega _k/\sigma )`$. In this case $`\sigma `$ is the physical parameter which, as we shall see below, can be interpreted as a “temperature” parameter. This particular choice corresponds to a system where the lowest frequency modes depart from equilibrium, while the high energy ones obey the Bose-Einstein distribution (typical situation in many non-equilibrium media, e.g. plasma heated up by ultrasound waves, hot fusion or ionosphere ionised by sun). In terms of the parameter $`\sigma `$ the Lagrange multiplier may be written as $`\beta (𝐤)=\frac{1}{\sigma }+\frac{1}{\omega _k}_{n=1}^{\mathrm{}}\frac{(1)^{n+1}}{n}\mathrm{exp}(n\omega _k/\sigma )`$. Notice that when $`\omega _k\sigma `$, $`\beta \sigma ^1`$, and we may see that $`f_{++}`$ approaches to the Bose-Einstein distribution with temperature $`\sigma `$. However, when $`\omega _k\sigma `$ the latter interpretation fails. Instead of the parameter $`\sigma `$, it may be useful to work with the expectation value of $`\beta (𝐤)`$:
$$\beta =\frac{d^3𝐤\beta (𝐤)e^{\omega _k/\sigma }}{d^3𝐤e^{\omega _k/\sigma }}=\frac{1}{\sigma }+\frac{\underset{n=1}{\overset{\mathrm{}}{}}\frac{(1)^{n+1}}{n(n+1)}K_1((n+1)/\sigma )}{K_2(/\sigma )},$$
(5)
where $`K_n`$ is the Bessel function of imaginary argument of order $`n`$. An interesting feature of Eq.(5) is that it is actually insensitive to the value of $``$ which is important if one wants to use $`1/\beta `$ as a “temperature”. The actual behaviour of $`\beta `$ is depicted in Fig.1
Let us now consider the renormalized expression for the expectation value of the energy momentum tensor jizba-tututi :
$$\theta _{\mu \nu }_{\mathrm{ren}}=N\frac{d^dk}{(2\pi )^d}k_\mu k_\nu [G_{++}(k)G(k)]i\frac{Ng_{\mu \nu }\delta m^2}{4}\frac{d^dk}{(2\pi )^d}[G_{++}(k)+G(k)],$$
with $`G`$ being the usual ($`T=0`$) causal Green function and $`\delta m^2=^2m_r^2`$ with $`m_r`$ being the ($`T=0`$) renormalized mass. The pressure per particle, in the high “temperature” expansion (i.e. for large $`\sigma `$ or small $`\beta `$) for the system described by the density matrix (3) may be worked out either in terms of $`\sigma `$, using the Mellin transform technique jizba-tututi ; landsman-weert :
$$P(\sigma )=\frac{1}{3N}\theta _i^i_{\mathrm{ren}}=\frac{\sigma ^4}{\pi ^2}\frac{\sigma ^2^2}{2\pi ^2}+\frac{\lambda _r}{8}\left(\frac{\sigma ^2^2}{64\pi ^4}\frac{\sigma ^43}{4\pi ^4}\right)+𝒪(\text{ln}(/\sigma );\lambda _r^2),$$
or in terms of $`1/\beta `$ using the Padé approximationjizba-tututi :
$`P(\beta )`$ $`=`$ $`0.0681122\beta ^40.0415368\beta ^2^2+\lambda _r(0.000647\beta ^4`$
$`+`$ $`0.0000164\beta ^2^2)+𝒪(^2\text{ln}(\beta );\lambda _r^2).`$
It is interesting to compare the previous two results with the high-temperature expansion of the same system in thermal equilibrium amelinocamelia-pi :
$$P(T)=\frac{T^4\pi ^2}{90}\frac{T^2^2}{24}+\frac{T^3}{12\pi }+\frac{\lambda _r}{8}\left(\frac{T^4}{144}\frac{T^3}{24\pi }+\frac{T^2^2}{16\pi ^2}\right)+𝒪\left(\text{ln}\left(\frac{}{T4\pi }\right)\right).$$
Particularly, the leading “temperature” coefficients in the first two expansions approximate to a very good accuracy the usual Stefan-Boltzmann constant for scalar theory. The latter vindicates the interpretation of $`\sigma `$ and $`1/\beta `$ as temperatures for high energy modes. The behaviour of both $`P(T)`$ and $`P(\sigma )`$ are shown in Fig.1.
## summary and outlook
One of the main advantages of the Jaynes-Gibbs construction is that one starts with constraints imposed by experiment/theory. The constraints directly determine the density matrix with the least informative content (the least prejudiced density matrix which is compatible with all information one has about the system) and consequently the generalized KMS conditions for the Dyson-Schwinger equations. We applied our method on a toy model system ($`O(N)\lambda \varphi ^4`$ theory), in the translationally invariant medium. The method presented, however, has a natural potential to be extensible to more general systems. Particularly to media where the translational invariance is lost. As an example we can mention systems which are in local thermal equilibrium. For such systems it is well knownjaynes ; landsman-weert that equilibrium $`\beta `$ must be replaced by $`\beta (𝐱)`$ (i.e. temperature which slowly varies with position). Obviously one may receive this result from the outlined Jaynes–Gibbs principle almost for free. Work on more complex systems is now in progress.
|
warning/0002/cond-mat0002131.html
|
ar5iv
|
text
|
# On the evaluation of the specific heat and general off-diagonal n-point correlation functions within the loop algorithm
## I Introduction
Numerical investigations of strongly correlated electron systems gained considerable importance in the last decade. The evaluation of non-diagonal correlation function and dynamical response function plays a major role in the context of correlated electron systems . On the other hand, there are only very few investigations of non-diagonal and/or higher-order correlation function in the context of quantum spin-systems. Indeed, it has been realized only recently, that non-diagonal correlation function might be calculated efficiently within the loop-algorithm . The loop-algorithm has established itself as the method of choice for quantum-Monte Carlo (MC) simulations of non-frustrated quantum spin systems.
The key observation here is the fact, that local updating dynamics in a MC simulation creates strongly correlated configurations for gapless quantum spin systems at low temperatures. Since the samples are then not statistically independent, the statistical error bars do decay only very slowly with the number of samples. One way to state this problem is to say, that the autocorrelation time $`\tau _{auto}`$ for the samples of spin-configurations created with the MC-walk increases (in generally exponentially) at low temperatures.
Most efficient MC procedures implement consequently global update dynamics. Examples of these procedures are the clusters algorithms . Designed to circumvent the critical slowing down, these methods have been intensively used to study classical statistical systems near critical points, where the problem of large $`\tau _{auto}`$ is very severe.
The loop algorithm can be considered as a generalization of classical clusters algorithms to quantum models. In fact, it gives a prescription on how global updates can be performed in quantum systems. As we will see this prescription lays on the geometric interpretation of the transformation from a quantum system to a statistical model of oriented loops. The MC procedure can be implemented then directly on the loops. It has the advantage that the updating dynamics defined on the loops generates statistically nearly independent configurations. The autocorrelation time is therefore about just MC step and the corresponding operators can be measured at every MC step avoiding both ’waiting times’ and substantial increments of the variance (statistical error bars).
In addition, a loop has another remarkable property; starting from an allowed spin configuration, constructing a loop and then flipping all spins in one loop (flipping the orientation of the loop) one obtains a new allowed configuration. This observation allows to compute the expectation value of operators not only in one configuration per MC step but in all configuration related to it by flipping any number of given loops. This procedure is usually called improved estimator .
The purpose of this work is to extend the algorithm to the computation of higher order (and non-diagonal) correlations functions. As we will see it involves dealing with two or more loop contributions. In particular we will focus on the specific heat $`c_V`$, which, in the past, has been considered a major challenge for Monte-Carlo simulations . We will show, that the direct evaluation of the higher-order (non-diagonal) correlation functions contributing to $`c_V`$ allows for improved estimators and such to gain one order of magnitude in computational efficiency. The method that we presented is valid in any dimension.
## II The Loop Algorithm
A nice review of the loop algorithm can be found in Ref. . Here we start with a short introduction in order to introduce the notation used further on for the evaluation of higher-order correlation functions.
The loop algorithm is most easily understood in the checkerboard picture for a discrete number of Trotter slices $`N_T`$; the generalization to continuous Trotter time is straightforward. This picture, which is based on the Suzuki-Trotter decomposition, describes in a graphical way how the interacting spin system wave function evolves in discrete imaginary time.
The Suzuki-Trotter formula maps a quantum spin system in dimension $`d`$ onto a classical spin in dimension $`d+1`$. The partition function of the original quantum spin model is hereby written in terms of the trace of a product of transfer matrices defined in the classical model.
To illustrate the method we consider an inhomogeneous one-dimensional XXZ model $`H=H_1+H_2`$ on a bipartite chain of length $`L`$:
$`H_1={\displaystyle \underset{i=2m}{}}H_i,H_2={\displaystyle \underset{j=2m+1}{}}H_j`$
$`H_i={\displaystyle \frac{J_i^{XY}}{2}}\left(S_i^+S_{i+1}^{}+S_i^{}S_{i+1}^+\right)+J_i^ZS_i^ZS_{i+1}^Z,`$where the sign of the term $`J_i^{XY}`$ has been choose to be negative by an appropriate rotation of the spins on one of the two sublattices. This is always possible on a bipartite lattice and allows for positive transfer matrix elements (absence of the sign problem). The decomposition $`H=H_1+H_2`$ allows for the use of Totter-Suzuki formula for the representation of the partition function $`Z=\text{Tr}\left[\mathrm{exp}(\frac{\beta }{N_T}H)\right]^{N_T}`$,
$`Z=\text{Tr}{\displaystyle \underset{n=1}{\overset{N_T}{}}}{\displaystyle \underset{\alpha _n}{}}\varphi _{\alpha _n}^{(n)}|\mathrm{exp}(\mathrm{\Delta }\tau H_1)\mathrm{exp}(\mathrm{\Delta }\tau H_2)|\varphi _{\alpha _{n+1}}^{(n+1)}+O(\mathrm{\Delta }\tau ^2),`$where $`\mathrm{\Delta }\tau =\beta /N_T`$. Here we have introduced representations of the unity operator $`_{\alpha _n}|\varphi _{\alpha _n}^{(n)}\varphi _{\alpha _n}^{(n)}|`$ in between any of the $`N_T`$ imaginary time slices.
Since $`H_1`$ and $`H_2`$ are sum of local operators that commute with each other, we may write the wave function as the product of the local basis in say z-component of spin, $`|\varphi _{\alpha _n}^{(n)}=_i|\sigma _i`$, with $`\sigma _i=,`$.
In the checkerboard lattice the interaction between two consecutive pairs of spins is graphically denoted by shaded plaquettes (see Fig. 1). There are two spins interacting per plaquette so a 4x4 transfer matrix $`T_i`$ can be defined in each plaquette, which depends only one the coupling constants.
For the XXZ-model the transfer matrix $`T_i`$ is in the basis $`(|,,|,,|,,|,)`$:
$`T_i=\text{e}^{\frac{\mathrm{\Delta }\tau J_i^z}{4}}\left(\begin{array}{cccc}\mathrm{exp}(\frac{\mathrm{\Delta }\tau J_i^Z}{2})& 0& 0& 0\\ 0& \mathrm{cosh}(\frac{\mathrm{\Delta }\tau J_i^{XY}}{2})& \mathrm{sinh}(\frac{\mathrm{\Delta }\tau J_i^{XY}}{2})& 0\\ 0& \mathrm{sinh}(\frac{\mathrm{\Delta }\tau J_i^{XY}}{2})& \mathrm{cosh}(\frac{\mathrm{\Delta }\tau J_i^{XY}}{2})& 0\\ 0& 0& 0& \mathrm{exp}(\frac{\mathrm{\Delta }\tau J_i^Z}{2})\end{array}\right).`$The partition function $`Z`$ is then, up to terms order $`O(\mathrm{\Delta }\tau ^2)`$, the trace of a product of transfer matrices:
$`Z=\text{Tr}\left[\mathrm{exp}(\beta H)\right]=\text{Tr}{\displaystyle \underset{n=1}{\overset{N_T}{}}}\left({\displaystyle \underset{i=2m}{}}T_i\right)\left({\displaystyle \underset{j=2m+1}{}}T_j\right).`$As a next step beyond this standard representation of $`d`$-dimensional quantum models in terms of classical statistical systems we expanded the transfer matrices $`T_i=_\gamma p_i^{(\gamma )}M^{(\gamma )}`$ in terms of certain matrices $`M^{(\gamma )}`$ such that the weight $`p_i^{(\gamma )}0`$ are non-negative. This is, in general, not possible for all models. For the XXZ with $`J_i^{XY}J_i^Z`$ we can choose:
$`M^{(1)}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right),M^{(2)}=\left(\begin{array}{cccc}1& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\end{array}\right),M^{(3)}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 1& 1& 0\\ 0& 1& 1& 0\\ 0& 0& 0& 0\end{array}\right),`$where $`p_i^{(1)}=\frac{1}{2}(\mathrm{exp}(\mathrm{\Delta }\tau J_i^z/2)+\mathrm{exp}(\mathrm{\Delta }\tau J_i^{XY}/2))\mathrm{exp}(\mathrm{\Delta }\tau J_i^Z/4)`$, $`p_i^{(2)}=\frac{1}{2}(\mathrm{exp}(\mathrm{\Delta }\tau J_i^z/2)\mathrm{exp}(\mathrm{\Delta }\tau J_i^{XY}/2))\mathrm{exp}(\mathrm{\Delta }\tau J_i^Z/4)`$ and $`p_i^{(3)}=\frac{1}{2}(\mathrm{exp}(\mathrm{\Delta }\tau J_i^z/2)+\mathrm{exp}(\mathrm{\Delta }\tau J_i^{XY}/2))\mathrm{exp}(\mathrm{\Delta }\tau J_i^Z/4)`$. We then obtain for the partition function
$$Z=\text{Tr}\underset{n=1}{\overset{N_T}{}}\underset{i=2m}{}\left(\underset{\gamma }{}p_i^{(\gamma )}M^{(\gamma )}\right)\underset{j=2m+1}{}\left(\underset{\gamma }{}p_j^{(\gamma )}M^{(\gamma )}\right)$$
(1)
Eq. (1) can be interpreted in a geometrical way (see Fig. 2). In the checkerboard picture the $`M^{(\gamma )}`$ matrices can be understood as different ways in which the worldlines can be broken in every plaquette and are usually called breakups. By taking one breakup per every plaquette we force the worldlines into closed paths which we call directed loops (see Fig. 3). A directed loop therefore follows the worldline of an up-spin when it evolves in positive Trotter-time direction and the world-line of a down-spin when it evolves in negative Trotter-time direction.
In Fig. 3 we show the graphic representation of the breakups $`M^{(\gamma )}`$. The lines now represent the directed loop segments.
Eq. (1) states that the partition function can be obtained as a sum over all breakups. As a sum over all breakups is equivalent to a sum over all loop configurations $`\{l\}`$ we may rewrite Eq. (1) as
$$Z=\underset{\{l\}}{}\rho (\{l\})\text{Tr}\underset{n=1}{\overset{N_T}{}}\underset{i=2m}{}M^{(\gamma _i)}\underset{j=2m+1}{}M^{(\gamma _j)},$$
(2)
where $`\rho (\{l\})=_ip_i^{(\gamma _i)}_jp_j^{(\gamma _j)}`$. Eq. (2) leads to a very efficient MC-algorithm : (a) Choose loop-breakups $`M^{(\gamma _i)}`$ with probabilities $`p_i^{(\gamma _i)}`$. (b) Construct the loop configuration $`\{l\}`$ and flip all loops with probability $`1/2`$. (c) Measure any desired operator in all $`2^{N_L(\{l\})}`$ spin configurations reachable with independent loop flips (improved estimators), where $`N_L(\{l\})`$ is the number of loops in the loop configuration $`\{l\}`$.
For later use we rewrite Eq. (2) in a form of traces over individual loops. Noting that vertical and diagonal loop segments do not change the spin-direction (see Fig. 3), we may associate the $`2\times 2`$ identity matrix $`\sigma ^0=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)`$ with vertical and diagonal loop segments. As horizontal loop segments do change the spin-direction, we associate the Pauli-matrix $`\sigma ^x=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)`$ with them. We then may rewrite Eq. (2) as
$$Z=\underset{\{l\}}{}\rho (\{l\})\underset{l\{l\}}{}\text{Tr}_l\underset{\mu }{}\sigma ^{\gamma _\mu },$$
(3)
where $`\mu `$ is an index running over loop $`l`$ and $`\gamma _\mu =0,x`$. $`\text{Tr}_l`$ denotes the trace over loop $`l`$. Since $`\text{Tr}_l_\mu \sigma ^{\gamma _\mu }=2`$, Eq. (3) is equivalent to a statistical mechanical model of oriented loops, $`Z=_{\{l\}}\rho (\{l\})2^{N_L(\{l\})}`$.
## III Correlation functions, improved estimators
The expectation value of an operator $`𝒪`$ is
$$𝒪=\text{Tr}(𝒪\mathrm{exp}(\beta H))=\underset{\alpha ,\beta }{}\varphi _\alpha |𝒪|\varphi _\beta \varphi _\beta |\mathrm{exp}(\beta H)|\varphi _\alpha $$
(4)
If $`𝒪`$ is diagonal in the basis $`\{|\varphi _\alpha \}`$ then this procedure is straightforward. The updating procedure generates a sequence of configurations $`c_{i_{MC}}`$ ($`i_{MC}=1\mathrm{}N_{MC}`$), according with the distribution function of the system. In these configurations $`𝒪`$ takes a well defined value $`𝒪(c_{i_{MC}})`$, therefore:
$$𝒪=\frac{1}{N_{MC}}\underset{i_{MC}}{}𝒪(c_{i_{MC}}).$$
(5)
The loop algorithm allows to measure an operator not only in $`c_{i_{MC}}`$ but in all configurations related by loop flippings. We illustrate the use of these improved estimators by computing $`𝒪=S_𝐱^zS_𝐲^z`$ (here indices $`𝐱`$ and $`𝐲`$ label both space and Trotter time (see Fig. 4). When $`𝐱`$ and $`𝐲`$ belong to different loops the orientation can be changed independently and the total contribution cancels. By the contrary when $`𝐱`$ and $`𝐲`$ are on the same loop the orientations of the loop in both sites are linked and these terms contribute for the two possible orientations of the loop.
We will consider now the problem of non diagonal operators. The expectation value of a non diagonal operator $`𝒪^{}`$ in the loop picture is, see Eq. (2):
$$𝒪^{}=\underset{\{l\}}{}\rho (\{l\})\text{Tr}𝒯\left(𝒪^{}\underset{n=1}{\overset{N_T}{}}\underset{i}{}M^{(\gamma _i)}\underset{j}{}M^{(\gamma _j)}\right),$$
(6)
where $`𝒯()`$ means proper imaginary time ordering. Let us take as an example the two-point correlator $`𝒪^{}=S_𝐱^+S_𝐲^{}`$,
Graphically the evaluation of an operator can be interpreted on the checkerboard framework as the insertion of a new kind of plaquette. In Fig. 5 we show the action of that operator in the checkerboard picture. We note that an off-diagonal operator in general reverses the direction of one or more loops. The loop configurations generated by the MC updating-procedure does, on the other hand, only generate loops with well defined loop orientations. Nevertheless there is a close connection between these two types of configurations which is easy to understand in graphical terms.
In Fig. 5 it is shown how the flipping of one spin ’propagates’ through the loop, changing the orientation of the loop from that point. Thinking in terms of oriented loops it is obvious that with only one of these flipping processes ($`S_x^+`$ or $`S_y^{}`$) per loop, it is not possible to close the loop consistently. To reestablish the original loop orientation it is necessary to have an even number of properly ordered $`S^{}`$ or $`S^+`$ operators on the same loop to close it consistently in terms of loop orientation variables. A loop which is not properly closed does not contribute to $`𝒪^{}`$. Then we can establish that for a two-point correlation function we only obtain a contribution when $`x`$ and $`y`$ belong to the same loop (see Fig. 6). Eq. (4) could suggest that measurements of non diagonal operators consume more computing time than diagonal operators, but using this graphical picture we note that both computations can be implemented in an equivalent way.
These ideas can be justified in formal terms using Eq. (3) and Eq. (6). The $`S_𝐱^+`$ and $`S_𝐲^{}`$ operators are placed in between of two $`\sigma ^\gamma `$ matrices belonging to neighboring plaquettes and traces can be taken again independently in each loop. We define the $`2\times 2`$ matrices $`\sigma ^+=\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right)`$ and $`\sigma ^{}=\left(\begin{array}{cc}0& 0\\ 1& 0\end{array}\right)`$. For positive loop-direction (with respect to the Trotter direction) $`S^+`$ is equivalent to $`\sigma ^+`$, for a directed loop segment with negative loop-direction $`S^+`$ is equivalent to $`\sigma ^{}`$. For $`S^{}`$ it is just the other way round. The loop direction of relevance here is the one before the insertion of either a $`S^+`$ or a $`S^{}`$ operator.
We start considering contributions to $`S_𝐱^+S_𝐲^{}`$ where the loop-direction at site $`𝐱`$ is up and down at site $`𝐲`$ (see Fig. 6). The expectation value of the non-diagonal operator $`S_𝐱^+S_𝐲^{}`$ then becomes (compare Eq. (3))
$$S_𝐱^+S_𝐲^{}\frac{1}{Z}\underset{\{l\}}{}\rho (\{l\})𝒯\left(\sigma _𝐱^+\sigma _𝐲^+\underset{l\{l\}}{}\text{Tr}_l\underset{\mu }{}\sigma ^{\gamma _\mu }\right).$$
(7)
Here $`𝒯`$ means proper time and space ordering. When $`\sigma _𝐱^+`$ and $`\sigma _𝐲^+`$ are placed in different loops, the traces taken in these two loops cancel. If they are in the same loop the trace taken in that loop equals 1 (and not 2), independently of the spin-configuration. We will prove this last point now. We start by writing the partial trace of the loop containing $`\sigma _𝐱^+`$ and $`\sigma _𝐲^{}`$ as
$`T^{(++)}=\text{Tr}_l\sigma _𝐱^+\left(\sigma ^x\right)^{z_1}\sigma _𝐲^+\left(\sigma ^x\right)^{z_2},`$where we neglected the $`\sigma ^0`$ matrices, as they are just the identity matrices. We note that $`z_1+z_2`$ is even since $`\left(\sigma ^x\right)^2=\sigma ^0`$ and because we are considering a loop which did contribute to the partition function $`Z`$ before the $`S_𝐱^+S_𝐲^{}`$ operators were inserted. The $`\sigma ^x`$ matrix corresponds to a horizontal loop segment and such to a change in loop direction. $`z_1`$ needs therefore to be odd (and therefore also $`z_2`$), since one needs an odd number of directional inversions to arrive to a negative loop direction at site $`𝐲`$, starting from a positive direction at site $`𝐱`$. We may therefore rewrite $`T^{(++)}`$ (using again $`\left(\sigma ^x\right)^2=\sigma ^0`$) as
$`T^{(++)}=\text{Tr}_l\sigma _𝐱^+\sigma ^x\sigma _𝐲^+\sigma ^x1,`$as one can easily evaluate. Similarly one can consider the case when the initial loop directions are both positive at sites $`𝐱`$ and $`𝐲`$. The expectation value of the non-diagonal operator $`S_𝐱^+S_𝐲^{}`$ becomes then in this case
$$S_𝐱^+S_𝐲^{}\frac{1}{Z}\underset{\{l\}}{}\rho (\{l\})𝒯\left(\sigma _𝐱^+\sigma _𝐲^{}\underset{l\{l\}}{}\text{Tr}_l\underset{\mu }{}\sigma ^{\gamma _\mu }\right).$$
(8)
The corresponding one-loop contributions then have the form
$`T^{(+)}=\text{Tr}_l\sigma _𝐱^+\left(\sigma ^x\right)^{z_1}\sigma _𝐲^{}\left(\sigma ^x\right)^{z_2}=\text{Tr}_l\sigma _𝐱^+\sigma _𝐲^{}1,`$since both $`z_1`$ and $`z_2`$ have to be even in this case. Similarly one can consider the two remaining cases of loop directions down/up and down/down at the sites $`𝐱`$ and $`𝐲`$. It is worthwhile noting, that one easily proves along these lines the expected result $`S_𝐱^+S_𝐲^+=0`$.
## IV General case n-point correlation functions
In the last section we have shown how the loop orientation is the fundamental variable to deal with the computation of correlation functions using improved estimators. In fact the problem of n-point correlation functions can also be reduced to the study of how the loop orientation is changed by the action of some operators.
We illustrate the case of two-loop terms for the four-point correlation function $`𝒪^{\prime \prime }=S_𝐱^+S_𝐲^{}S_𝐱^{}^+S_𝐲^{}^{}`$. Here we consider the case relevant for the specific heat were $`(𝐱,𝐲)`$ and $`(𝐱^{},𝐲^{})`$ are pairs of real-space nearest neighbor (n.n.) sites at the same Trotter time. This operator can generate several different kinds of contributions. The first one is the case of two disconnected one-loop contributions (see Fig. 7). This is the case if $`S_𝐱^+`$ and $`S_𝐲^{}`$ act in one loop and $`S_𝐱^{}^+`$ and $`S_𝐲^{}^{}`$ in a second loop. A second contribution arises if $`S_𝐱^+`$ and $`S_𝐲^{}^{}`$ act in one loop and $`S_𝐲^{}`$ and $`S_𝐱^{}^+`$ in a second loop (see Fig. 8). We call this contribution a connected two-loop contribution. A third contribution arises when all four sites act on the same loop.
The evaluation of a single off-diagonal four-point operator $`𝒪^{\prime \prime }`$ does not pose a problem within the loop algorithm. For the case of interest, the specific heat a few additional points need to be kept in mind. The specific heat $`c_V`$ is given by
$$c_V=\frac{\beta ^2}{LN_T^2}\left[\underset{𝐱,𝐱^{}}{}(𝐒_𝐱𝐒_𝐲)(𝐒_𝐱^{}𝐒_𝐲^{})\left(\underset{𝐱}{}𝐒_𝐱𝐒_𝐲\right)^2\right],$$
(9)
where, again, $`(𝐱,𝐲)`$ and $`(𝐱^{},𝐲^{})`$ are pairs of (real-space) n.n. sites on the Trotter lattice. The first term of Eq. (9) is a local energy-energy correlation function. When, $`𝐱`$ and $`𝐲`$ belong to a loop and $`𝐱^{}`$ and $`𝐲^{}`$ to another, we generate two-loop disconnected terms (as the one illustrated in Fig. 7) that can be computed from the expectation value of the internal energy, the second term of specific heat. The energy in a given MC-configuration, $`E_{i_{MC}}`$, can be written as a sum of the energy in the $`N_L(i_{MC})`$ loops in this MC-configuration:
$`E_{i_{MC}}={\displaystyle \underset{l=1}{\overset{N_L(i_{MC})}{}}}E_{i_{MC}}^l.`$
With this definition we obtain
$`c_V^{(ind)}={\displaystyle \frac{1}{N_{MC}}}{\displaystyle \underset{i_{MC}}{}}{\displaystyle \underset{lk}{}}E_{i_{MC}}^lE_{i_{MC}}^k={\displaystyle \frac{1}{N_{MC}}}{\displaystyle \underset{i_{MC}}{}}\left[\left({\displaystyle \underset{l}{}}E_{i_{MC}}^l\right)^2{\displaystyle \underset{l}{}}\left(E_{i_{MC}}^l\right)^2\right],`$where $`c_V=c_V^{(conn)}+c_V^{(ind)}`$. For the evaluation of the connected term $`c_V^{(conn)}`$ one has to evaluate the off-site terms, $`c_V^{(off)}`$, where the pairs $`(𝐱,𝐲)`$ and $`(𝐱^{},𝐲^{})`$ are disjunct, separately from the on-site terms, $`c_V^{(on)}`$, where they are not disjunct: $`c_V^{(conn)}=c_V^{(off)}+c_V^{(on)}`$. By spin-algebra the on-site terms reduce to general two-point correlation functions. The (connected) off-site contributions fall in three categories, depending on the number $`S^z`$ operators involved (four, two or zero). The contributions with four $`S^z`$ operators have one and two loop contributions. A connected term with two $`S^z`$ operators has no two-loop contribution. Every correlation with two $`S^z`$ operators has the form $`S_𝐱^zS_𝐲^zS_𝐱^{}^+S_𝐲^{}^{}`$. If the indices $`𝐱`$ and $`𝐲`$ are not in the same loop the two $`S^z`$ operators act in different loops and their traces cancel for the reason explained in section III. The same reasoning is valid for $`𝐱^{}`$ and $`𝐲^{}`$ with the operators $`S^+`$ and $`S^{}`$. Finally, terms with no $`S^z`$ operators can have two loop contributions (see Fig. 8) and also one-loop contributions when the $`S^+`$ and $`S^{}`$ are properly ordered along the loop to close the loop coherently in terms of loop orientation. On the left of Fig. 9 we see that an arbitrary insertion of the operators $`S^+`$ and $`S^{}`$ can produce a conflict on the orientation of the loop. Technically, the value of the trace taken along the loop will depend on the structure of the correlator. This structure determines the order of the insertion of the $`\sigma ^+`$ and $`\sigma ^{}`$ matrices. For example the trace along the loop on the left of Fig. 9 is:
$`\text{Tr}(\sigma _𝐱^{}\sigma ^x\sigma _𝐱^{}^{}\sigma _𝐲^{}\sigma ^x\sigma _𝐲^{}^+)=0.`$For the loop on the right of Fig. 9 it is:
$`\text{Tr}(\sigma _𝐱^{}\sigma ^x\sigma _𝐲^{}\sigma _𝐱^{}^+\sigma ^x\sigma _𝐲^{}^+)=1.`$
It is possible to evaluate certain off-diagonal operators $`𝒪`$ by an alternative method. The condition is, that the operator can be expressed by a sum of local operators which do involve the same pairs of sites $`l,l^{}`$ as the Hamilton-operator $`H=_{l,l^{}}H_{l,l^{}}`$: $`𝒪=_{l,l^{}}𝒪_{l,l^{}}`$. It is then possible to compute $`𝒪`$ by a reweighting method. The idea is to extend the plaquette of the checkerboard representation by new internal degrees of freedom, $`_\beta |\varphi _\beta \varphi _\beta |`$ (see Fig. 10). The reweighted matrix element of $`𝒪_{x,x^{}}`$ is then
$$𝒪_{\alpha _n,\alpha _{n+1}}^{(n)}(x,x^{})=\underset{\beta }{}\frac{\varphi _{\alpha _n}^{(n)}|𝒪_{x,x^{}}|\varphi _\beta \varphi _\beta |\mathrm{exp}(\mathrm{\Delta }\tau H_{x,x^{}})|\varphi _{\alpha _{n+1}}^{(n+1)}}{\varphi _{\alpha _n}^{(n)}|\mathrm{exp}(\mathrm{\Delta }\tau H_{x,x^{}})|\varphi _{\alpha _{n+1}}^{(n+1)}},$$
(10)
where $`x`$ and $`x^{}`$ denote combined space-time indices. For a given spin-configuration $`c_{i_{MC}}=\{\varphi _{\alpha _n}^{(n)}|(n=1,\mathrm{},N_T)\}`$ the off-diagonal expectation value of $`𝒪(c_{i_{MC}})`$ is $`𝒪(c_{i_{MC}})=1/N_T_{x,x^{},(n)}𝒪_{\alpha _n,\alpha _n+1}^{(n)}(x,x^{})`$ and $`𝒪=1/N_{MC}_{i_{MC}}𝒪(c_{i_{MC}})`$ (see Eq. (5)).
The reweighting method may also be applied to specific heat, which is the sum of products of local operators.
From the point of view of the complexity of the algorithm, measuring four-point correlation functions requires more computing time than two-point correlation functions. For the latter is only necessary to know whether or not two sites are in the same loop. This information can be obtained at the same time the loop is constructed and consequently the computing time remains proportional to $`LN_T`$. For n-point correlation functions the situation is more complex. In this case, there are contributions involving two or more loops and at the same time non-diagonal operators give different contributions depending on how they are ordered on the loop. In practice this depends on the shape of the loops in each configuration. A rigorous study of the performance of the method must include an analysis of the behavior of the statistical errors as a function of the temperature, size, number and type of operator involved in the correlation functions and the details of the Hamiltonian. This detailed analysis of technical aspects of n-point correlations will be presented elsewhere.
## V Results
As an application of the rules explained in this paper we have computed the specific heat for a Heisenberg chain and for a ladder with $`J_{}=0.5J`$ (which corresponds to the ratio for the ladder-compound Sr<sub>14</sub>Cu<sub>24</sub>O<sub>41</sub> ).
In Fig. 11 we compare exact diagonalization results with the results using the method described above and the reweighting method for the same number of MC steps. The error bars in these two methods are also compared. For the lowest temperature the error bar with improved estimators are 6 times smaller. Taking into account that error bars decay as $`\frac{1}{\sqrt{N_{MC}}}`$ we expect that without using improved estimators 36 times more MC steps are necessary to get equal size error bars. The statistical errors are amplified by the factor $`\beta ^2`$. This factor and the substraction of similarly large numbers lead to large error bars at low temperatures.
In the Fig. 12 we present results for the specific heat of a 100-site Heisenberg chain. To reproduce the linear regime at low temperatures it is necessary to perform a careful extrapolation to $`\mathrm{\Delta }\tau 0`$ taking half a million of MC steps for each $`\mathrm{\Delta }\tau `$ values and 10 different $`N_T`$ values ranging from 20 to 200.
In Fig. 13 we present results for the two-leg ladder of $`2\times 201`$ sites with twisted boundary conditions (i.e. for $`J_{}=0`$ this system corresponds to a L=402-site Heisenberg chain).
## VI Conclusions
We have presented detailed rules on how to evaluate general, off-diagonal n-point Greens functions within the loop algorithm. These rules have a very simple interpretation in the picture of oriented loops. They state that the loop-orientation has to close coherently whenever a certain number of non-diagonal operators are inserted. We have shown how to apply these rules to the case of the specific heat and presented results for the 1D-Heisenberg model and a ladder system.
## VII Acknowledgments
We would like to acknowledge discussions with Matthias Troyer, Naoki Kawashima and Andreas Klümper and the support of the German Science Foundation. We acknowledge the hospitality of the ITP in Santa Barbara. This research was supported by the National Science Foundation under Grant No. PHY94-07194.
|
warning/0002/math0002162.html
|
ar5iv
|
text
|
# Untitled Document
Maps of Surface Groups to Finite Groups
with No Simple Loops in the Kernel
by
Charles Livingston
Abstract: Let $`F_g`$ denote the closed orientable surface of genus $`g`$. What is the least order finite group, $`G_g`$, for which there is a homomorphism $`\psi :\pi _1(F_g)G_g`$ so that no nontrivial simple closed curve on $`F_g`$ represents an element in Ker($`\psi `$)? For the torus it is easily seen that $`G_1=Z_2\times Z_2`$ suffices. We prove here that $`G_2`$ is a group of order 32 and that an upper bound for the order of $`G_g`$ is given by $`g^{2g+1}`$. The previously known upper bound was greater than $`2^{g2^{2g}}`$.
For any compact surface $`F`$ there exists a finite group $`G`$ and a homomorphism $`\psi :\pi _1(F)G`$ such that no nontrivial element in the kernel of $`\psi `$ can be represented by a simple closed curve. Such a homomorphism is said to have nongeometric kernel. Casson, Gabai, and Skora have each constructed examples of this (see Section 2 for details). The presence of such examples raises a variety of questions relating to the characterization of the finite groups that can occur in this way. This paper addresses the problem of determining the relationship between the genus of $`F`$ and the order of $`G`$. In the case that $`F`$ is a torus a complete analysis is straightforward. For instance, the natural projection $`\psi :\pi _1(F)H_1(F;Z_2)Z_2\times Z_2`$ has nongeometric kernel.
Our first result concerns the genus 2 closed orientable surface, $`F_2`$. Casson’s construction yields a group of order $`2^{38}`$. Skora reduced this order considerably by producing a group of order $`2^9`$. In Section 3 a group of order $`2^5=32`$, $`G_2`$, is constructed for which there is a homomorphism $`\psi _2:\pi _1(F_2)G_2`$ having nongeometric kernel. In Section 4 it is proved that no such example can be constructed using a group of order less than 32.
The example in Section 3 is generalized to construct examples for arbitrary genus surfaces in Section 5. The order of the groups constructed is quite small compared to previously constructed examples. As the examples directly generalize the minimal genus 2 example, there is the possibility that they are minimal as well.
Acknowledgements Thanks are due to Allan Edmonds for pointing out the proof of Theorem 4.2. The work in Section 5 was motivated by discussions with Dennis Johnson.
1 Notation and Conventions
Throughout this paper all surfaces will be closed and orientable. References to basepoints for the fundamental group of a space are omitted. Since the property of being in the kernel of a homomorphism depends only on the conjugacy class of an element, such omissions will not affect the arguments.
By a simple loop on a surface we mean an embedding of the circle $`S^1`$ .
We will say that a homomorphism $`\psi :\pi _1(F)G`$ has geometric kernel if some nontrivial element in the kernel can be represented by a simple loop. Otherwise $`\psi `$ has nongeometric kernel.
2 Basic Examples
In this section a procedure of Casson is used to construct for each surface $`F`$ a finite group $`G`$ and a surjective homomorphism $`\psi :\pi _1(F)G`$ such that $`\psi `$ has nongeometric kernel. The orders of the groups involved is computed for contrast with the examples produced in Section 5.
The statement that $`\psi `$ has nongeometric kernel can be reinterpreted in terms of covering spaces as follows. Corresponding to Ker$`(\psi )`$ there is a connected regular covering space $`p:\stackrel{~}{F}F`$ with $`p_{}(\pi _1(F))=`$ Ker$`(\psi )`$. An element in $`\pi _1(F)`$ is in Ker$`(\psi )`$ if and only if when represented by a closed path, the path lifts to a closed path in $`\stackrel{~}{F}`$. Hence a simple loop on $`F`$ represents an element in Ker$`(\psi )`$ if and only if it can be lifted to a simple loop in $`\stackrel{~}{F}`$. Conversely, if $`p:\stackrel{~}{F}F`$ is a regular covering space with the property that no nontrivial simple loop on $`F`$ lifts to $`\stackrel{~}{F}`$ then the natural projection $`\psi :\pi _1(F)\pi _1(F)/p_{}(\pi _1(\stackrel{~}{F}))`$ has nongeometric kernel.
Construction Given a surface $`F`$, construct the covering space $`p:\stackrel{~}{F}F`$ corresponding to the kernel of the projection $`\pi _1(F)H_1(F;Z_2)`$. Simple nonseparating loops on $`F`$ represent generators of $`H_1(F;Z_2)`$ and hence do not lift to $`\stackrel{~}{F}`$ . Nontrivial separating simple loops do lift, but each preimage on $`\stackrel{~}{F}`$ is nonseparating on $`\stackrel{~}{F}`$.
Now construct the covering $`q:\overline{F}\stackrel{~}{F}`$ corresponding to the kernel of the projection $`\pi _1(\stackrel{~}{F})H_1(\stackrel{~}{F};Z_2)`$. As no nonseparating simple loop on $`\stackrel{~}{F}`$ lifts to $`\overline{F}`$ it is apparent that no nontrivial simple loop on $`F`$ lifts to $`\overline{F}`$.
It remains to show that the covering $`pq:\overline{F}F`$ is regular; that is, that $`p_{}q_{}(\pi _1(\overline{F}))`$ is normal in $`\pi _1(F)`$. Observe that $`q_{}(\pi _1(\overline{F}`$)) is a characteristic subgroup of $`\pi _1(\stackrel{~}{F})`$ and $`p_{}(\pi _1(\stackrel{~}{F}))`$ is a characteristic subgroup of $`\pi _1(F)`$. Since a characteristic subgroup of a characteristic subgroup is characteristic, $`p_{}q_{}(\pi _1(\overline{F})`$) is characteristic in $`\pi _1(F)`$, and is hence normal.
Order of $`\pi _1(F)/<p_{}q_{}(\pi _1(\overline{F}))>`$
The order of this finite group is equal to the degree of the covering $`pq`$. Suppose that $`F`$ is of genus $`g`$. The Euler characterisitic of $`F`$ is $`22g`$ . Since $`\stackrel{~}{F}`$ is a $`2^{2g}`$ fold cover of $`F`$, the Euler characteristic of $`\stackrel{~}{F}`$ is $`2^{2g}(22g)`$. The genus of $`\stackrel{~}{F}`$ is $`\frac{1}{2}(22^{2g}(22g))=\frac{1}{2}((g1)2^{2g+1}+2)=g^{}`$. The covering $`q:\overline{F}\stackrel{~}{F}`$ is of degree $`2^{2g^{}}`$. The degree of $`pq:\overline{F}F`$ is the product of these two degrees: $`2^{2g^{}}2g=2^{(g1)2^{2g+1}+2+2g}`$.
Note The construction of Gabai differs considerably from the one above. He notes that every simple curve is in the complement of some index three subgroup of $`\pi _1(F)`$; nonseparating curves are not in the kernel of some map to $`Z_3`$ and separating curves are mapped to a 3–cycle in the third symmetric group, $`S_3`$, under some homomorphism and hence map to the complement of an index three subgroup of $`S_3`$. Hence the quotient map $`\pi _1(F)\pi _1(F)/H`$ has nongeometric kernel, where $`H`$ is the intersection of all index three subgroups of $`\pi _1(F)`$. We have been unable to find reasonable bounds on the size of this quotient.
3 A small genus 2 example
The groups constructed in the previous section are of very large order. If $`F`$ is of genus 2, the corresponding group is of order $`2^{38}`$. This section presents a description of a group of order 32, $`G_2`$, and a homomorphism $`\psi _2`$ of the fundamental group of the genus 2 surface to $`G_2`$, such that $`\psi _2`$ has a nongeometric kernel. The next section contains a proof that this example is minimal.
For the remainder of this section $`F`$ will denote a genus 2 surface.
Construction of $`G_2`$ For our purposes, the easiest way to describe $`G_2`$ is as follows. Define a group structure on the set $`(Z_2)^4\times Z_2`$ by defining the product by
$$(a_1,b_1,a_2,b_2,ϵ)(a_1^{},b_1^{}a_2^{},b_2^{},ϵ^{})=(a_1+a_1^{},b_1+b_1^{},b_2+a_2^{},b_2+b_2^{},ϵ+ϵ^{}+b_1a_1^{}+b_2a_2^{}).$$
The operations within the parenthesis are addition and multiplication in $`Z_2`$. The verification that this defines a group structure can be done by a direct calculation, which is left to the reader. The group is denoted $`32_{42}`$ in , and $`\mathrm{\Gamma }_5a_1`$ in the notation of .
An essential calculation for later purposes is that of commutators in $`G_2`$. A direct computation yields
$$[(a_1,b_1,a_2,b_2,ϵ),(a_1^{},b_1^{}a_2^{},b_2^{},ϵ^{})]=(0,0,0,0,(b_1a_1^{}b_1^{}a_1)+(b_2a_2^{}b_2^{}a_2)).$$
$`(1)`$
From this it is apparent that both the center and commutator subgroup of $`G_2`$ consists of the set $`(0,0,0,0)\times Z_2`$. The abelianization of $`G_2`$ is $`(Z_2)^4`$, given by the projection $`(Z_2)^4\times Z_2(Z_2)^4\times \{0\}`$.
Construction of $`\psi _2`$ Let $`\{x_1,y_1,x_2,y_2\}`$ be a standard generating set of $`\pi _1(F)`$ so that $`\pi _1(F)`$ has presentation $`<x_1,y_1,x_2,y_2,[x_1,y_1][x_2,y_2]=1>`$. This set projects to a standard symplectic basis of $`H_1(F;Z_2)`$, $`\{|x_1|,|y_1|,|x_2|,|y_2|\}`$
Define $`\psi _2:\pi _1(F)G_2`$ be setting:
$$\psi _2(x_1)=(1,0,0,0)\times (0),$$
$$\psi _2(y_1)=(0,1,0,0)\times (0),$$
$$\psi _2(x_2)=(0,0,1,0)\times (0),$$
$$\psi _2(x_2)=(0,0,0,1)\times (0).$$
Using (1) it is easily verified that this gives a well defined surjective representation.
The key observation is that $`\psi _2`$ has the following property: if $`\omega _1`$ and $`\omega _2`$ are elements of $`\pi _1(F)`$, then
$$[\psi _2(\omega _1),\psi _2(\omega _2)]=(0,0,0,0)\times (|\omega _1||\omega _2|),$$
$`(2)`$
where $`|\omega _1||\omega _2|`$ is the $`Z_2`$ intersection number of the classes in $`H_1(F;Z_2)`$, represented by $`\omega _1`$ and $`\omega _2`$. This follows from (1) along with the fact that if $`(a_1,b_1,a_2,b_2)`$ and $`(a_1^{},b_1^{},a_2^{},b_2^{})`$ are classes in $`H_1(F;Z_2)`$, then $`(a_1,b_1,a_2,b_2)(a_1^{},b_1^{},a_2^{},b_2^{})=(b_1a_1^{}b_1^{}a_1)+(b_2a_2^{}b_2^{}a_2)`$. (Note also that the natural map $`\pi _1(F)H_1(F;Z_2)`$ factors through $`G_2`$ via $`\psi _2`$.)
The kernel of $`\psi _2`$ is nongeometric Suppose that there is a simple loop $`\gamma `$ representing a nontrivial element $`\omega `$ in Ker$`(\psi _2)`$. Our first observation is that $`\gamma `$ can be chosen to be separating. If $`\gamma `$ is nonseparating, pick a simple loop $`\gamma ^{}`$ meeting $`\gamma `$ transversely in exactly one point. Let $`\omega ^{}`$ be the element of $`\pi _1(F)`$ represented by $`\gamma ^{}`$. Clearly, $`[\omega ,\omega ^{}]`$ is in the kernel of $`\psi _2`$ and it is represented by a separating simple loop.
Since $`\gamma `$ is now assumed to be separating, it bounds a punctured torus on $`F`$. This follows from the classification of surfaces. Hence $`\omega =[\omega _1,\omega _2]`$, where $`\omega _1`$ and $`\omega _2`$ are represented by simple loops meeting transversely in one point. From this one computes using (2) that $`\psi _2(w)=[\psi _2(\omega _1),\psi _2(\omega _2)]=(0,0,0,0)\times (|\omega _1||\omega _2|)=(0,0,0,0)\times (1)`$, which is nontrivial in $`G_2`$. This contradicts the assumption that $`w`$ Ker$`(\psi _2).`$
4 Minimality of G
The goal of this section is to prove that if $`F`$ is of genus 2 and the order of $`G`$ is less than 32, than any homomorphism $`\varphi :\pi _1(F)G`$ has geometric kernel.
Here is a summary of the argument. We first prove that any $`\varphi :\pi _1(F)G`$ has geometric kernel if $`G`$ is a cyclic extension of an abelian group, that is, if $`G`$ contains a normal abelian subgroup with cyclic quotient. The approach used to prove this was pointed out by Allan Edmonds. The argument depends on an analysis of the action of the homeomorphism group of $`F`$ on the set of representations of $`\pi _1(F)`$ to $`G`$. We next note that with the exception of two groups of order 24, $`SL_2(Z_3)`$ and $`S_4`$, all groups of order less than 32 are cyclic extensions of abelian groups. This can be proved by a case–by–case analysis using the Sylow theorems. More easily, group tables such as provide the necessary information. The proof is completed using specialized arguments for $`SL_2(Z_3)`$ and $`S_4`$.
Cyclic Extension of Abelian Groups Fix a group $`G`$. The group of basepoint preserving homeomorphisms of $`F`$ acts on the set of representations of $`\pi _1(F)`$ to $`G`$, as follows. If $`h`$ is a homeomorphism of $`F`$, send a representation $`\varphi `$ to $`\varphi h_{}`$. Notice that $`\varphi `$ has geometric kernel if and only if $`\varphi h_{}`$ has geometric kernel. The following is a result of Nielsen ; a proof can be found in .
4.1 Lemma. If $`G`$ is a cyclic group and $`\varphi :\pi (F)G`$ is a surjective homomorphism, then there is a homeomorphism $`h`$ of $`F`$ such that $`\varphi h_{}(x_1)`$ generates $`G`$ , and $`\varphi h_{}(y_1)`$, $`\varphi h_{}(x_2)`$ and $`\varphi h_{}(y_2)`$ are all trivial.
4.2 Theorem. If $`G`$ contains an abelian normal subgroup $`N`$ such that $`G/N`$ is cyclic, than any surjective homomorphism $`\varphi :\pi _1(F)G`$ has a geometric kernel.
Proof Denote the quotient map $`GG/N`$ by $`\rho `$. Applying the lemma, we can assume that $`\rho \varphi (x_2)`$ and $`\rho \varphi (y_2)`$ are both trivial. Hence $`\varphi (x_2)`$ and $`\rho \varphi (y_2)`$ are both in $`N`$. The commutator $`[x_2,y_2]`$ is represented by a simple loop and is in the kernel of $`\varphi `$, since $`\varphi ([x_2,y_2])`$ is in the commutator subgroup of an abelian group.
Exceptional Groups
Case 1 We begin by recalling that $`SL_2(Z_3)`$ is isomorphic to the semidirect product of the quaternionic 8–group, $`Q`$, with $`Z_3`$. We will use the standard notation for elements in $`Q`$ . The generator of $`Z_3`$ will be denoted $`t`$. The action of $`Z_3`$ on $`Q`$ is given by $`tit^1=j`$, $`tjt^1=k`$ and $`tkt^1=i`$. Note that $`t(1)t^1=1`$ and that $`1Q`$ is hence central in $`SL_2(Z_3)`$.
Suppose $`\varphi :\pi _1(F)SL_2(Z_3)`$ is a surjective representation with nongeometric kernel. Applying the lemma to the composition $`\pi _1(F)SL_2(Z_3)SL_2(Z_3)/Q=Z_3`$ we can assume that $`\varphi (x_1)=tq_1`$, $`\varphi (y_1)=q_2`$, $`\varphi (x_2)=q_3`$, and $`\varphi (y_2)=q_4`$, where each $`q_i`$ is in $`Q`$.
Since $`[x_2,y_2]`$ is represented by a simple loop, $`[q_3,q_4]1`$. Hence $`[q_3,q_4]=1Q`$. It follows that $`[tq_1,q_2]=1`$. Note that $`q_2\pm 1`$, so $`q_2=\pm i`$, $`\pm j`$, or $`\pm k`$. From the commutator relation, $`tq_1q_2q_1^1t^1=q_2^1`$ For any two quaternions, $`q_1q_2q_1^1=q_2^{\pm 1}`$. Hence, $`tq_2t^1=q_2^{\pm 1}`$. However, this is impossible, given that $`q_2\pm 1`$ and the action of $`t`$ on $`Q`$.
Case 2 The symmetric group $`S_4`$ is the semidirect product of $`Z_2\times Z_2`$ with $`S_3`$. As a subgroup, the $`Z_2\times Z_2`$ is given by the set $`\{(1),(12)(34),(13)(24),(14)(23)\}`$ The $`S_3`$ is given by the set $`\{(1),(12),(13),(23),(123),(321)\}`$.
Let $`\varphi :\pi _1(F)S_4`$ be a surjective representation with nongeometric kernel. The main result of applied to the composition $`\pi _1(F)S_4S_4/(Z_2\times Z_2)S_3`$ shows that by applying a homeomorphism we can arrange that $`\varphi `$ takes on the values $`\varphi (x_1)=(12)n_1`$, $`\varphi (y_1)=n_2`$, $`\varphi (x_2)=(123)n_3`$, and $`\varphi (y_2)=n_4`$ where each $`n_i`$ is in $`Z_2\times Z_2`$.
Since both $`y_1`$ and $`y_2`$ are represented by simple loops, neither $`n_2`$ nor $`n_4`$ are trivial. Also, $`[x_1,y_1]`$ is represented by a simple loop, so $`[(12)n_1,n_2]1`$. It follows that $`n_2(12)(34)`$. There are two other possibilities for $`n_2`$.
Suppose that $`n_2=(13)(24)`$. There are three possible values of $`n_4`$ to be considered. Because $`y_1y_2`$ is realized by a simple loop, $`n_4(13)(24)`$. It is easily seen that $`y_1x_2y_2^1x_2^1`$ is realized by a simple loop. Hence $`n_4(12)(34)`$. Finally $`n_4(14)(23)`$, because $`y_2x_1y_1^1x_1^1`$ can also be represented by a simple loop.
We proceed similarly if $`n_2=(14)(23)`$. Clearly $`n_4(14)(23)`$. Because $`y_2x_1y_1^1x_1^1`$ is realized by a simple loop, $`n_4(13)(24)`$. Finally, it is again easily seen that $`y_1x_2^1y_2^1x_2`$ is realized by a simple loop. This implies that $`n_4(12)(34)`$. All possibilities have now been eliminated.
5 Generalizations The group constructed in Section 3, $`G_2`$, can be generalized to a group $`G_k`$ such that for the genus $`k`$ surface $`F_k`$ there is a homomorphism $`\varphi _k:\pi _1(F_k)G_k`$ with nongeometric kernel. The arguments are similar to those of Section 3 and are only outlined here.
Define $`G_k`$ by defining a product on the set $`(Z_k)^{2k}\times Z_k`$ as follows.
$$(a_1,b_1,a_2,b_2,\mathrm{}b_k,ϵ)(a_1^{},b_1^{},a_2^{},b_2^{},\mathrm{}b_k^{},ϵ^{})=(a_1+a_1^{},b_1+b_1^{},a_2+a_2^{},b_2+b_2^{},\mathrm{},b_k+b_k^{},ϵ+ϵ^{}+b_ia_i^{})$$
Sums and products within the parenthesis are in $`Z_k`$. That this defines a group is a straightforward calculation.
There is a natural representation $`\varphi _k:\pi _1(F_k)G_k`$ as before. In this case the essential observation is
$$[\varphi _k(\omega _1),\varphi _k(\omega _2)]=(0,0,\mathrm{},0)\times (|\omega _1||\omega _2|),$$
$`(1)`$
where $`(|\omega _1||\omega _2|)`$ is the $`Z_k`$ intersection number of the classes in $`H_1(F_k;Z_k)`$ represented by $`\omega _1`$ and $`\omega _2`$.
If $`\varphi _k`$ had geometric kernel, there would be a separating simple loop representing an element in the kernel. Using the classification of surfaces, that element would be of the form
$$[\omega _1,\omega _1^{}][\omega _2,\omega _2^{}]\mathrm{}[\omega _m,\omega _m^{}]$$
with $`m<k`$ and $`(|\omega _i||\omega _i^{}|)=1`$ for all $`i`$. A contradiction follows as in Section 3.
Remark The order of the group $`G`$ just constructed is $`g^{2g+1}`$. This number should be contrasted to the order found in Section 2, $`2^{(g1)2^{2g+1}+2+2g}`$. The first is obviously much smaller than the second. The results of this paper, along with our difficulties in trying to find smaller examples, leads us to conjecture that $`g^{2g+1}`$ represents the least possible order.
References
Edmonds, A. Surface Symmetry I, Michigan J. Math 29 (1982) 171–183.
Edmonds, A. Surface Symmetry II, Michigan J. Math 30 (1983) 143–154.
Hall, M. and Senior, J.K. The groups of order $`2^n(n6)`$ Macmillan, New York (1964).
Nielson, J. Die Struktur periodischer Transformationen von Flachen, Dansk Vid. Selsk., Mat.-Fys. Medd. 15 (1937), 1–77.
Skora, R. Dissertation, Department of Mathematics, University of Texas, Austin, Texas, 1984.
Thomas, A.D. and Wood, G.V. Group Tables, Shiva Publishing Limited, Kent, Great Britian, 1980.
Department of Mathematics
Indiana University
Bloomington, IN 47405
livingst@indiana.edu
|
warning/0002/cond-mat0002375.html
|
ar5iv
|
text
|
# Skyrmions in Spinor Bose-Einstein Condensates
## Abstract
We show that spinor Bose-Einstein condensates not only have line-like vortex excitations, but in general also allow for point-like topological excitations, i.e., skyrmions. We discuss the static and dynamic properties of these skyrmions for spin-1/2 and ferromagnetic spin-1 Bose gases.
Introduction. — An understanding of quantum magnetism is important for a large number of phenomena in physics. Three well-know examples are high-temperature superconductivity, quantum phase transitions and the quantum Hall effect. Moreover, it appears that magnetic properties will also be very important in another area, namely Bose-Einstein condensation in trapped atomic gases. The latter has come about because of two independent experimental developments. The first development is the realization of an optical trap for atoms, whose operation no longer requires the gas to be doubly spin-polarized . The second development is the creation of a two-component Bose condensate , which by means of rf-fields can be manipulated so as to make the two components essentially equivalent . As a result the behavior of both spin-1 and spin-1/2 Bose gases can now be experimentally explored in detail.
Theoretically, the ground-state structure of these gases has recently been worked out by a number of authors and also the first studies of the line-like vortex excitations have appeared . However, an immediate question that comes to mind is whether the spin degrees of freedom allow for other topological excitations that do not have an analogy in the case of a single component Bose condensate. It is the main purpose of this Letter to show that the answer to this question is in general affirmative. In particular, we show that spinor Bose-Einstein condensates have so-called skyrmion excitations, which are topological nontrivial point-like spin textures . Having done so, we then turn to the investigation of the precise texture and the dynamics of these skyrmions.
Topological considerations. — To find all possible topological excitations of a spinor condensate, we need to know the full symmetry of the macroscopic wavefunction $`\mathrm{\Psi }(𝐱)\sqrt{n(𝐱)}\zeta (𝐱)`$, where $`n(𝐱)`$ is the total density of the gas, $`\zeta (𝐱)`$ is a normalized spinor that determines the average local spin by means of $`𝐒(𝐱)=\zeta _a^{}(𝐱)𝐒^{ab}\zeta _b(𝐱)`$, and $`𝐒`$ are the usual spin matrices obeying the commutation relations $`[S_\alpha ,S_\beta ]=iϵ_{\alpha \beta }^{}{}_{}{}^{\gamma }S_\gamma `$. Note that here, and in the following, summation over repeated indices is always implicitly implied. From the work of Ho we know that in the case of spin-1 bosons we have to consider two possibilities, since the effective interaction between two spins can be either antiferromagnetic or ferromagnetic. In the antiferromagnetic case the ground-state energy is minimized for $`𝐒(𝐱)=\mathrm{𝟎}`$, which implies that the parameter space for the spinor $`\zeta (𝐱)`$ is only $`U(1)\times S^2`$ because we are free to chooce both its overall phase and the spin quantization axis. In the ferromagnetic case the energy is minimized for $`|𝐒(𝐱)|=1`$ and the parameter space corresponds to the full rotation group $`SO(3)`$. Using the same arguments, we find that for spin-1/2 bosons the order parameter space of the ground state is always equivalent to $`SU(2)`$ .
What do these results tell us about the possible topological excitations? For line-like defects or vortices, we can assume $`\zeta (𝐱)`$ to be independent of one direction and the spinor represents a mapping from a two-dimensional plane into the order parameter space. If the vortex is singular this will be visible on the boundary of the two-dimensional plane and we need to investigate the properties of a continuous mapping from a circle $`S^1`$ into the order parameter space $`G`$, i.e., from the first homotopy group $`\pi _1(G)`$. Since $`\pi _1(SU(2))=\pi _1(SO(3))=Z_2`$ and $`\pi _1(U(1)\times S^2)=Z`$, we conclude that a spin-1/2 and a ferromagnetic spin-1 condensate can have only vortices with a winding number equal to 1, whereas an antiferromagnetic spin-1 condensate can have vortices with winding numbers that are an arbitrary integer. Physically, this means that by traversing the boundary of the plane, the spinor can wind around the order parameter at most once or an arbitrary number of times, respectively. This conclusion is identical to the one obtained prevously by Ho .
If the vortex is nonsingular, however, the spinor $`\zeta (𝐱)`$ will be identical everywhere on the boundary of the two-dimensional plane and it effectively represents a mapping from the surface of a three-dimensional sphere $`S^2`$ into the order parameter space. We then need to consider the second homotopy group $`\pi _2(G)`$. For this we have that $`\pi _2(SU(2))=\pi _2(SO(3))=0`$ and $`\pi _2(U(1)\times S^2)=Z`$. Hence nonsingular or coreless vortices are only possible for a spin-1 condensate with antiferromagnetic interactions. It therefore appears that the nonsingular spin texture discussed in Ref. , is topologically unstable and can be continously deformed into the ground state by ‘local surgery’ .
We are now in a position to discuss point-like defects. Since the boundary of a three-dimensional gas is also the surface of a three-dimensional sphere, singular point-like defects are also determined by the second homotopy group $`\pi _2(G)`$. Such topological excitations thus only exists in the case of a spin-1 Bose gas with antiferromagnetic interactions. We call these excitations singular skyrmions, although in view of the work of ’t Hooft and Polyakov it would be justified to call them monopoles . For nonsingular point-like defects the spinor $`\zeta (𝐱)`$ will again be identical on the boundary of the three-dimensional gas. As a result, the configurations space is compactified to the surface of a four-dimensional sphere $`S^3`$ and we need to determine the third homotopy group $`\pi _3(G)`$. For this we find $`\pi _3(SU(2))=\pi _3(SO(3))=\pi _3(U(1)\times S^2)=Z`$. Hence nonsingular skyrmion excitations exists in all three cases.
Skyrmion texture. — In view of the fact that magnetic gradients are needed to stabilize the spinor condensate in the case of a spin-1 Bose gas with antiferromagnetic interactions , we from now on consider only spin-1/2 and ferromagnetic spin-1 Bose gases, which have only nonsingular skyrmions as we have seen. Due to the absense of a core, the fluctuations in the density $`\delta n(𝐱)=n(𝐱)n(𝐱)`$ will be small compared to the average density $`n(𝐱)`$ and the energy of the skyrmion can be determined by
$`E[\zeta ]={\displaystyle \frac{1}{2}}{\displaystyle 𝑑𝐱𝑑𝐱^{}\delta n(𝐱)\chi ^1(𝐱,𝐱^{})\delta n(𝐱^{})}`$ (1)
$`+{\displaystyle 𝑑𝐱n(𝐱)\left(\frac{\mathrm{}^2}{2m}|\mathbf{}\zeta (𝐱)|^2\mu 𝐁𝐒(𝐱)\right)}.`$ (2)
Here, $`\chi (𝐱,𝐱^{})`$ is the static density-density response function defined by
$`{\displaystyle \frac{\mathrm{}^2}{4m}}\left(\mathbf{}\left({\displaystyle \frac{1}{n(𝐱)}}\mathbf{}\right)+16\pi a\right)\chi (𝐱,𝐱^{})`$ (3)
$`=\delta (𝐱𝐱^{}),`$ (4)
$`a`$ is the appropriate scattering length, $`𝐁`$ is either a ficticious magnetic field, caused by resonant rf-fields, or a real homogeneous magnetic field and $`\mu `$ is the corresponding magnetic moment of the atoms in the trap. Moreover, the spinor $`\zeta (𝐱)`$ can now be conveniently parametrized as
$$\zeta (𝐱)=\mathrm{exp}\left\{\frac{i}{S}\mathrm{\Omega }^\alpha (𝐱)S_\alpha \right\}\zeta ^\mathrm{Z},$$
(5)
where $`S`$ denotes the spin of the atoms and $`\zeta ^\mathrm{Z}`$ is a constant spinor that minimizes the Zeeman energy, i.e., it obeys $`\zeta _a^\mathrm{Z}=\delta _{aS}`$ if we use the direction of the magnetic field as the quantization axis.
This last equation explicitly shows that the topology of the order parameter space is a sphere of radius $`\pi `$ with opposite points on the surface identified. If we now assume that a maximally symmetric shape of the skyrmion spin texture is allowed, we can take $`𝛀(𝐱)=𝐱\omega (x)/x\widehat{𝐱}\omega (x)`$, with $`\omega (x)`$ a monotonically decreasing function obeying $`\omega (0)=2\pi `$ and $`lim_x\mathrm{}\omega (x)=0`$. For this ansatz we see that by traversing the whole configuration space, we exactly cover the order parameter space once. The above spin texture therefore corresponds to a skyrmion with a topological winding number
$$W=\frac{3}{4\pi ^4}𝑑𝛀=\frac{1}{8\pi ^4}𝑑𝐱ϵ^{ijk}ϵ_{\alpha \beta \gamma }_i\mathrm{\Omega }^\alpha _j\mathrm{\Omega }^\beta _k\mathrm{\Omega }^\gamma $$
(6)
equal to 1 and located at the center of the trap.
To obtain the precise spin texture of the skyrmion we have to find the function $`\omega (x)`$ that minimizes the energy. In general this leads, due to the nonlocality of the density-density response function, to a complicated nonlinear integro-differential equation, which can only be solved numerically. However, we can gain a lot of physical insight in the size and stability of the skyrmion by the following analysis. Considering first the ideal case of zero magnetic field $`𝐁`$ and solving for the density fluctuations induced by the spin texture, we have to minimize the gradient energy
$`E^{\mathrm{grad}}[\zeta ]={\displaystyle 𝑑𝐱n(𝐱)\frac{\mathrm{}^2}{2m}|\mathbf{}\zeta (𝐱)|^2}`$ (7)
$`{\displaystyle \frac{\mathrm{}^4}{8m^2}}{\displaystyle 𝑑𝐱𝑑𝐱^{}|\mathbf{}\zeta (𝐱)|^2\chi (𝐱,𝐱^{})|\mathbf{}\zeta (𝐱^{})|^2}.`$ (8)
For a skyrmion of size $`\lambda `$ in the center of the trap this energy can be estimated to be given by
$`E^{\mathrm{grad}}(\lambda )n(\mathrm{𝟎}){\displaystyle \frac{2\pi \mathrm{}^2}{m}}\left(\lambda {\displaystyle \frac{3\lambda }{(\lambda /\xi )^2+1}}\right),`$ (9)
with $`\xi =(16\pi |a|n(\mathrm{𝟎}))^{1/2}`$ the typical coherence length of the condensate. It has a minimum when $`\lambda /\xi `$ is equal to $`((\sqrt{33}5)/2)^{1/2}0.61`$. Therefore, our maximally symmetric ansatz is indeed justified, even for an anisotropic trap, as long as the coherence length is much smaller than the size of the condensate, i.e., we are in the so-called Thomas-Fermi limit.
For this result we have assumed the scattering length $`a`$ to be positive. In that case the skyrmion is thus stabilized by the fact that gradients in the spin texture on the one hand lead to an increase in the average kinetic energy of the spinor condensate, but one the other hand also lead to a reduction of the density, which for repulsive interactions results in a decrease in the average interaction energy. For a negative scattering length, gradients in the spin texture lead to an enhancement of the density, but this, due to the attractive interactions, also reduces the average interaction energy. Therefore, we expect skyrmions also to be stablilized in that case. Indeed, using the same estimates as before, we find that $`E^{\mathrm{grad}}(\lambda )`$ is now minimized for $`\lambda /\xi `$ equal to $`((\sqrt{33}+5)/2)^{1/2}2.3`$. However, due to the intrinsic instability of a condensate with attractive interactions , this implies that the size of the skyrmion is essentially equal to the size of the condensate itself and that its spin texture will be strongly affected by finite-size effects. In particular, it in general becomes anisotropic.
Note finally, that the above mechanism for the stability of the skyrmion excitations is quite different from that in the quantum Hall effect near filling fraction $`\nu =1`$. There skyrmions are electrically charged and their size is determined by a competition between the Coulomb and Zeeman interactions . In the case of a spinor Bose-Einstein condensate, the Zeeman interaction plays a much less important role and for small magnetic fields $`B8\pi |a|\mathrm{}^2n(\mathrm{𝟎})/m\mu `$ only directs the spins at large distances of the skyrmion. Larger magnetic fields will tend to decrease the size of the skyrmion so as to reduce the cost in Zeeman energy.
Skyrmion dynamics. — The most important dynamical variable of the skyrmion arises from the fact that the Euler-Lagrange equations for the skyrmion spin texture is invariant under a space independent rotation of the average local spin $`𝐒(𝐱)`$ around the magnetic field direction $`\widehat{𝐁}`$ . Mathematically, this means that if $`\zeta ^{\mathrm{sk}}(𝐱)`$ describes a skyrmion, then $`\mathrm{exp}\{i\vartheta \widehat{𝐁}𝐒\}\zeta ^{\mathrm{sk}}(𝐱)`$ describes also a skyrmion with the same winding number and energy. The dynamics of the variable $`\vartheta (t)`$ associated with this symmetry is determined by the full action for the spin texture $`S[\zeta ]=𝑑t(T[\zeta ]E[\zeta ])`$, which contains the time-derivative term
$$T[\zeta ]=𝑑𝐱n(𝐱,t)\zeta ^{}(𝐱,t)i\mathrm{}\frac{}{t}\zeta (𝐱,t).$$
(10)
Hence, subsituting $`\zeta (𝐱,t)=\mathrm{exp}\{i\vartheta (t)\widehat{𝐁}𝐒\}\zeta ^{\mathrm{sk}}(𝐱)`$ and making use of the conservation of total particle number $`N`$ to introduce the change of the average local spin projection on the magnetic field $`\mathrm{\Delta }S_{||}(𝐱)=\widehat{𝐁}𝐒(𝐱)S`$ induced by the skyrmion, we obtain apart from an unimportant boundary term that
$$T[\zeta ]=\mathrm{}\frac{\vartheta (t)}{t}𝑑𝐱n(𝐱,t)\mathrm{\Delta }S_{||}(𝐱).$$
(11)
Taking now again the density fluctuations into account, we finally find that the dynamics of the rotation angle $`\vartheta (t)`$ is determined by the action
$$S[\vartheta ]=𝑑t\left\{\frac{\vartheta (t)}{t}\mathrm{}\mathrm{\Delta }S_{||}^{\mathrm{tot}}+\frac{1}{2}I\left(\frac{\vartheta (t)}{t}\right)^2\right\},$$
(12)
where $`\mathrm{\Delta }S_{||}^{\mathrm{tot}}`$ is the change of the total spin along the magnetic field direction and the ‘moment of inertia’ of the skyrmion equals
$$I=\mathrm{}^2𝑑𝐱𝑑𝐱^{}\mathrm{\Delta }S_{||}(𝐱)\chi (𝐱,𝐱^{})\mathrm{\Delta }S_{||}(𝐱^{}).$$
(13)
The importance of this result is twofold. First, from the action in Eq. (12) we see that at the quantum level the hamiltonian for the dynamics of the wave function $`\mathrm{\Psi }(\vartheta ,t)`$ becomes
$$H=\frac{1}{2I}\left(\frac{\mathrm{}}{i}\frac{}{\vartheta }\mathrm{}\mathrm{\Delta }S_{||}^{\mathrm{tot}}\right)^2.$$
(14)
Therefore, the ground state wave function is given by $`\mathrm{\Psi }_0(\vartheta )=e^{iK\vartheta }/\sqrt{2\pi }`$, with $`K`$ an integer that is as close as possible to $`\mathrm{\Delta }S_{||}^{\mathrm{tot}}`$. In this way we thus recover the fact that according to quantum mechanics the total number of spin-flips associated with the skyrmion texture must be an integer. More precisely, we have actually shown that the many-body wave function describing the skyrmion is an eigenstate of the operator $`\widehat{𝐁}𝐒^{\mathrm{tot}}`$ with eigenvalue $`NSK`$. Note that physically this is equivalent to the way in which ‘diffusion’ of the overall phase of a Bose-Einstein condensate leads to the conseration of particle number .
Furthermore, the existence of this internal degree of freedom becomes especially important when we deal with more than one skyrmion in the condensate. In that case every skyrmion can have its own orientation and we expect the interaction between two skyrmions to have a Josephson-like contribution proportional to $`\mathrm{cos}(\vartheta _1\vartheta _2)`$. As a result the phase diagram of a gas of skyrmions can become extremely rich and contain both quantum as well as classical, i.e., nonzero temperature, phase transitions . In this context, it is interesting to mention two important differences with the situation in the quantum Hall effect. First, the fact that the spin projection $`K`$ of the skyrmion is an integer shows that these excitations have an integer spin and are therefore bosons . In the quantum Hall case the skyrmions are fermions with half-integer spin, due to the presence of a topological term in the action $`S[\zeta ]`$ for the spin texture . Second, in a spinor Bose-Einstein condensate the skyrmions are not pinned by disorder and are in principle free to move. Both differences will clearly have important consequences for the many-body physics of a skyrmion gas.
Focussing again on the single skyrmion dynamics, we can make the last remark more quantitative by using the ansatz $`\zeta (𝐱,t)=\zeta ^{\mathrm{sk}}(𝐱𝐮(t))`$ for the texture of a moving skyrmion, which is expected to be accurate for small velocities $`𝐮(t)/t`$. Considering for illustrative purposes also only the case of a maximally symmetric skyrmion near the center of the trap with $`u/\xi 1`$, we find in a similar way as before that the action for the center-of-mass motion of the skyrmion becomes
$$S[𝐮]=𝑑t\frac{1}{2}M\left(\frac{𝐮(t)}{t}\right)^2,$$
(15)
where the mass, which in the more general anisotropic case is of course a tensor, is now simply given by
$$M=\frac{m^2}{3}𝑑𝐱𝑑𝐱^{}\chi (𝐱,𝐱^{})𝐯_\mathrm{s}(𝐱)𝐯_\mathrm{s}(𝐱^{})$$
(16)
in terms of the superfluid velocity of the spinor condensate $`𝐯_\mathrm{s}(𝐱)i\mathrm{}\zeta _{}^{\mathrm{sk}}{}_{}{}^{}(𝐱)\mathbf{}\zeta ^{\mathrm{sk}}(𝐱)/m`$. The skyrmions thus indeed behave in this respect as ordinary particles. It should be noted that in contrast to Eq. (12) there appears no term linear in $`𝐮(t)/t`$ in the action $`S[𝐮]`$. This is a result of the fact that we have performed all our calculations at zero temperature. In the presence of a normal component, we anticipate the appearance of such a linear term with an imaginary coefficient. This will lead to damping of the center-of-mass motion of the skyrmion.
I would like to thank Michiel Bijlsma, Gerard ’t Hooft, David Olive, and Jan Smit for useful discussions. This work is supported by the Stichting Fundamenteel Onderzoek der Materie (FOM), which is financially supported by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO).
|
warning/0002/gr-qc0002071.html
|
ar5iv
|
text
|
# Open universes and avoidance of the cosmological singularity
## I Introduction
In recent papers we claimed that canonical Einstein’s general relativity(GR) with an extra scalar field and its conformal formulation (called ’Jordan frame GR’ therein), are different but physically equivalent representations of a same theory. Our claim was based on the following argument. The spacetime coincidences (coordinates) are not affected by a conformal rescaling of the spacetime metric of the kind
$$\widehat{g}_{ab}=\mathrm{\Omega }^2g_{ab},$$
(1)
where $`\mathrm{\Omega }^2`$ is a smooth, nonvanishing function on the spacetime manifold. Correspondingly, the experimental observations (measurements), being nothing but just verifications of these coincidences, are unchanged too by the transformations (1.1). This means that canonical GR and its conformal are experimentally indistinguishable. However, we are dissatisfied with the exposition of our ideas in reference since it may lead to misunderstanding of our view-point and, correspondingly, to misinterpretation of the results presented therein.
The main objection against our claim in ref. can be based on the following argument. In canonical GR the matter fields couple minimally to the metric that determines metrical relations on a Riemann spacetime, say $`\widehat{𝐠}`$. In this case matter particles follow the geodesics of the metric $`\widehat{𝐠}`$ (on Riemann geometry), and their masses are constant over the spacetime manifold, i.e., it is the metric which matter ’feels’ or, if convenient, the ’physical’ metric. Under the conformal rescaling (1.1), the matter fields become non-minimally coupled to the conformal metric $`𝐠`$. Hence, matter particles do not follow the geodesics of this last metric. Furthermore, it is not the metric that determines metrical relations on a Riemann manifold. This line of reasoning leads to the following conclusion. Although canonical GR and its conformal may be physically equivalent theories, nevertheless, the physical metric is that which determines metrical relations on a Riemann spacetime. Hence, the result that, in the conformal formulation of general relativity, flat (barotropic) Friedmann-Robertson-Walker (FRW) universes are free of the cosmological singularity (in the region $`\frac{3}{2}\omega \frac{4}{3}`$, $`0<\gamma <2`$ of the parameter space), is not relevant. In fact, the conformal (singularity-free) metric $`𝐠`$, is not the physical metric.
In the present paper, we shall show that the above conclusion is wrong. It is a consequence of the long-standing confusion in the understanding of metric theories of gravity, when conformal transformations of the metric are concerned . There is a missing detail when these transformations are studied in the literature. In fact, under the conformal rescaling of the metric (1.1), not only the Lagrangian of the theory is mapped into its conformal Lagrangian, but the spacetime geometry itself is mapped too into a conformal geometry. In this last geometry, metrical relations involve both the conformal metric $`𝐠`$ and the conformal factor $`\mathrm{\Omega }^2`$ generating the transformation (1.1). Hence, in the conformal Lagrangian, the matter fields should ’feel’ both the metric $`𝐠`$ and the scalar function $`\mathrm{\Omega }`$. I.e., the matter particles would not follow the geodesics of the conformal metric alone. The result is that, under the transformation (1.1), the ’physical’ metric of the untransformed geometry is effectively mapped into the ’physical’ metric of the conformal geometry. This missing detail is the source of the long-standing confusion in former studies of conformal transformations of the kind (1.1). We aim section II of this paper, precisely, at a clarification of this situation, since it is of great importance for the understanding of our view-point.
Yet another question respecting metric theories of spacetime has been long avoided in the literature. It is linked with a program Brans and Dicke outlined in their classical paper . In that paper the authors made evident their hope that, the physical content of a given theory of spacetime should be contained in the invariants of the group of position-dependent transformations of units and coordinate transformations. The last part of this program has been already worked-out. By the present, invariance under the group of general coordinate transformations is the minimal requirement any theory, pretending to be a real alternative for the formulation of the laws of gravity, must fulfil<sup>*</sup><sup>*</sup>*All known metric theories of spacetime, including GR, Brans-Dicke theory and scalar-tensor theories in general, fulfil this requirement.. It is evident that any consistent formulation of a given effective theory of spacetime must be invariant, also, under the group of the transformations of the units of length, time and mass. However, this part of the program Brans and Dicke outlined in , has not been yet worked-out sufficiently. This important subject is treated in detail in section III of the present paper. In this section we study the invariance properties of the different formulations of Brans-Dicke (BD) theory and general relativity, under a particular, one-parameter Abelian group of transformations, that can be identified with the group of transformations of the units of length, time and reciprocal mass. We shall show that the only consistent formulation of the laws of gravity (among those studied here), is the conformal representation of general relativity.
In this context, the result presented in ref. that flat FRW universes, in the frame of conformal GR, are free of the cosmological singularity (in the given region of the parameter space), acquires a new dimension. The occurrence of spacetime singularities may be interpreted as a result of a wrong choice of the formulation of the given spacetime theory. In section IV we further extend the result of paper to open FRW universes.
A discussion on the results obtained in the present and former papers (), and on the physical significance of our view-point, is given in section V.
## II Effective theories of spacetime and conformal geometries
In this section we shall investigate the effect of a conformal transformation of the kind (1.1) on the laws of gravity and on the geometry. For illustration we shall study general relativity with an extra scalar field. The canonical Lagrangian for this theory is
$$\widehat{}_{GR}=\sqrt{\widehat{g}}(\widehat{R}\alpha (\widehat{}\widehat{\varphi })^2)+16\pi \sqrt{\widehat{g}}L_{matter},$$
(2)
where $`\widehat{R}`$ is the Ricci scalar given in terms of the untransformed metric $`\widehat{𝐠}`$, $`(\widehat{}\widehat{\varphi })^2=\widehat{g}^{mn}\widehat{\varphi }_{,m}\widehat{\varphi }_{,n}`$, $`\alpha `$ is a free parameter ($`\alpha 0`$) and $`L_{matter}`$ is the Lagrangian of the matter fields. When $`\varphi `$ is a constant, or when $`\alpha =0`$ (arbitrary $`\varphi `$), we recover usual Einstein’s theory. Under the conformal rescaling (1.1) with $`\mathrm{\Omega }^2=e^{\widehat{\varphi }}`$, the Lagrangian (2.1) is mapped into its conformal Lagrangian
$$_{GR}=\sqrt{g}e^{\widehat{\varphi }}(R(\alpha \frac{3}{2})(\widehat{\varphi })^2)+16\pi \sqrt{g}e^{2\widehat{\varphi }}L_{matter},$$
(3)
where now $`R`$ is the Ricci scalar given in terms of the conformal metric $`𝐠`$. This Lagrangian can be given a more usual, Brans-Dicke form, after the change of variable $`\widehat{\varphi }\varphi =e^{\widehat{\varphi }}`$,
$$_{GR}=\sqrt{g}(\varphi R(\alpha \frac{3}{2})\frac{(\varphi )^2}{\varphi })+16\pi \sqrt{g}\varphi ^2L_{matter}.$$
(4)
Due to the minimal coupling of the scalar field $`\widehat{\varphi }`$ to the curvature in canonical GR (Lagrangian (2.1)), the effective gravitational constant $`\widehat{G}`$ (set equal to unity in (2.1)) is a real constant. The minimal coupling of the matter fields to the metric in (2.1), warrants that matter particles follow the geodesics of the metric $`\widehat{𝐠}`$, that is the metric which defines metrical relations on Riemann spacetimes. Hence, the inertial mass $`\widehat{m}`$ of some elementary particle is constant too over the spacetime. This leads that the dimensionless gravitational coupling constant $`\widehat{G}\widehat{m}^2`$ ($`c=\mathrm{}=1`$) is constant in spacetime unlike BD theory, where $`\widehat{G}\widehat{m}^2`$ evolves like $`\varphi ^1`$. This is a conformal invariant feature of general relativity since dimensionless constants do not change under (1.1). In other words, in conformal general relativity, the dimensionless gravitational constant $`Gm^2`$ is a constant as well ($`G`$ and $`m`$ are the effective gravitational constant and the inertial mass of some elementary particle respectively, given in the conformal GR). However, in this last case (Lagrangian (2.2) or (2.3)), the effective gravitational constant varies like $`Ge^{\widehat{\varphi }}`$ (or $`\varphi ^1`$). Since $`\widehat{m}^2=Gm^2`$ ($`\widehat{G}=1`$), then, in this formulation of general relativity, particle masses vary over the spacetime manifold like
$$m=e^{\frac{1}{2}\widehat{\varphi }}\widehat{m}.$$
(5)
According to Dicke , the conformal transformation (1.1) (in $`\mathrm{\Omega }^2=e^{\widehat{\varphi }}=\varphi `$) can be interpreted as a transformation of the units of length, time and reciprocal mass. In fact, if one chooses the arc-length as one’s unit of length and time, and the inertial mass of some elementary particle as one’s unit of mass, then, we see that $`ds=e^{\frac{1}{2}\widehat{\varphi }}d\widehat{s}`$, while $`m^1=e^{\frac{1}{2}\widehat{\varphi }}\widehat{m}^1`$, i.e., these measuring scales change in the same way over the manifold.
A careful looking at eqs.(2.1), (2.2,2.3) shows that the Einstein’s laws of gravity derivable from (2.1), change under the units transformation (1.1). This seems to be a serious drawback of canonical GR (and BD theory and scalar-tensor theories in general) since, it is obvious that, in any consistent theory of spacetime, the laws of physics must be invariant under a change of the units of length, time and mass. This subject will be discussed in detail in the next section. It will be shown that the transformation (1.1) with $`\mathrm{\Omega }^2=e^{\widehat{\varphi }}=\varphi `$, can not be taken properly as a units transformation. It is just a transformation that allows ’jumping’ from one formulation of the given spacetime theory into its conformal.
In ref. we claimed that canonical GR (Lagrangian (2.1)) and its conformal (Lagrangian (2.2) or (2.3)) are physically equivalent theories, since they are indistinguishable from the observational point of view. However, it is of common belief, that only one of the conformally related metrics is the ’physical’ metric, i.e., that which determines metrical relations on the spacetime manifold. The reasoning line leading to this conclusion is based on the following analysis. Take, for instance, general relativity with an extra scalar field. Due to the minimal coupling of the matter fields to the metric in (2.1), the matter particles follow the geodesics of the metric $`\widehat{𝐠}`$,
$$\frac{d^2x^a}{d\widehat{s}^2}+\widehat{\mathrm{\Gamma }}_{mn}^a\frac{dx^m}{d\widehat{s}}\frac{dx^n}{d\widehat{s}}=0,$$
(6)
where $`\widehat{\mathrm{\Gamma }}_{bc}^a=\frac{1}{2}\widehat{g}^{an}(\widehat{g}_{bn,c}+\widehat{g}_{cn,b}\widehat{g}_{bc,n})`$ are the Christoffel symbols of the metric $`\widehat{𝐠}`$. These coincide with the geodesics of the Riemann geometry, where metrical relations are given by $`\widehat{𝐠}`$ through the expressions for the scalar product of two vectors $`\widehat{𝐗}`$ and $`\widehat{𝐘}`$, $`\widehat{g}(\widehat{𝐗},\widehat{𝐘})=\widehat{g}_{mn}\widehat{X}^m\widehat{Y}^n`$, the line-element $`d\widehat{s}^2=\widehat{g}_{mn}dx^mdx^n`$, etc. It is the reason why canonical GR, based on the Lagrangian (2.1), is naturally linked with Riemann geometryThe same is true for the Jordan frame formulation of Brans-Dicke theory since, in this frame, the matter fields couple minimally to the spacetime metric.. The units of this geometry are constant over the manifold. This requirement is realized in canonical general relativity and in the Jordan frame formulation of Brans-Dicke theory, through the assumption that there exists a large class of physical systems (such like atoms) having properties that are independent of location . This is equivalent to say that one can take some quantities associated with these systems (for instance the atom radius and transition energies) as one’s units of measurement. These will be constant over the manifold. Hence, for instance, the arc-length between two successive events on a geodesic curve will be point-independent, as required by the behavior of the units of measure of Riemann geometry.
On the other hand, since the matter fields are non-minimally coupled to the metric in the conformal GR, the matter particles would not follow the geodesics of the conformal metric $`𝐠`$. These will follow curves that are solutions of the equation conformal to (2.5) instead
$$\frac{d^2x^a}{ds^2}+\mathrm{\Gamma }_{mn}^a\frac{dx^m}{ds}\frac{dx^n}{ds}+\frac{\varphi _{,n}}{2\varphi }(\frac{dx^n}{ds}\frac{dx^a}{ds}g^{na})=0,$$
(7)
where now $`\mathrm{\Gamma }_{bc}^a`$ are the Christoffel symbols of the metric $`𝐠`$ conformal to $`\widehat{𝐠}`$. Hence, if one assumes that the spacetime geometry is fixed to be Riemannian and that it is unchanged under the conformal rescaling (1.1) with $`\mathrm{\Omega }^2=\varphi `$, one effectively arrives at the conclusion that $`\widehat{𝐠}`$ is the ’physical’ metric. However, this assumption is wrong and is the source of the long-standing confusion in the understanding of the meaning of the conformal transformations of the metric (1.1). We shall show this in the remainder of this section.
### A Conformal Riemann geometry
Let $`\lambda (t)`$ be a curve on the spacetime manifold with local coordinates $`x^a(t)`$ and let $`𝐗`$ with local coordinates $`X^a=\frac{dx^a}{dt}`$, be a vector tangent to $`\lambda (t)`$. The covariant derivative of a given vector field $`\widehat{𝐘}`$ along $`\lambda `$ is given byWe use the symbology of reference .
$$\frac{\widehat{D}\widehat{Y}^a}{t}=\frac{\widehat{Y}^a}{t}+\widehat{\gamma }_{mn}^a\widehat{Y}^m\frac{dx^n}{dt},$$
(8)
where $`\widehat{\gamma }_{bc}^a`$ is a symmetric connection on the manifold. The vector $`\widehat{𝐘}`$ is said to be parallely transported along $`\lambda `$ if $`\frac{\widehat{D}\widehat{𝐘}}{t}=0`$. In particular, if one considers the covariant derivative of the tangent vector itself along $`\lambda `$, then, the curve $`\lambda (t)`$ is said to be a geodesic curve if $`\frac{\widehat{D}}{t}(\frac{}{t})_\lambda `$ is parallel to $`(\frac{}{t})_\lambda `$. In other words, there exists a function $`f`$ such that $`X^n\widehat{D}_nX^a=fX^a`$. For such a curve, one can find an affine parameter $`\widehat{v}(t)`$ along the curve such that $`\frac{\widehat{D}}{\widehat{v}}(\frac{}{\widehat{v}})_\lambda =0`$. The associated tangent vector $`\widehat{𝐕}=(\frac{}{\widehat{v}})_\lambda `$ is parallel to $`\widehat{𝐗}`$ and its scale is determined by $`\widehat{V}(\widehat{v})=1`$. It obeys the equations
$$\widehat{V}^n\widehat{D}_n\widehat{V}^a=0.$$
(9)
Given a metric $`\widehat{𝐠}`$ on the manifold $`\widehat{}`$, the Riemann geometry is fixed by the following postulate. There is a unique torsion-free (symmetric) connection on $`\widehat{}`$ defined by the condition that the covariant derivative of $`\widehat{𝐠}`$ is zero, i.e. $`\widehat{D}_c\widehat{g}_{ab}=0`$. With the connection defined in such a way, parallel transfer of vectors, for instance, $`\widehat{𝐘}`$
$$\frac{\widehat{D}\widehat{Y}^a}{t}=\frac{\widehat{Y}^a}{t}+\widehat{\gamma }_{mn}^a\widehat{Y}^m\frac{dx^n}{dt}=0,$$
(10)
preserves scalar products defined by $`\widehat{𝐠}`$, in particular
$$dg(\widehat{𝐘},\widehat{𝐘})=0,$$
(11)
where $`g(\widehat{𝐘},\widehat{𝐘})=\widehat{g}_{mn}\widehat{Y}^m\widehat{Y}^n`$. The laws of parallel transport (2.9) and length preservation (2.10) together lead that the symmetric connection $`\widehat{\gamma }_{bc}^a`$ on a Riemann manifold, coincides with the Christoffel symbols of the metric $`\widehat{𝐠}`$; $`\widehat{\gamma }_{bc}^a=\widehat{\mathrm{\Gamma }}_{bc}^a`$. Then, eq.(2.8) defining a geodesic curve on the Riemann manifold, can be written as
$$\frac{d^2x^a}{d\widehat{v}^2}+\widehat{\mathrm{\Gamma }}_{mn}^a\frac{dx^m}{d\widehat{v}}\frac{dx^n}{d\widehat{v}}=0.$$
(12)
If $`\widehat{𝐕}`$ is a time-like vector, then, in particular, the affine parameter $`\widehat{v}`$ can be set equal to the arc-length $`\widehat{s}`$.
Suppose that, under the conformal rescaling (1.1) with $`\mathrm{\Omega }^2=\varphi `$, the vectors $`\widehat{𝐘}`$ and $`\widehat{𝐕}`$ transform in the following ways,
$$\widehat{Y}^a=h(\varphi )Y^a,\widehat{V}^a=j(\varphi )V^a,$$
(13)
where $`h`$ and $`j`$ are smooth functions of the scalar field $`\varphi `$. Hence, the law (2.10) of preservation of length of a vector $`\widehat{𝐘}`$ in Riemann geometry, is transformed into the following law of length transport for the conformal vector $`𝐘`$,
$$dg(𝐘,𝐘)=d[\mathrm{ln}(\varphi h^2)]g(𝐘,𝐘).$$
(14)
This resembles the law of length transport in Weyl geometry. Hence, given a Riemann geometry on $`\widehat{}`$, under (1.1) with $`\mathrm{\Omega }^2=\varphi `$, it is transformed into a Weyl geometry on $``$. The parallel transport law conformal to (2.9), can be written as
$$\frac{DY^a}{t}+\frac{}{t}(\mathrm{ln}h)Y^a=0,$$
(15)
where the following definition for the conformal covariant derivative has been used
$$\frac{DY^a}{t}=\frac{Y^a}{t}+\gamma _{mn}^aY^m\frac{dx^n}{dt}=0.$$
(16)
$`\gamma _{bc}^a`$ is the symmetric connection on the Weyl manifold. It is related with the Christoffel symbols of the conformal metric $`𝐠`$ through
$$\gamma _{bc}^a=\mathrm{\Gamma }_{bc}^a+\frac{1}{2}\varphi ^1(\varphi _{,b}\delta _c^a+\varphi _{,c}\delta _b^ag_{bc}g^{an}\varphi _{,n}).$$
(17)
If one applies eq.(2.14) to the vector $`𝐕=(\frac{}{v})_\lambda `$ that is tangent to the curve $`\lambda `$ (it fulfils $`V^nD_nV^a=fV^a`$, where$`f=\frac{}{v}(\mathrm{ln}j)`$), then, we obtain the local coordinate expression for the geodesic equation that is conformal to (2.11),
$$\frac{d^2x^a}{dv^2}+\mathrm{\Gamma }_{mn}^a\frac{dx^m}{dv}\frac{dx^n}{dv}+\frac{\varphi _{,n}}{2\varphi }(2\frac{dx^n}{dv}\frac{dx^a}{dv}g^{na})+\frac{j_{,n}}{j}\frac{dx^n}{dv}\frac{dx^a}{dv}=0.$$
(18)
The main feature of Weyl geometry, being based on the laws of parallel transport (2.14) and length transport (2.13), is that the units of measure of this geometry are point-dependent. In particular, the law (2.13) applied to the arc-length, leads that the line-element changes from point to point in spacetime like $`ds^2\varphi ^1`$. In other words, the arc-length between two neighboring events on a geodesic curve is different for different points on the curve. Hence, Weyl geometry represents a generalization of Riemann geometry to include units of measure with point-dependent length.
### B Linkage between effective theories of spacetime and conformal geometries
When an effective theory of spacetime is approached, some conclusions raise. First, theories with minimal coupling of the matter fields to the metric are naturally linked with Riemann geometry with constant units of measure, while theories with non-minimal coupling are linked with Weyl geometry with varying length units of measure. Conformal GR is an example of this last case. In fact, we see that the geodesic equation for a time-like vector $`𝐕=(\frac{}{s})_\lambda `$ on Weyl geometry (eq.(2.17) with $`v=s`$ and $`j=\varphi ^{\frac{1}{2}}`$) exactly coincides with the equation (2.6) defining free-motion time-like path in conformal general relativity. The second conclusion is connected with the fact that, in Weyl geometry, metrical relations are given by the metric $`𝐠`$ and by the scalar field $`\varphi `$. This last field determines how metrical relations change from point to point on the Weyl manifold (see eq.(2.13)). Hence, on Weyl geometry, any matter field would ’feel’ both the metric $`𝐠`$ and the scalar field $`\varphi `$, i.e., matter would be coupled both to $`𝐠`$ and to $`\varphi `$ (non-minimal coupling). Following the same reasoning-line leading to the identification of $`\widehat{𝐠}`$ as the physical metric in canonical general relativity, we reach to the following crucial conclusion. The conformal transformation (1.1) with $`\mathrm{\Omega }^2=\varphi `$, maps the ’physical’ metric $`\widehat{𝐠}`$ on Riemann geometry into the ’physical’ metric $`𝐠`$ on its conformal (Weyl) geometry. Correspondingly, geodesic curves on Riemann spacetimes are mapped into geodesic curves on Weyl spacetimes. In particular, incomplete geodesics on a Riemann spacetime can, in principle, be mapped into complete geodesics on the conformal spacetimes. Hence, spacetime singularities that plage the canonical (Riemannian) general relativity, may be avoided in conformal general relativity linked with Weyl geometry. This subject will be treated in section IV, for Friedmann-Robertson-Walker spacetimes.
## III Effective theories of spacetime and transformations of units
We shall study two kinds of Lagrangians for pure gravity,
$$_1=\sqrt{g}(R\alpha (\varphi )^2),$$
(19)
and
$$_2=\sqrt{g}(\varphi R(\alpha \frac{3}{2})\frac{(\varphi )^2}{\varphi }),$$
(20)
in respect to their transformation properties under rescalings of the units of length, time and reciprocal mass of the kind (1.1)<sup>§</sup><sup>§</sup>§Lagrangian (3.2) can be obtained from (3.1) if we rescale $`𝐠\varphi 𝐠`$ and change $`\varphi \mathrm{ln}\varphi `$.. We shall interested, in particular, in the following conformal transformation:
$$\stackrel{~}{g}_{ab}=\varphi ^\sigma g_{ab},$$
(21)
where $`\sigma `$ is some arbitrary parameter. Under (3.3) the Lagrangian $`_1`$ changes into
$$\stackrel{~}{}_1=\sqrt{\stackrel{~}{g}}(\varphi ^\sigma \stackrel{~}{R}+((3\sigma \frac{3}{2}\sigma ^2)\varphi ^{2\sigma }\alpha \varphi ^\sigma )(\stackrel{~}{}\varphi )^2),$$
(22)
so the laws of gravity it describes, change under (3.3). In particular, in the conformal frame (magnitudes with tilde), the effective gravitational constant depends on $`\varphi `$ due to the non-minimal coupling between the scalar field $`\varphi `$ and the curvature. Lagrangian (3.2), on the other hand, is mapped into
$$\stackrel{~}{}_2=\sqrt{\stackrel{~}{g}}(\varphi ^{1\sigma }\stackrel{~}{R}\frac{(\alpha \frac{3}{2}3\sigma +\frac{3}{2}\sigma ^2)}{(1\sigma )^2}\varphi ^{\sigma 1}(\stackrel{~}{}\varphi ^{1\sigma })^2).$$
(23)
Hence, if we introduce a new scalar field variable
$$\stackrel{~}{\varphi }=\varphi ^{1\sigma },$$
(24)
and redefine the free parameter of the theory
$$\stackrel{~}{\alpha }=\frac{\alpha +3\sigma (\sigma 2)}{(1\sigma )^2},$$
(25)
then, the Lagrangian $`\stackrel{~}{}_2`$ (eq.(3.5)) can be put in the following form
$$\stackrel{~}{}_2=\sqrt{\stackrel{~}{g}}(\stackrel{~}{\varphi }\stackrel{~}{R}(\alpha \frac{3}{2})\frac{(\stackrel{~}{}\stackrel{~}{\varphi })^2}{\stackrel{~}{\varphi }}),$$
(26)
i.e., the Lagrangian $`_2`$ is invariant in form under the conformal transformation (3.3), the scalar field redefinition (3.6) and the parameter transformation (3.7). To our knowledge, these transformations were first studied in . The composition of two successive transformations (3.3), (3.6) and (3.7), with parameters $`\sigma _1`$ and $`\sigma _2`$, gives a transformation of the same kind with parameter $`\sigma _3`$,
$$\sigma _3=\sigma _1+\sigma _2\sigma _1\sigma _2.$$
(27)
The identity transformation is that with $`\sigma =0`$, while the inverse of the transformation with parameter $`\sigma `$, is a transformation with parameter $`\overline{\sigma }=\frac{\sigma }{1\sigma }`$. Hence, if we exclude the value $`\sigma =1`$, then (3.3), (3.6) and (3.7) form a one-parameter Abelian group of transformations ($`\sigma _3(\sigma _1,\sigma _2)=\sigma _3(\sigma _1,\sigma _2)`$). Since (3.3) can be interpreted as a transformation of the units of length, time and reciprocal mass, we shall call this as the one-parameter group of transformations of these units. We see that the transformation (3.3) with $`\sigma =1`$ does not belong to this group (the inverse does not exist) and, hence, it can not be interpreted properly as a units transformationTransformation of units is a group-theoretic technique..
A very important conclusion raises. Since any consistent theory of spacetime must be invariant under the one-parameter group of transformations of the units of length, time and mass, then, spacetime theories whose Lagrangian for pure gravity is of the kind $`_1`$, are not consistent theories while, those based on Lagrangians of the kind $`_2`$ may, in principle, be consistent formulations of a spacetime theory. Hence, for instance, canonical GR and the Einstein frame formulation of BD theory are not consistent formulations of the laws of gravity.
We shall now consider, separately, the following matter Lagrangians
$$\sqrt{g}\varphi ^2L_{matter},$$
(28)
and
$$\sqrt{g}L_{matter}.$$
(29)
The Lagrangian (3.11) shows minimal coupling of matter to the metric, while Lagrangian (3.10) shows non-minimal coupling. Under (3.3) the Lagrangian (3.10) is transformed in the following way,
$$\sqrt{g}\varphi ^2L_{matter}=\sqrt{\stackrel{~}{g}}\varphi ^{22\sigma }L_{matter},$$
(30)
and, hence, considering the scalar field redefinition (3.6), we complete the demonstration that (3.10) is invariant in form under the one-parameter group of transformations of the units of length, time and reciprocal mass. Unfortunately, it is straightforward that the Lagrangian (3.11) with minimal coupling, is not invariant under this group of units transformations. The consequence is that Brans-Dicke theory (its Jordan frame formulation) based on the Lagrangian
$$_{BD}=_2+16\pi \sqrt{g}L_{matter},$$
(31)
is not yet a consistent theory of spacetime. The only surviving theory is the conformal general relativity based on the Lagrangian (2.3), i.e.,
$$_{GR}=_2+16\pi \sqrt{g}\varphi ^2L_{matter}.$$
(32)
This theory provides a consistent formulation of the laws of gravity. In fact, the laws derived from (2.3) (or (3.14)) are invariant under the one-parameter Abelian group of units transformations (3.3), (3.6) and (3.7) as required. We think, it is a very serious argument to take in mind while studying different alternatives for a final theory of spacetime.
It is not casual that the laws under which Weyl geometry is based (laws of parallel transport (2.14) and length transport (2.13)), are invariant too under the transformations of this group. In fact, as we have already shown, the conformal formulation of general relativity (Lagrangian (3.14)), being invariant under the transformations of this group, is naturally linked with Weyl geometry. Unlike this, Riemann geometry is not invariant under (3.3) and (3.6). The parallel transport law (2.9) and the length preservation law (2.9), change under these transformations. Hence, Riemann geometry is not a consistent framework for the interpretation of the laws of gravity while Weyl geometry does it.
Finally, we shall remark that the conformal rescaling of the metric (1.1) with $`\mathrm{\Omega }^2=\varphi `$ (the case $`\sigma =1`$ in (3.3)) can not be interpreted properly as a units transformation, since it does not belong to the one-parameter Abelian group studied above. The role of this particular transformation is just to warrant ’jumping’ from the inconsistent canonical formulation of GR, into the consistent conformal formulation of this theory. At the same time, it permits ’jumping’ from the inconsistent Riemann geometry into the consistent Weyl geometry.
## IV Conformal general relativity and the cosmological singularity
In reference we showed that the cosmological singularity that is always present in the canonical (Einstein frame) formulation of general relativity, may be removed when we work in conformal representation of this theory (in the ’plus’ branch of the solution). This result was obtained for Friedmann-Robertson-Walker(FRW), flat universes filled with a barotropic perfect fluid. The absence of singularities in the conformal frame was expected since, in the ’+’ branch of the solution $`R_{mn}k^mk^n`$ may be negative definite (in the given region of the parameter space) for any non-spacelike vector $`𝐤`$. This means that the relevant singularity theorems may not holdThe vanishing of the cosmological singularity can be explained, in other words, as due to the fact that, under (1.1) with $`\mathrm{\Omega }^2=\varphi `$, incomplete geodesics on a Riemann FRW spacetime, are mapped into complete geodesics on the conformal (Weyl) spacetime, in a given region of the parameter space.. This is in contradiction with the canonical formulation where $`\widehat{R}_{mn}\widehat{k}^m\widehat{k}^n`$ is non-negative and a space-like singularity at the beginning of time $`t=0`$ always occurs. These are not alternative theories, but just alternative representations of the same theory . Both of them are observationally equivalent. However, observational evidence should be interpreted either on the grounds of Riemann geometry (with constant units of measure) or on the grounds of Weyl geometry (with varying length units of measure), depending on the formulation of general relativity we chose for modeling the universe.
In the light of our previous discussion in sections II and III, the result that the cosmological singularity may be removed (in some region of the parameter space of the theory), in the conformal representation of general relativity, is very interesting since, it suggests that the spacetime singularity may be a spurious entity due to a wrong choice of the formulation of GR. In fact, the conformal transformation (1.1) with $`\mathrm{\Omega }^2=\varphi `$, effectively removes the cosmological singularity occurring in canonical Einstein’s GR on Riemann manifolds. As we have already shown, this last representation of general relativity provides an inconsistent formulation of the laws of gravity. Weyl geometry, on the contrary, is a consistent framework where to formulate the laws of gravitation.
Another argument in this direction is connected with the fact that the canonical formulation of general relativity is a classical theory of spacetime and, it is the hope that, when a final quantum theory of gravity will be worked out, the singularity will be removed. In the conformal representation of general relativity no singularity occurs even without including quantum considerations. Other arguments come from string theory where a scalar field (the dilaton) is always coupled to the curvature.
In this section we shall further extend the results of to open FRW universes filled with a barotropic perfect fluid.
The field equations derivable from (2.1), for open FRW universes, can be reduced to the following equation for determining the untransformed scale factor $`\widehat{a}`$,
$$(\frac{\dot{\widehat{a}}}{\widehat{a}})^2\frac{1}{\widehat{a}^2}=\frac{M}{\widehat{a}^{3\gamma }}+\frac{\alpha N}{6\widehat{a}^6},$$
(33)
where, in order that the results here were consistent with the symbology used in ref., we should realize that $`\alpha =\omega +\frac{3}{2}`$. $`N`$ and $`M`$ are arbitrary integration constants and the barotropic index $`\gamma `$ is in the range $`0<\gamma <2`$. While deriving eq.(4.1) we have taken into account that, in the untransformed frame, the ordinary matter energy density is given by $`\widehat{\mu }=\frac{3}{8\pi }\frac{M}{\widehat{a}^{3\gamma }}`$. After integrating once the wave equation $`\widehat{\mathrm{}}\widehat{\varphi }=0`$ for the scalar field $`\widehat{\varphi }`$, we obtain:
$$\dot{\widehat{\varphi }}=\pm \frac{\sqrt{N}}{\widehat{a}^3}.$$
(34)
This equation has been considered also while deriving eq.(4.1). The curvature scalar for an open FRW universe is found to be
$$\widehat{R}=\frac{3M}{\widehat{a}^6}[(43\gamma )\widehat{a}^{3(2\gamma )}\frac{\alpha N}{3M}].$$
(35)
Hence at $`\widehat{a}=0`$ there is a curvature singularity which corresponds to the initial cosmological singularity. In fact, with the help of eq.(4.1) for $`\widehat{a}1`$, we find that $`t\widehat{a}^3`$. Hence the correspondence $`\widehat{a}0t0`$.
Our goal here is to show that this curvature singularity is removed when we ’jump’ to the conformal formulation of general relativity. For this purpose we shall first study the Raychaudhuri equation for a congruence of fluid lines and then we shall study, in detail, the behaviour of the relevant magnitudes and relationships in the conformal formulation of GR. Finally we shall study the particular case with $`\alpha =0`$ ($`\omega =\frac{3}{2}`$) for dust-filled and radiation-filled universes since, in these cases, exact analytic solutions are easily found.
The Raychaudhuri equation for a congruence of fluid lines without vorticity and shear, with the time-like tangent vector $`\widehat{k}^a=\delta _0^a`$, can be written (in the canonical formulation of GR) as:
$$\dot{\widehat{\theta }}=\widehat{R}_{00}\frac{1}{3}\widehat{\theta }^2,$$
(36)
where the overdot means derivative with respect to the untransformed proper time $`t`$ and $`\widehat{\theta }`$ is the volume expansion. In eq.(4.4) we took the reversed sense of time $`\mathrm{}t0`$, i.e. $`\widehat{a}`$ runs from infinity to zero. This equation can be finally written as:
$$\dot{\widehat{\theta }}=\frac{3}{\widehat{a}^2}\frac{9\gamma M}{2\widehat{a}^{3\gamma }}\frac{3}{2}\frac{\alpha N}{\widehat{a}^6}.$$
(37)
From it one sees that all terms in the right-hand side induce contraction and hence a spacetime singularity is expected to occur (the global singularity at $`\widehat{a}=0`$).
The evolution of the volume expansion in the conformal frame, given in terms of the untransformed scale factor, can be easily found from eq.(4.5) if we realize that $`\theta =e^{\frac{\widehat{\varphi }}{2}}(\widehat{\theta }\frac{3}{2}\widehat{\varphi })`$. We find that
$$(\frac{d\theta }{d\tau })^\pm =3\frac{e^{\widehat{\varphi }^\pm }}{\widehat{a}^6}\{\widehat{a}^4\frac{3}{2}\gamma M\widehat{a}^{3(2\gamma )}\frac{1}{2}(\alpha +\frac{1}{2})N\pm \frac{\sqrt{N}}{2}\sqrt{\widehat{a}^4+M\widehat{a}^{3(2\gamma )}+\frac{1}{6}\alpha N}\},$$
(38)
where $`\tau `$ is the proper time in the conformal frame. It is related with $`t`$ through $`d\tau =e^{\frac{1}{2}\widehat{\varphi }^\pm }dt`$. The ’+’ and ’-’ signs in eq.(4.6) correspond to two possible branches of the solution in conformal general relativity. From (4.6) one sees that for the ’+’ branch of the solution, the last term in brackets induces expansion.
We are interested now in the limiting case $`\widehat{a}1`$, since the singularity in the untransformed frame is found at $`\widehat{a}=0`$. In this case for $`\alpha =0`$ ($`\omega =\frac{3}{2}`$) eq.(4.6) can be written as:
$$(\frac{d\theta }{d\tau })^\pm \pm \frac{3\sqrt{NM}e^{\widehat{\varphi }_0}}{2\widehat{a}^{\frac{3}{2}(\gamma +2)}}\mathrm{exp}[\frac{2}{3}\sqrt{\frac{N}{M}}\frac{\widehat{a}^{\frac{3}{2}(2\gamma )}}{(2\gamma )}],$$
(39)
for $`\gamma >\frac{2}{3}`$. $`\widehat{\varphi }_0`$ is some integration constant. For $`\gamma =\frac{2}{3}`$ we obtain:
$$(\frac{d\theta }{d\tau })^\pm \pm \frac{3\sqrt{N(M+1)}e^{\widehat{\varphi }_0}}{2\widehat{a}^4}\mathrm{exp}[\frac{1}{2}\sqrt{\frac{N}{M+1}}\widehat{a}^2],$$
(40)
while for $`\gamma <\frac{2}{3}`$;
$$(\frac{d\theta }{d\tau })^\pm \pm \frac{3\sqrt{N}e^{\widehat{\varphi }_0}}{2\widehat{a}^4}\mathrm{exp}[\frac{\sqrt{N}}{2}\widehat{a}^2].$$
(41)
For arbitrary positive $`\alpha `$ ($`\omega >\frac{3}{2}`$), in the limit $`\widehat{a}1`$, eq.(4.6) can be written in the following way:
$$(\frac{d\theta }{d\tau })^\pm \frac{Ne^{\widehat{\varphi }_0}}{2\widehat{a}^{6\sqrt{\frac{6}{\alpha }}}}(3\alpha \pm 2\sqrt{6\alpha }\frac{3}{2}),$$
(42)
for $`0<\gamma <2`$.
A careful analysis of eq.(4.6) shows that, for big $`\widehat{a}`$ the first three terms in brackets prevail over the last one and, consequently, contraction is favored until $`\widehat{a}`$ becomes sufficiently small ($`\widehat{a}1`$). In this case, when $`\alpha `$ is in the range $`0\alpha \frac{1}{6}`$ ($`\frac{3}{2}\omega \frac{4}{3}`$), in the ’+’ branch of the solution, there are not enough conditions for further contraction and the formation of the global singularity is not allowed. A cosmological wormhole is obtained instead. For further analysis of what happen in this case, we need to write down the relevant magnitudes and relationships, in the conformal frame, in terms of the untransformed scale factor. We shall interested in the behaviour of these magnitudes and relationships for small $`\widehat{a}1`$ (when the condition for further contraction ceases to hold), in the ’+’ branch of the solution. In this case the scale factor is found to be
$$a^+e^{\frac{1}{2}\widehat{\varphi }_0}\widehat{a}^{1\frac{1}{2}\sqrt{\frac{6}{\alpha }}}.$$
(43)
The conformal Ricci scalar is
$$R^+\frac{3}{2}Ne^{\widehat{\varphi }_0}\widehat{a}^{\sqrt{\frac{6}{\alpha }}6},$$
(44)
while for the proper time $`\tau `$ we have that
$$\tau ^+\frac{2e^{\frac{1}{2}\widehat{\varphi }_0}\widehat{a}^{3\frac{1}{2}\sqrt{\frac{6}{\alpha }}}}{\sqrt{\frac{\alpha N}{6}}(6\sqrt{\frac{6}{\alpha }})},$$
(45)
for $`\alpha \frac{1}{6}`$ and
$$\tau ^+\frac{e^{\frac{1}{2}\widehat{\varphi }_0}}{\sqrt{\frac{\alpha N}{6}}}\mathrm{ln}\widehat{a},$$
(46)
for $`\alpha =\frac{1}{6}`$. Hence, if we choose the ’+’ branch of the solution, for $`\alpha \frac{1}{6}`$, $`R^+`$ is bounded for $`\widehat{a}0`$. In this limit $`a^++\mathrm{}`$ while $`\tau ^+\mathrm{}`$. A similar analysis shows that, for big $`\widehat{a}`$, $`\widehat{a}\mathrm{}a^\pm \mathrm{}`$ and $`\tau ^\pm +\mathrm{}`$. For intermediate values in the range $`0<\widehat{a}<\mathrm{}`$ the curvature scalar $`R^+`$ is well behaved and bounded. The scale factor $`a`$ is a minimum at some intermediate time $`\tau _{}`$. Hence in the conformal representation of general relativity, if we choose the ’+’ branch of the solution, the following picture takes place. If we restrict $`\omega `$ to fit into the range $`\frac{3}{2}\omega \frac{4}{3}`$ and for $`0<\gamma <2`$, then, the universe evolves from the infinite past $`\tau =\mathrm{}`$ when he had an infinite size, through a bounce at some intermediate $`\tau _{}`$ when he reached a minimum size $`a_{}`$, into the infinite future $`\tau =+\mathrm{}`$ when he will reach again an infinite size. As illustrations to this behaviour we shall study the particular cases with $`\omega =\frac{3}{2}`$ ($`\alpha =0`$) for dust-filled and radiation-filled universes since, in these very particular situations exact analytic solutions can be easily found.
For a radiation-filled universe ($`\gamma =\frac{4}{3}`$) the equation (4.1) with $`\alpha =0`$ can be written as:
$$\dot{\widehat{a}}=\sqrt{M\widehat{a}^2+1},$$
(47)
and, after integration we obtain for the untransformed scale factor
$$\widehat{a}=\sqrt{t^2M}.$$
(48)
The proper time $`t`$ is constrained to the range $`|t|\sqrt{M}`$ or $`\sqrt{M}t+\mathrm{}`$ (the case $`\mathrm{}t\sqrt{M}`$ corresponds to the time reversed solution). The scale factor, in the conformal frame, is then found to be
$$a^\pm =\frac{\sqrt{t^2M}}{\sqrt{\varphi _0}}\mathrm{exp}[\pm \frac{1}{2}\frac{\sqrt{N}}{M}\frac{t}{\sqrt{t^2M}}],$$
(49)
while the curvature scalar:
$$R^\pm =\frac{3}{2}N\varphi _0\frac{\mathrm{exp}[\pm \frac{\sqrt{N}}{M}\frac{t}{\sqrt{t^2M}}]}{(t^2M)^3}.$$
(50)
The relationship between the proper time $`t`$ measured in the untransformed frame and its conformal (for the ’+’ branch of the solution that is the case of interest) is given by
$$\tau ^+=\frac{t}{\sqrt{M\varphi _0}}\mathrm{exp}[\frac{\sqrt{N}}{2M}\frac{t}{\sqrt{t^2M}}].$$
(51)
A careful analysis of eq.(4.17) shows that $`a^+`$ is a minimum at some time that is a root of the algebraic equation $`t^4Mt^2\frac{N}{4}=0`$. The curvature singularity occurring in the Einstein’s formulation of general relativity at $`t=\sqrt{M}`$, is removed in the conformal frame, where $`R^+`$ is bounded and well behaved for all times in the range $`\sqrt{M}t+\mathrm{}`$ ($`\mathrm{}\tau +\mathrm{}`$).
For a dust-filled universe ($`\gamma =1`$) the untransformed scale factor can be given the form:
$$\widehat{a}=\frac{4M}{\eta ^24},$$
(52)
where the time variable $`\eta `$ has been introduced through $`dt=\frac{\widehat{a}^2}{M}d\eta `$ and is constrained to the range $`2\eta +\mathrm{}`$ (the case $`\mathrm{}\eta 2`$ corresponds to the time reversed solution). In the conformal formulation of GR we have that
$$a^\pm (\eta )=\frac{4M}{\sqrt{\varphi _0}}\frac{\mathrm{exp}[\frac{\sqrt{N}}{24M^2}\eta (\eta ^212)]}{\eta ^24},$$
(53)
and the relationship between the proper time $`\tau `$ and $`\eta `$ is given by the following expression:
$$\tau ^\pm =\frac{16M}{\sqrt{\varphi _0}}𝑑\eta \frac{\mathrm{exp}[\frac{\sqrt{N}}{24M^2}\eta (\eta ^212)]}{(\eta ^24)^2}.$$
(54)
The curvature singularity occurring in the canonical Einstein’s GR at time $`\eta =2`$ is removed again in the conformal representation of the theory. The ’+’ branch scale factor $`a^+`$ is a minimum at some $`\eta _{}`$ that is a root of the algebraic equation $`\eta ^48\eta ^2+\frac{16M^2}{\sqrt{N}}\eta +16=0`$.
Finally we shall remark the fact that the cosmological singularity is removed, in conformal GR, only for a given range of the parameter $`\alpha `$ (or $`\omega `$). It can be taken just as a restriction on the values this parameter can take. A physical consideration why we chose the ’+’ branch (the non-singular branch) instead of the ’-’ branch is based on the following analysis. We shall note that in the conformal formulation of GR, $`e^{\widehat{\varphi }}`$ plays the role of an effective gravitational constant $`G`$. For the ’-’ branch $`G`$ runs from zero to an infinite value, i.e. gravity becomes stronger as the universe evolves and, in the infinite future it dominates over the other interactions, that is in contradiction with the usual picture. On the contrary, for the ’+’ branch, $`G`$ runs from an infinite value to zero and hence gravitational effects are weakened as the universe evolves, as required.
## V Discussion
In this section, we would like to present some reflections originated by the results we have obtained in this and in the former papers (reference ). These results tell us that, flat and open universes, are free of the cosmological singularity in the conformal formulation of general relativity, in the given region of the parameter space ($`\frac{3}{2}\omega \frac{4}{3}`$, $`0<\gamma <2`$). Both canonical formulation with the cosmological singularity and the conformal one without them, are just two different representations of the same theory: general relativity. Respecting experimental observations none of these pictures is preferred over the other. However a question is to be raised. Is the cosmological singularity a spurious object (an artifact) of general relativity due to a wrong choice of the representation of this theory?. While trying to answer this fundamental question we should be very careful since experimental observations can not help us. In fact, the untransformed FRW universe with the cosmological singularity and its conformal, FRW, wormhole universe, are related through the conformal transformation (1.1) with $`\mathrm{\Omega }^2=\varphi `$, and experimental measurements are insensible to this transformation.
In section II we have shown that matter particles non-minimally coupled to the metric $`𝐠`$, follow the geodesics of Weyl geometry. The metric $`𝐠`$, together with the scalar field $`\varphi `$, define metrical relations on Weyl spacetimes. Hence, if we follow the line of reasoning, leading to the identification of the metric $`\widehat{𝐠}`$ as the physical metric in canonical GR, we then reach to the conclusion that, $`𝐠`$ is the physical metric in the conformal representation of general relativity. On the other hand, in section III, we showed that conformal general relativity, linked with Weyl geometry, provides a consistent formulation of the laws of gravity. In fact, the effective Lagrangian of this theory (eq.(2.3) or (3.14)), is invariant under the one-parameter group of transformations of the units of length, time and mass studied in that section. Unlike this, canonical general relativity, is not such a consistent theory of spacetime. This means that the singularity-free character of FRW (flat and open) spacetimes, in conformal GR, is relevant enough: for flat and open FRW spacetimes, the answer to the question raised at the beginning of this section, should be positive. The cosmological singularity is an artifact of general relativity, due to the wrong choice of the formulation of this theory.
As it is understood at present, the canonical formulation of general relativity is a classical theory of spacetime. Hence this formulation of GR can not take account of the global singularity at the beginning of time. It is expected that, when a final quantum gravity theory will be worked out, the cosmological singularity will be removed. Unlike this, as we have shown for flat () and open universes (present paper), the conformal formulation of general relativity can take account of the full (singularity-free) evolution of the universe in spacetime without resorting to quantum considerations. It is true, however, if the throat radius of the cosmological wormhole, in the frame of conformal GR, is much greater than the Planck length. Since the throat radius depends on the integration constants $`N`$, $`M`$, and $`\widehat{\varphi }_0`$ this means that appropriate initial conditions should be given.
ACKNOWLEDGMENT
We acknowledge many colleagues for their interest in the ideas developed in this paper. Their criticism, together with the referee’s criticism, contributed to the improvement of the present version of the paper. We, also, thank MES of Cuba by financial support.
|
warning/0002/hep-ph0002252.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
The SAMPLE collaboration at MIT-Bates has recently reported a value for the strange-quark magnetic form factor measured using backward angle parity-violating (PV) electron-proton scattering :
$$G_M^{(s)}(Q^2=0.1\text{GeV}^2/c^2)=0.61\pm 0.27\pm 0.19,$$
(1)
where the first error is experimental and the second is theoretical. The dominant contribution to the theoretical error is uncertainty associated with radiative corrections to the axial vector term in the backward angle left-right asymmetry $`A_{LR}`$:
$$A_{LR}Q_W^P+Q_W^N\frac{G_M^n}{G_M^p}+Q_W^{(0)}\frac{G_M^{(s)}}{G_M^p}(14\mathrm{sin}^2\theta _W)\sqrt{1+1/\tau }\frac{G_A^p}{G_M^p},$$
(2)
where $`Q_W^P`$ and $`Q_W^N`$ are the proton and neutron weak charges, respectively, $`Q_W^{(0)}`$ is the SU(3)-singlet weak charge<sup>*</sup><sup>*</sup>*Note that in Ref. , the weak charges are denoted $`\xi _V^{p,n,(0)}`$ ., $`\theta _W`$ is the weak mixing angle, and $`\tau =Q^2/4M_n^2`$. The axial form factor is normalized at the photon point as
$$G_A^p(0)=g_A[1+R_A^p]$$
(3)
where $`g_A=1.267\pm 0.004`$ is the nucleon’s axial charge as measured in neutron $`\beta `$-decay and $`R_A^p`$ denotes process-dependent electroweak radiative corrections to the $`V(e)\times A(p)`$ scattering amplitude.
The radiative correction $`R_A^p`$ is the subject of the present study. It was first analyzed in Ref. and found to be large, negative in sign, and plagued by considerable theoretical uncertainty. Generally, $`R_A^p`$ contains two classes of contributions. The first involve electroweak radiative corrections to the elementary $`V(e)\times A(q)`$ amplitudes, where $`q`$ is any one of the quarks in the proton. These radiative corrections, referred to henceforth as “one-quark” radiative corrections, are calculable in the Standard Model. They contain little theoretical uncertainty apart from the gentle variation with Higgs mass and long-distance QCD effects involving light-quark loops in the $`Z\gamma `$ mixing tensor. The one-quark contributions can be large, due to the absence from loops of the small $`(14\mathrm{sin}^2\theta _W)`$ factor appearing at tree level (see Eq. (2) ) and the presence of large logarithms of the type $`\mathrm{ln}(m_q/M_Z)`$.
A second class of radiative corrections, which we refer to as “many-quark” corrections, involve weak interactions among quarks in the proton. In this paper, we focus on those many-quark corrections which generate an axial vector coupling of the photon to the proton (see Figure 1). This axial vector $`pp\gamma `$ interaction, also known as the anapole moment (AM), has the form
$$^{AM}=\frac{e}{\mathrm{\Lambda }_\chi ^2}\overline{N}(a_s+a_v\tau _3)\gamma _\mu \gamma _5N_\nu F^{\nu \mu }.$$
(4)
(Here, we have elected to normalize the interaction to the scale of chiral symmetry breaking, $`\mathrm{\Lambda }_\chi =4\pi F_\pi `$.) These many-quark anapole contributions to $`R_A^p`$, which are independent of the electroweak gauge parameter, were first studied in Ref. and found in Ref. to carry significant theoretical uncertainty. The scale of this uncertainty was estimated in Ref. , and this value was used to obtain the theoretical error in Eq. (1). (Note that the central value for $`G_M^{(s)}`$ given in Eq. (1) is obtained from the experimental asymmetry using the calculation of Ref. ).
In order to better constrain the error in $`G_M^{(s)}`$ associated with $`R_A^p`$, the SAMPLE collaboration performed a second backward angle PV measurement using quasielastic (QE) scattering from the deuteron. The asymmetry $`A_{LR}(\text{QE})`$ is significantly less sensitive to $`G_M^{(s)}`$ than is $`A_{LR}(ep)`$, but retains a strong dependence on $`R_A^{T=1}`$, the isovector part of $`R_A^p`$. The calculation of Ref. found the uncertainty in $`R_A^p`$ to be dominated by this isovector component—$`R_A^{T=1}0.34\pm 0.20`$—and the goal of the deuterium measurement was, therefore, to constrain the size of this largest term. A preliminary deuterium result was reported at the recent Bates25 Symposium at MIT, and suggests that $`R_A^{T=1}`$ has the same negative sign as computed in Ref. but has considerably larger magnitude, possibly of order unity. Combining this result with the previous $`A_{LR}(ep)`$ measurement would yield a nearly vanishing value for $`G_M^{(s)}`$, rather than the large and positive value quoted in Eq. (1).
The prospective SAMPLE result for $`R_A^{T=1}`$ is remarkable, indicating that a higher-order electroweak radiative correction is of the same magnitude as, and cancels against, the tree-level amplitude! The occurance of such enhanced electroweak radiative corrections is rare. Nevertheless, there does exist at least one other instance in which higher-order electroweak processes can dominate the axial vector hadronic response, namely, the nuclear anapole moment. The anapole moment of a heavy nucleus grows as $`A^{2/3}`$ (see, e.g. Refs. and references therein). Because of the scaling with mass number, the nuclear AM contribution to a $`V(e)\times A(\text{nucleus})`$ amplitude can be considerably larger than the corresponding tree-level $`Z^0`$-exchange amplitude, and this $`A^{2/3}`$ enhancement is consistent with the size of the Cesium AM recently determined by the Boulder group using atomic parity-violation. The reason behind the enhancement of $`R_A^{T=1}`$ for the few-nucleon system, however, is not understood. The goal of the present paper is to investigate whether there exist c onventional, hadronic physics effects which can explain the enhancement apparently implied by the SAMPLE deuterium measurement.
In order to address this question, we revisit the analysis of Ref. . Following Ref. , we re-cast that analysis into the framework of heavy baryon chiral perturbation theory (HBChPT) . We carry out a complete calculation of $`R_A^{T=1}`$ and $`R_A^{T=0}`$ to order $`1/\mathrm{\Lambda }_\chi ^2`$, including loop diagrams not considered in Refs. . We also extend those analyses to include decuplet as well as octet intermediate states, magnetic insertions, and SU(3) chiral symmetry. As in Ref. , we estimate the chiral counterterms at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ using vector meson saturation. However, we go beyond that previous analysis and determine the sign of this vector meson contribution phenomenologically. We find that decuplet intermediate states and magnetic insertions do not contribute up to the chiral order at which we truncate. Also, the effect of SU(3) symmetry, in the guise of kaon loops, is generally smaller than the pion loops considered previously. In the end, we express our results in terms of effective PV hadronic couplings. Some of these couplings may be determined from nuclear and hadronic PV experiments or detailed calculations (for reviews, see Refs. ), while others are presently unconstrained by measurement. Guided by phenomenology and the dimensional analysis of Ref. , we estimate the range of possible values for the new couplings. We suspect that our estimates are overly generous. Nevertheless, we find that – even under liberal assumptions – the AM contributions to $`R_A^{T=1}`$ appear unable to enhance the one-quark corrections to the level apparently observed by the SAMPLE collaboration and, in our conclusions, we speculate on possible additional sources of enhancement not considered here.
The remainder of the paper is organized as follows. In Section 2, we relate the anapole couplings $`a_{s,v}`$ to the radiative corrections, $`R_A^{T=0,1}`$, and in Section 3, we outline our formalism for computing these couplings in HBChPT. A reader already familiar with this formalism may wish to skip to Section 4, where we compute the chiral loop contributions to the nucleon anapole moment through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. We also include the leading $`1/m_N`$ terms in the heavy baryon expansion, which generate contributions of $`𝒪(1/\mathrm{\Lambda }_\chi m_N)`$. Section 5 contains the vector meson estimate of the chiral counterterms and the determination of the sign, while Section 6 gives our numerical estimate of the AM contributions to $`R_A^{T=0,1}`$. We briefly discuss the phenomenology of hadronic and nuclear PV and what that phenomenology may imply about the scale of the unknown low-energy constants. Section 7 summarizes our conclusions. The Appendices give a detailed discussion of (A) our formalism, (B) the full set of hadronic PV Lagrangians allowed under SU(3) symmetry, and (C) graphs, nominally present at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ but whose contributions vanish.
## II Anapole contributions to $`R_A`$
The electron-nucleon parity violating amplitude is generated by the diagrams in Figure 2. At tree level this amplitude reads
$$iM^{PV}=iM_{AV}^{PV}+iM_{VA}^{PV},$$
(5)
where
$$iM_{AV}^{PV}=i\frac{G_\mu }{2\sqrt{2}}l^{\lambda 5}<N|J_\lambda |N>$$
(6)
and
$`iM_{VA}^{PV}`$ $`=`$ $`i{\displaystyle \frac{G_\mu }{2\sqrt{2}}}l^\lambda <N|J_{\lambda 5}|N>`$ (7)
$`=`$ $`i{\displaystyle \frac{14\mathrm{sin}^2\theta _W}{2\sqrt{2}}}g_AG_\mu \overline{e}\gamma ^\lambda e\overline{N}\tau _3\gamma _\lambda \gamma _5N.`$ (8)
at tree-level in the Standard Model (Figure 2a). Here, $`J_\lambda `$ ($`J_{\lambda 5}`$) and $`l_\lambda `$ ($`l_{\lambda 5}`$) denote the vector (axial vector) weak neutral currents of the quarks and electron, respectively . The anapole moment interaction of Eq. (4) generates additional contributions to $`M_{VA}^{PV}`$ when a photon is exchanged between the nucleon and the electron (Figure 2b). The corresponding amplitude is
$$iM_{AM}^{PV}=i\frac{(4\pi \alpha )}{\mathrm{\Lambda }_\chi ^2}\overline{e}\gamma ^\lambda e\overline{N}(a_s+a_v\tau _3)\gamma _\lambda \gamma _5N.$$
(9)
Note that unlike $`iM_{VA}^{PV}`$, $`iM_{AM}^{PV}`$ contains no $`(14\mathrm{sin}^2\theta _W)`$ suppression. Consequently, the relative importance of the anapole interaction is enhanced by $`1/(14\mathrm{sin}^2\theta _W)10`$. This enhancement may be seen explicitly by converting Eqs. (6) and (9) into $`R_A^{T=0,1}`$:
$$R_A^{T=0}|_{\text{anapole}}=\frac{8\sqrt{2}\pi \alpha }{G_\mu \mathrm{\Lambda }_\chi ^2}\frac{1}{14\mathrm{sin}^2\theta _W}\frac{a_s}{g_A}$$
(10)
$$R_A^{T=1}|_{\text{anapole}}=\frac{8\sqrt{2}\pi \alpha }{G_\mu \mathrm{\Lambda }_\chi ^2}\frac{1}{14\mathrm{sin}^2\theta _W}\frac{a_v}{g_A}$$
(11)
The constants $`a_{s,v}`$ contain contributions from loops generated by the Lagrangians given in Section 3 and from counterterms in the tree-level effective Lagrangian of Eq. (4):
$$a_{s,v}=a_{s,v}^L+a_{s,v}^{CT}.$$
(12)
In HBChPT, only the parts of the loop amplitudes non-analytic in quark masses can be unambigously indentified with $`a_{s,v}^L`$. The remaining analytic terms are included in $`a_{s,v}^{CT}`$. In what follows, we compute explicityly the various loop contributions up through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$, while in principle, $`a_{s,v}^{CT}`$ should be determined from experiment. In Section 5, however, we discuss a model estimate for $`a_{s,v}^{CT}`$.
Before proceeding with details of the calculation, it is useful to take note of the scales present in Eqs. (10). The constants $`a_{s,v}`$ are generally proportional to a product of strong and weak meson-baryon couplings. The former are generally of order unity, while the size of weak, PV couplings can be expressed in terms of $`g_\pi =3.8\times 10^8`$, the scale of charged current contributions. One then expects the AM contributions to the axial radiative corrections to be of order
$$R_A^{T=0,1}\frac{8\sqrt{2}\pi \alpha }{G_\mu \mathrm{\Lambda }_\chi ^2}\frac{1}{14\mathrm{sin}^2\theta _W}\frac{g_\pi }{g_A}0.01.$$
(13)
In some cases, the PV hadronic couplings may be an order of magnitude larger than $`g_\pi `$. Alternatively, chiral singularities arising from loops may also enhance the AM effects over the scale in Eq. (13). Thus, as we show below, the net effect of the AM is anticipated to be a 10-20 % contribution to $`R_A^{T=0,1}`$.
## III Notations and Conventions
Since much of the formalism for HBChPT is standard, we relegate a detailed summary of our conventions to Appendix A. However, some discussion of the effective Lagrangians used in computing chiral loop contributions to $`a_{s,v}`$ is necessary here. Specifically, we require the parity-conserving (PC) and parity-violating (PV) Lagrangians involving pseudoscalar meson, spin-$`1/2`$ and spin-$`3/2`$ baryon, and photon fields. For the moment, we restrict ourselves to SU(2) flavor symmetry and generalize to SU(3) later. The relativistic PC Lagrangian for $`\pi `$, $`N`$, $`\mathrm{\Delta }`$, and $`\gamma `$ interactions needed here is
$`^{PC}`$ $`=`$ $`{\displaystyle \frac{F_\pi ^2}{4}}TrD^\mu \mathrm{\Sigma }D_\mu \mathrm{\Sigma }^{}+\overline{N}(i𝒟_\mu \gamma ^\mu m_N)N+g_A\overline{N}A_\mu \gamma ^\mu \gamma _5N`$ (19)
$`+{\displaystyle \frac{e}{\mathrm{\Lambda }_\chi }}\overline{N}(c_s+c_v\tau _3)\sigma ^{\mu \nu }F_{\mu \nu }^+N`$
$`T_i^\mu [(i𝒟_\alpha ^{ij}\gamma ^\alpha m_\mathrm{\Delta }\delta ^{ij})g_{\mu \nu }{\displaystyle \frac{1}{4}}\gamma _\mu \gamma ^\lambda (i𝒟_\alpha ^{ij}\gamma ^\alpha m_\mathrm{\Delta }\delta ^{ij})\gamma _\lambda \gamma ^\nu `$
$`+{\displaystyle \frac{g_1}{2}}g_{\mu \nu }A_\alpha ^{ij}\gamma ^\alpha \gamma _5+{\displaystyle \frac{g_2}{2}}(\gamma _\mu A_\nu ^{ij}+A_\mu ^{ij}\gamma _\nu )\gamma _5+{\displaystyle \frac{g_3}{2}}\gamma _\mu A_\alpha ^{ij}\gamma ^\alpha \gamma _5\gamma _\nu ]T_j^\nu `$
$`+g_{\pi N\mathrm{\Delta }}[\overline{T}_i^\mu (g_{\mu \nu }+z_0\gamma _\mu \gamma _\nu )\omega _i^\nu N+\overline{N}\omega _i^\nu (g_{\mu \nu }+z_0\gamma _\nu \gamma _\mu )T_i^\mu ]`$
$`ie{\displaystyle \frac{c_\mathrm{\Delta }q_i}{\mathrm{\Lambda }_\chi }}\overline{T}_i^\mu F_{\mu \nu }^+T_i^\nu +[{\displaystyle \frac{ie}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu (d_s+d_v\tau _3)\gamma ^\nu \gamma _5F_{\mu \nu }^+N+h.c.]`$
where $`𝒟_\mu `$ is a chiral and electromagnetic (EM) covariant derivative, $`\mathrm{\Sigma }=\mathrm{exp}(i\stackrel{}{\tau }\stackrel{}{\pi }/F_\pi )`$ is the conventional non-linear representation of the pseudoscalar field, $`N`$ is a nucleon isodoublet field, $`T_\mu ^i`$ is the $`\mathrm{\Delta }`$ field in the isospurion formalism, $`F^{\mu \nu }`$ is the photon field strength tensor, and $`A_\mu `$ is the axial field involving the pseudoscalars
$$A_\mu =\frac{D_\mu \pi }{F_\pi }+𝒪(\pi ^3)$$
(20)
with $`D_\mu `$ being the EM covariant derivative. Explicit expressions for the fields and the transformation properties can be found in Appendix A. The constants $`c_s,c_v`$ determined in terms of the nucleon isoscalar and isovector magnetic moments, $`c_\mathrm{\Delta }`$ is the $`\mathrm{\Delta }`$ magnetic moment, $`d_s,d_v`$ are the nucleon and delta transition magnetic moments, and $`z_0`$ is the off-shell parameter which is not relevant in the present work . Our convention for $`\gamma _5`$ is that of Bjorken and Drell .
In order to obtain proper chiral counting for the nucleon, we employ the conventional heavy baryon expansion of $`^{PC}`$, and in order to cosistently include the $`\mathrm{\Delta }`$ we follow the small scale expansion proposed in . In this approach energy-momenta and the delta and nucleon mass difference $`\delta `$ are both treated as $`𝒪(ϵ)`$ in chiral power counting. The leading order vertices in this framework can be obtained via $`P_+\mathrm{\Gamma }P_+`$ where $`\mathrm{\Gamma }`$ is the original vertex in the relativistic Lagrangian and
$$P_\pm =\frac{1\pm \overline{)}v}{2}.$$
(21)
are projection operators for the large, small components of the Dirac wavefunction respectively. Likewise, the $`O(1/m_N)`$ corrections are generally propotional to $`P_+\mathrm{\Gamma }P_{}/m_N`$. In previous work the parity conserving $`\pi N\mathrm{\Delta }\gamma `$ interaction Lagrangians have been obtained to $`O(1/m_N^2)`$. We collect some of the relevant terms below:
$`_v^{PC}`$ $`=`$ $`\overline{N}[ivD+2g_ASA]Ni\overline{T}_i^\mu [ivD^{ij}\delta ^{ij}\delta +g_1SA^{ij}]T_\mu ^j`$ (25)
$`+g_{\pi N\mathrm{\Delta }}[\overline{T}_i^\mu \omega _\mu ^iN+\overline{N}\omega _\mu ^iT_i^\mu ]`$
$`+{\displaystyle \frac{1}{2m_N}}\overline{N}\{(vD)^2D^2+[S_\mu ,S_\nu ][D^\mu ,D^\nu ]`$
$`ig_A(SDvA+vASD)\}N+\mathrm{}`$
where $`S_\mu `$ is the Pauli-Lubanski spin operator and $`\delta m_\mathrm{\Delta }m_N`$.
The PV analog of Eq. (19) can be constructed using the chiral fields $`X_{L,R}^a`$ defined in Appendix A and the spacetime transformation properties of the various fields in Eq. (19). We find it convenient to follow the convention in Ref. and separate the PV Lagrangian into its various isospin components. The hadronic weak interaction has the form
$$_W=\frac{G_\mu }{\sqrt{2}}J_\lambda J^\lambda +\text{h.c.},$$
(26)
where $`J_\lambda `$ denotes either a charged or neutral weak current built out of quarks. In the Standard Model, the strangeness conserving charged currents are pure isovector, whereas the neutral currents contain both isovector and isoscalar components. Consequently, $`_W`$ contains $`\mathrm{\Delta }T=0,1,2`$ pieces and these channels must all be accounted for in any realistic hadronic effective theory.
Again for simplicity, we restrict our attention first to the light quark SU(2) sector. (A general SU(3) PV meson-baryon Lagrangian is given in the Appendix and is considerably more complex.) We quote the relativistic Lagrangians, but employ the heavy baryon projections, as described above, in computing loops. It is straightforward to obtain the corresponding heavy baryon Lagrangians from those listed below, so we do not list the PV heavy baryon terms below. For the $`\pi N`$ sector we have
$`_{\mathrm{\Delta }T=0}^{\pi N}`$ $`=`$ $`h_V^0\overline{N}A_\mu \gamma ^\mu N`$ (27)
$`_{\mathrm{\Delta }T=1}^{\pi N}`$ $`=`$ $`{\displaystyle \frac{h_V^1}{2}}\overline{N}\gamma ^\mu NTr(A_\mu X_+^3){\displaystyle \frac{h_A^1}{2}}\overline{N}\gamma ^\mu \gamma _5NTr(A_\mu X_{}^3)`$ (30)
$`{\displaystyle \frac{h_\pi }{2\sqrt{2}}}F_\pi \overline{N}X_{}^3N`$
$`_{\mathrm{\Delta }T=2}^{\pi N}`$ $`=`$ $`h_V^2^{ab}\overline{N}[X_R^aA_\mu X_R^b+X_L^aA_\mu X_L^b]\gamma ^\mu N`$ (33)
$`{\displaystyle \frac{h_A^2}{2}}^{ab}\overline{N}[X_R^aA_\mu X_R^bX_L^aA_\mu X_L^b]\gamma ^\mu \gamma _5N.`$
The above Lagrangian was first given by Kaplan and Savage (KS). However, the coefficients used in our work are slightly different from those of Ref. since our definition of $`A_\mu `$ differs by an overall phase (see Appendix A). Moreover, the coefficient of the second term in the original PV $`\mathrm{\Delta }T=2`$ $`NN\pi \pi `$ Lagrangian in Eq. (2.18) was misprinted in the work of KS, and should be $`2h_A^2`$ in their notation instead of $`h_V^2`$ as given in Eq. (2.18) of .
The term proportional to $`h_\pi `$ contains no derivatives and, at leading-order in $`1/F_\pi `$, yields the PV $`NN\pi `$ Yukawa coupling traditionally used in meson-exchange models for the PV NN interaction . The PV $`\gamma `$-decay of <sup>18</sup>F can be used to constrain the value of $`h_\pi `$ in a nuclear model-independent way as discussed in Ref. , resulting in $`h_\pi =(0.7\pm 2.2)g_\pi `$ . Future PV experiments are planned using light nuclei to confirm the <sup>18</sup>F result. The coupling $`h_\pi `$ has also received considerable theoretical attention and is particularly interesting since it receives no charged current contributions at leading order.
Unlike the PV Yukawa interaction, the vector and axial vector terms in Eqs. (27-33) contain derivative interactions. The terms containing $`h_A^1,h_A^2`$ start off with $`NN\pi \pi `$ interactions, while all the other terms start off as $`NN\pi `$. Such derivative interactions have not been included in conventional analyses of nuclear and hadronic PV experiments. Consequently, the experimental constraints on the low-energy constants $`h_V^i`$, $`h_A^i`$ are unknown. The authors of Ref. used simple dimensional arguments and factorization limits to estimate their values, and we present additional phenomenological considerations in Section 6 below. We emphasize, however, that the present lack of knowledge of these couplings introduces additional uncertainties into $`R_A^{T=0,1}`$.
In addition to purely hadronic PV interactions, one may also write down PV EM interactions involving baryons and mesonsNote that the hadronic derivative interactions of Eqs. (27-33) also contain $`\gamma `$ fields as required by gauge-invariance. The anapole interaction of Eq. (4) represents one such interaction, arising at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ and involving no $`\pi `$’s. There also exist terms at $`𝒪(1/\mathrm{\Lambda }_\chi )`$ which include at least one $`\pi `$:
$$^{\gamma N}PV=\frac{c_1}{\mathrm{\Lambda }_\chi }\overline{N}\sigma ^{\mu \nu }[F_{\mu \nu }^+,X_{}^3]_+N+\frac{c_2}{\mathrm{\Lambda }_\chi }\overline{N}\sigma ^{\mu \nu }F_{\mu \nu }^{}N+\frac{c_3}{\mathrm{\Lambda }_\chi }\overline{N}\sigma ^{\mu \nu }[F_{\mu \nu }^{},X_+^3]_+N.$$
(34)
The corresponding PV Lagrangians involving a $`N\mathrm{\Delta }`$ transition are somewhat more complicated. The analogues of Eqs. (27-33) are
$`_{\mathrm{\Delta }I=0}^{\pi \mathrm{\Delta }N}`$ $`=`$ $`f_1ϵ^{abc}\overline{N}i\gamma _5[X_L^aA_\mu X_L^b+X_R^aA_\mu X_R^b]T_c^\mu `$ (36)
$`+g_1\overline{N}[A_\mu ,X_{}^a]_+T_a^\mu +g_2\overline{N}[A_\mu ,X_{}^a]_+T_a^\mu +\text{h.c.}`$
$`_{\mathrm{\Delta }I=1}^{\pi \mathrm{\Delta }N}`$ $`=`$ $`f_2ϵ^{ab3}\overline{N}i\gamma _5[A_\mu ,X_+^a]_+T_b^\mu +f_3ϵ^{ab3}\overline{N}i\gamma _5[A_\mu ,X_+^a]_{}T_b^\mu `$ (41)
$`+g_3\overline{N}[(X_L^aA_\mu X_L^3X_L^3A_\mu X_L^a)(X_R^aA_\mu X_R^3X_R^3A_\mu X_R^a)]T_a^\mu `$
$`+g_4\{\overline{N}[3X_L^3A^\mu (X_L^1T_\mu ^1+X_L^2T_\mu ^2)+3(X_L^1A^\mu X_L^3T_\mu ^1+X_L^2A^\mu X_L^3T_\mu ^2)`$
$`2(X_L^1A^\mu X_L^1+X_L^2A^\mu X_L^22X_L^3A^\mu X_L^3)T_\mu ^3](LR)\}+\text{h.c.}`$
$`_{\mathrm{\Delta }I=2}^{\pi \mathrm{\Delta }N}`$ $`=`$ $`f_4ϵ^{abd}^{cd}\overline{N}i\gamma _5[X_L^aA_\mu X_L^b+X_R^aA_\mu X_R^b]T_c^\mu `$ (45)
$`+f_5ϵ^{ab3}\overline{N}i\gamma _5[X_L^aA_\mu X_L^3+X_L^3A_\mu X_L^a+(LR)]T_b^\mu `$
$`+g_5^{ab}\overline{N}[A_\mu ,X_{}^a]_+T_b^\mu +g_6^{ab}\overline{N}[A_\mu ,X_{}^a]_+T_b^\mu +\text{h.c.},`$
where the terms containing $`f_i`$ and $`g_i`$ start off with single and two pion vertices, respectively.
Finally, we consider PV $`\gamma \mathrm{\Delta }N`$ interactions:
$`_{PV}^{\gamma \mathrm{\Delta }N}`$ $`=`$ $`ie{\displaystyle \frac{d_1}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu F_{\mu \nu }^+N+ie{\displaystyle \frac{d_2}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu [F_{\mu \nu }^+,X_+^3]_+N`$ (49)
$`+ie{\displaystyle \frac{d_3}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu [F_{\mu \nu }^+,X_+^3]_{}N+ie{\displaystyle \frac{d_4}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu \gamma _5F_{\mu \nu }^{}N`$
$`+ie{\displaystyle \frac{d_5}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu \gamma _5[F_{\mu \nu }^+,X_{}^3]_+N+ie{\displaystyle \frac{d_6}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu \gamma _5[F_{\mu \nu }^{},X_+^3]_+N`$
$`+ie{\displaystyle \frac{d_7}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu [F_{\mu \nu }^{},X_{}^3]_+N+ie{\displaystyle \frac{d_8}{\mathrm{\Lambda }_\chi }}\overline{T}_3^\mu \gamma ^\nu [F_{\mu \nu }^{},X_{}^3]_{}N+\text{h.c.}.`$
The PV $`\gamma \mathrm{\Delta }N`$ vertices $`d_{13}`$, $`d_{46}`$ and $`d_{78}`$ are associated at leading order in $`1/F_\pi `$ with zero, one and two pion vertices, respectively. All the vertices in (27)-(49) are $`𝒪(p)`$ or $`𝒪(1/\mathrm{\Lambda }_\chi )`$ except $`h_\pi `$, which is Yukawa interaction and of $`O(p^0)`$. As we discuss in Appendix C, we do not require PV interactions involving two $`\mathrm{\Delta }`$ fields.
## IV Chiral Loops
The contributions to $`a_{s,v}`$ arising from the Lagrangians of Eqs . (27-33) are shown in Figure 3. We regulate the associated integrals using dimensional regularization (DR) and absorb the divergent—$`1/(d4)`$—terms into the counterterms, $`a_{s,v}^{CT}`$. The leading contributions arise from the PV Yukawa coupling $`h_\pi `$ contained in the loops of 3a-f. To $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$, the diagrams 3e,f containing a photon insertion on a nucleon line do not contribute. The reason is readily apparent from examination of the integral associated with the amplitude of Figure 3e:
$`iM_{3e}=ie_Nh_\pi v\epsilon {\displaystyle \frac{\sqrt{2}g_A}{F_\pi }}{\displaystyle \frac{d^Dk}{(2\pi )^D}\frac{i(Sk)}{vk}\frac{i}{v(q+k)}\frac{i}{k^2m_\pi ^2+iϵ}}`$ (50)
$`=ie_Nh_\pi v\epsilon {\displaystyle \frac{2\sqrt{2}g_A}{F_\pi }}S_\mu {\displaystyle _0^{\mathrm{}}}s𝑑s{\displaystyle _0^1}𝑑u{\displaystyle \frac{d^Dk}{(2\pi )^D}\frac{k_\mu }{[k^2+svk+usvq+m_\pi ^2]^3}},`$ (51)
where $`q_\mu `$ is the photon momentum, $`\epsilon `$ is the photon polarization vector, $`s`$ has the dimensions of mass, and we have Wick rotated to Euclidean momenta in the second line. From this form it is clear that $`iM_{3e}Sv=0`$. The sum of the non-vanishing diagrams Figure 3a-d yields a gauge invariant leading order result, which is purely isoscalar:
$$a_s^L(Y1)=\frac{\sqrt{2}}{24}eg_Ah_\pi \frac{\mathrm{\Lambda }_\chi }{m_\pi }.$$
(52)
As the PV Yukawa interaction is of order $`O(p^0)`$, we need to consider higher order corrections involving this interaction, which arise from the $`1/m_N`$ expansion of the nucleon propagator and various vertices. Since $`P_+1P_{}=0`$, there is no $`1/m_N`$ correction to the PV Yukawa vertex. From the $`1/m_N`$ $`\overline{N}N`$ terms in Eq. (19) we have
$$a_s^L(Y2)=\frac{7\sqrt{2}}{48\pi }eg_Ah_\pi \frac{\mathrm{\Lambda }_\chi }{m_N}\mathrm{ln}(\frac{\mu }{m_\pi })^2,$$
(53)
where $`\mu `$ is the subtraction scale introduced by DR. Finally, the $`1/m_N`$ correction to the strong $`\pi NN`$ vertex, contained in the term $`g_A`$ in Eq. (19), yields
$$a_s^L(Y3)=\frac{\sqrt{2}}{48\pi }eg_Ah_\pi \frac{\mathrm{\Lambda }_\chi }{m_N}\mathrm{ln}(\frac{\mu }{m_\pi })^2.$$
(54)
These terms are also isoscalar, and the results in Eqs. (52-54) are fully contained in the previous analyses of Refs. .
For the interactions in Eqs. (27-33) containing $`h_V^i`$, the eight diagrams Figure 3a-h must be considered. Their contribution is purely isovector—
$$a_v^L(V)=\frac{1}{6}eg_A(h_V^0+\frac{4}{3}h_V^2)\mathrm{ln}(\frac{\mu }{m_\pi })^2.$$
(55)
—and was not included in previous analyses.
The contribution generated from the two-pion PV axial vertices in Eqs. (30-33) comes only from the loop Figure 3i and contains both isovector and isoscalar components:
$$a_s^L(A)+a_v^L(A)\tau _3=\frac{1}{3}e(h_A^1+h_A^2\tau _3)\mathrm{ln}(\frac{\mu }{m_\pi })^2.$$
(56)
a result first computed in Ref. .
In principle, a variety of additional contributions will arise at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. For example, insertion of the nucleon magnetic moments (i.e. the terms in Eq. (19) containing $`c_{s,v}`$) into the loops Figure 3e,f—resulting in the loops of Figure 5a,b—would in principle generate terms of $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ when the PV Yukawa interaction is considered. As shown in Appendix C, however, such contributions vanish at this order. Similarly, the entire set of $`\mathrm{\Delta }`$ intermediate state contributions shown in Figure 4, as well as those generated by $`_{PV}^{\gamma N}`$ and $`_{PV}^{\gamma \mathrm{\Delta }N}`$ in Figure 6, vanish up to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. The reasons for the vanishing of these various possible contributions is discussed in Appendix C. Thus, the complete set of SU(2) loop contributions up to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ are given in Eqs. (52-56).
Because $`m_cm_s>>m_sm_{u,d}`$ and $`\mathrm{\Lambda }_\chi >>m_s`$, it may be appropriate to treat the lightest strange and non-strange hadrons on a similar footing and extend the foregoing discussion to SU(3) chiral symmetry. A similar philosophy has been adopted by several authors in studying the axial charges and magnetic moments of the lightest baryons . In what follows, we consider the possibility that kaon loop contributions, introduced by the consideration of SU(3) symmetry, may further enhance the anapole contribution to $`R_A`$.
Before proceeding along these lines, however, one must raise an important caveat. When kaon loop corrections are included in a HBChPT analysis, higher order chiral corrections may go as powers of $`m_K/\mathrm{\Lambda }_\chi 0.5`$. Consequently, the convergence of the SU(3) chiral expansion remains a subject of debate. Fortunately, no such factors appear in the present analysis through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ so that at this order, we find that kaon loop effects in $`R_A`$ are generally tiny compared to those involving pion loops. Whether or not higher-order terms (e.g., those of $`𝒪(1/\mathrm{\Lambda }_\chi ^2\times m_K/\mathrm{\Lambda }_\chi )`$ contribute as strongly as those considered here remains a separate, open question.
To set our notation, we give the leading strong-interaction SU(3) Lagrangian. Since the $`K^0`$ and $`\eta `$ are neutral, loops containing these mesons do not contribute to the AM through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ and we do not include their strong couplings below. For the proton the possible intermediate states are $`\mathrm{\Sigma }^0K^+,\mathrm{\Lambda }K^+`$ while for the neutron only $`\mathrm{\Sigma }^{}K^+`$ can appear. The necessary vertices derive from
$``$ $`=`$ $`2g_A\overline{N}SAN+2g_{N\mathrm{\Lambda }K}[(\overline{N}SK)\mathrm{\Lambda }+\overline{\mathrm{\Lambda }}(SK^{}N)]`$ (58)
$`+2g_{N\mathrm{\Sigma }K}[SK^{}\overline{\mathrm{\Sigma }}N+\overline{N}\mathrm{\Sigma }SK],`$
where $`g_{N\mathrm{\Lambda }K}=[(1+2\alpha )/\sqrt{6}]g_A`$, $`g_{N\mathrm{\Sigma }K}=(12\alpha )g_A`$ with $`g_A=D+F`$, $`\alpha =F/(D+F)`$ and $`D,F`$ are the usual $`SU(3)`$ symmetric and antisymmetric coupling constants.
The general pesudoscalar octet and baryon octet PV Lagrangians are given in the Appendix B. They contain four independent PV Yukawa couplings, 20 axial vector couplings ($`h_A`$-type), and 22 vector couplings ($`h_V`$-type). For simplicity, we combine the SU(3) couplings into combinations specific to various hadrons—e.g. the leading PV Yukawa interactions are
$`_{\text{Yukawa}}^{1\pi }=ih_\pi (\overline{p}n\pi ^+\overline{n}p\pi ^{})ih_{p\mathrm{\Sigma }^0K}(\overline{p}\mathrm{\Sigma }^0K^+\overline{\mathrm{\Sigma }^0}pK^{})`$ (59)
$`ih_{n\mathrm{\Sigma }^{}K}(\overline{n}\mathrm{\Sigma }^{}K^+\overline{\mathrm{\Sigma }}^{}nK^{})ih_{p\mathrm{\Lambda }K}(\overline{p}\mathrm{\Lambda }K^+\overline{\mathrm{\Lambda }}pK^{})+\mathrm{}`$ $`.`$ (60)
In terms of the SU(3) couplings listed in Appendix B, the $`h_{BBM}`$ have the form
$`h_\pi `$ $`=`$ $`2\sqrt{2}(h_1+h_2)`$ (61)
$`h_{p\mathrm{\Sigma }^0K}`$ $`=`$ $`[h_1h_2+\sqrt{3}(h_3h_4)]`$ (62)
$`h_{n\mathrm{\Sigma }^{}K}`$ $`=`$ $`\sqrt{2}h_{p\mathrm{\Sigma }^0K}`$ (63)
$`h_{p\mathrm{\Lambda }K}`$ $`=`$ $`\left[{\displaystyle \frac{h_1}{\sqrt{3}}}+\sqrt{3}h_2+h_3+3h_4\right].`$ (64)
Similarly, we write for the vector PV interaction
$`_V^{1\pi }`$ $`=`$ $`{\displaystyle \frac{h_V^{pn\pi ^+}}{\sqrt{2}F_\pi }}\overline{p}\gamma ^\mu nD_\mu \pi ^+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{\sqrt{2}F_\pi }}\overline{p}\gamma ^\mu \mathrm{\Sigma }^0D_\mu K^+`$ (66)
$`{\displaystyle \frac{h_V^{n\mathrm{\Sigma }^{}K^+}}{\sqrt{2}F_\pi }}\overline{n}\gamma ^\mu \mathrm{\Sigma }^{}D_\mu K^+{\displaystyle \frac{h_V^{p\mathrm{\Lambda }K^+}}{\sqrt{2}F_\pi }}\overline{p}\gamma ^\mu \mathrm{\Lambda }D_\mu K^++h.c.+\mathrm{},`$
and for the axial PV two pion and kaon interactions
$`_A^{2\pi }`$ $`=`$ $`i{\displaystyle \frac{h_A^{p\pi }}{F_\pi ^2}}\overline{p}\gamma ^\mu \gamma _5p(\pi ^+D_\mu \pi ^{}\pi ^{}D_\mu \pi ^+)+i{\displaystyle \frac{h_A^{pK}}{F_\pi ^2}}\overline{p}\gamma ^\mu \gamma _5p(K^+D_\mu K^{}K^{}D_\mu K^+)`$ (69)
$`+i{\displaystyle \frac{h_A^{n\pi }}{F_\pi ^2}}\overline{n}\gamma ^\mu \gamma _5n(\pi ^+D_\mu \pi ^{}\pi ^{}D_\mu \pi ^+)+i{\displaystyle \frac{h_A^{nK}}{F_\pi ^2}}\overline{n}\gamma ^\mu \gamma _5n(K^+D_\mu K^{}K^{}D_\mu K^+)`$
$`+\mathrm{}.`$
Expressions for these PV vector and axial coupling constants in terms of SU(3) constants appear in Appendix B. For illustrative purposes, it is useful to express the nucleon-pion couplings in terms of the $`h_{V,A}^i`$ of Eqs. (27-33) for the SU(2) sector:
$`h_V^{pn\pi ^+}`$ $`=`$ $`h_V^0+{\displaystyle \frac{4}{3}}h_V^2`$ (70)
$`h_A^{p\pi }`$ $`=`$ $`h_A^1+h_A^2`$ (71)
$`h_A^{n\pi }`$ $`=`$ $`h_A^1h_A^2.`$ (72)
The leading order contributions to $`a_{s,v}`$ arise only from the loops in Figure 3 where a photon couples to a charged meson. The charged kaon loop contributions to the $`a_{s,v}`$ can be obtained from the corresponding formulae for the $`\pi `$-loop terms by making simple replacements of couplings and masses. For example, for the PV Yukawa interactions, these replacements are: (a) for the proton case, $`m_\pi m_K,h_\pi h_{p\mathrm{\Sigma }^0K}`$, $`g_Ag_{N\mathrm{\Sigma }^0K^+}=g_{N\mathrm{\Sigma }K}/\sqrt{2}`$ for $`\mathrm{\Sigma }^0K^+`$ intermediate states and $`h_\pi h_{p\mathrm{\Lambda }K}`$, $`g_Ag_{N\mathrm{\Lambda }K}`$ for $`\mathrm{\Lambda }K^+`$; (b) for the neutron case, $`h_\pi h_{n\mathrm{\Sigma }^{}K}`$, $`g_Ag_{N\mathrm{\Sigma }^{}K^+}=g_{N\mathrm{\Sigma }K}`$ for $`\mathrm{\Sigma }^{}K^+`$ intermediate state for the neutron case. Similar replacements hold for the vector PV coupling contributions. For the axial PV two-pion contribution we need only make the replacement $`h_A^{p\pi }h_A^{pK},m_\pi m_K`$.
Upon making these substitutions, we obtain the complete heavy baryon loop contribution to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ in SU(3):
$`a_s^L`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{24}}g_Ah_\pi [{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_\pi }}+{\displaystyle \frac{3}{\pi }}{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_N}}\mathrm{ln}({\displaystyle \frac{\mu }{m_\pi }})^2]`$ (78)
$`{\displaystyle \frac{\sqrt{3}}{144}}(1+2\alpha )g_Ah_{p\mathrm{\Lambda }K}[{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_K}}+{\displaystyle \frac{3}{\pi }}{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_N}}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2]`$
$`+{\displaystyle \frac{\sqrt{2}}{32}}(12\alpha )g_Ah_{n\mathrm{\Sigma }^{}K}[{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_K}}+{\displaystyle \frac{3}{\pi }}{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_N}}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2]`$
$`{\displaystyle \frac{1}{6}}(h_A^{p\pi }+h_A^{n\pi })\mathrm{ln}({\displaystyle \frac{\mu }{m_\pi }})^2{\displaystyle \frac{1}{6}}(h_A^{pK}+h_A^{nK})\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2`$
$`+{\displaystyle \frac{1}{12}}(12\alpha )g_A(h_V^{n\mathrm{\Sigma }^{}K^+}+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{\sqrt{2}}})\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2`$
$`{\displaystyle \frac{\sqrt{6}}{72}}(1+2\alpha )g_Ah_V^{p\mathrm{\Lambda }K^+}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2`$
$`a_v^L`$ $`=`$ $`{\displaystyle \frac{\sqrt{2}}{96}}(12\alpha )g_Ah_{n\mathrm{\Sigma }^{}K}[{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_K}}+{\displaystyle \frac{3}{\pi }}{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_N}}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2]`$ (84)
$`{\displaystyle \frac{\sqrt{3}}{144}}(1+2\alpha )g_Ah_{p\mathrm{\Lambda }K}[{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_K}}+{\displaystyle \frac{3}{\pi }}{\displaystyle \frac{\mathrm{\Lambda }_\chi }{m_N}}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2]`$
$`{\displaystyle \frac{1}{6}}(h_A^{p\pi }h_A^{n\pi })\mathrm{ln}({\displaystyle \frac{\mu }{m_\pi }})^2{\displaystyle \frac{1}{6}}(h_A^{pK}h_A^{nK})\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2`$
$`{\displaystyle \frac{1}{6}}g_Ah_V^{pn\pi ^+}\mathrm{ln}({\displaystyle \frac{\mu }{m_\pi }})^2`$
$`+{\displaystyle \frac{1}{12}}(12\alpha )g_A(h_V^{n\mathrm{\Sigma }^{}K^+}+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{\sqrt{2}}})\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2`$
$`{\displaystyle \frac{\sqrt{6}}{72}}(1+2\alpha )g_Ah_V^{p\mathrm{\Lambda }K^+}\mathrm{ln}({\displaystyle \frac{\mu }{m_K}})^2.`$
## V Low-energy Constants and Vector Mesons
A pure ChPT treatment of the anapole contributions to $`R_A`$ would use a measurment of the axial term in $`A_{LR}(ep)`$ and $`A_{LR}(QE)`$, together with the non-analytic, long-distance loop contributions, $`a_{s,v}^L`$, to determine the low-energy constants, $`a_{s,v}^{CT}`$. In the present case, however, we wish to determine whether there exist reasonable hadronic mechanisms which can enhance the low-energy constants to the level suggested by the SAMPLE results. Thus, we attempt to estimate $`a_{s,v}^{CT}`$ theoretically.
Because they are governed in part by the short-distance ($`r>1/\mathrm{\Lambda }_\chi `$) strong interaction, $`a_{s,v}^{CT}`$ are difficult to compute from first principles in QCD. Nevertheless, experience with ChPT in the pseudscalar meson sector and with the phenomenology of nucleon EM form factors suggests a reasonable model approach. It is well known, for example, that in the $`𝒪(p^4)`$ chiral Lagrangian describing pseudoscalar interactions, the low-energy constants are well-described by the exchange of heavy mesons. In particular, the charge radius of the pion receives roughly a 7% long-distance loop contribution, while the remaining 93% is saturated by $`t`$-channel exchange of the $`\rho ^0`$. Similarly, in the baryon sector, dispersion relation analyses of the isovector and isoscalar nucleon electromagnetic form factors indicate important contributions from the lightest vector mesons. Thus, it seems reasonable to assume that $`t`$-channel exchange of vector mesons also plays an important role in the short-distance physics associated with the anapole moment.
With these observations in mind, we estimate the coefficients $`a_{s,v}^{CT}`$ in the approximation that they are saturated by $`t`$-channel exchange of the lightest vector mesons, as shown in Figure 7. Here parity-violation enters through the vector meson-nucleon interaction vertices. We also use a similar picture for the electromagnetic nucleon form factors to determine the overall phase of $`a_{s,v}^{CT}`$ in the vector meson dominance approximation. To that end we require the PC and PV vector meson-nucleon Lagrangians :
$`_{\rho NN}^{PC}`$ $`=`$ $`g_{\rho NN}\overline{N}[\gamma _\mu +\kappa _\rho {\displaystyle \frac{i\sigma _{\mu \nu }q^\nu }{2m_N}}]\tau \rho ^\mu N`$ (85)
$`_{\omega NN}^{PC}`$ $`=`$ $`g_{\omega NN}\overline{N}[\gamma _\mu +\kappa _\omega {\displaystyle \frac{i\sigma _{\mu \nu }q^\nu }{2m_N}}]\omega ^\mu N`$ (86)
$`_{\varphi NN}^{PC}`$ $`=`$ $`g_{\varphi NN}\overline{N}[\gamma _\mu +\kappa _\varphi {\displaystyle \frac{i\sigma _{\mu \nu }q^\nu }{2m_N}}]\varphi ^\mu N`$ (87)
and
$`_{\rho NN}^{PV}`$ $`=`$ $`\overline{N}\gamma ^\mu \gamma _5\rho _\mu ^0[h_\rho ^1+(h_\rho ^0+{\displaystyle \frac{h_\rho ^2}{\sqrt{6}}})\tau _3]N`$ (88)
$`_{\omega NN}^{PV}`$ $`=`$ $`\overline{N}\gamma ^\mu \gamma _5\omega _\mu [h_\omega ^0+h_\omega ^1\tau _3]N`$ (89)
$`_{\varphi NN}^{PV}`$ $`=`$ $`\overline{N}\gamma ^\mu \gamma _5\varphi _\mu [h_\varphi ^0+h_\varphi ^1\tau _3]N.`$ (90)
(Note that we have adopted a different convention for $`\gamma _5`$ than used in Ref. .) The coupling constants $`h_{\rho ,\omega ,\varphi }^i`$ were estimated in Refs. and have also been constrained by a variety of hadronic and nuclear parity-violating experiments (for a review, see Ref. ).
For the $`V\gamma `$ transition amplitude, we use
$$_{V\gamma }=\frac{e}{2f_V}F^{\mu \nu }V_{\mu \nu },$$
(91)
where $`e`$ is the charge unit, $`f_V`$ is the $`\gamma `$-$`V`$ conversion constant ($`V=\rho ^0,\omega ,\varphi `$), and $`V_{\mu \nu }`$ is the corresponding vector meson field tensor. (This gauge-invariant Lagrangian ensures that the diagrams of Figure 7 do not contribute to the charge of the nucleon.) The amplitude of Figure 7 then becomes
$$a_s^{CT}(VMD)=\frac{h_\rho ^1}{f_\rho }(\frac{\mathrm{\Lambda }_\chi }{m_\rho })^2+\frac{h_\omega ^0}{f_\omega }(\frac{\mathrm{\Lambda }_\chi }{m_\omega })^2+\frac{h_\varphi ^0}{f_\varphi }(\frac{\mathrm{\Lambda }_\chi }{m_\varphi })^2,$$
(92)
$$a_v^{CT}(VMD)=\frac{h_\rho ^0+h_\rho ^2/\sqrt{6}}{f_\rho }(\frac{\mathrm{\Lambda }_\chi }{m_\rho })^2+\frac{h_\omega ^1}{f_\omega }(\frac{\mathrm{\Lambda }_\chi }{m_\omega })^2+\frac{h_\varphi ^1}{f_\varphi }(\frac{\mathrm{\Lambda }_\chi }{m_\varphi })^2.$$
(93)
The parity violating rho-pole contribution was first derived in . However, the relative sign between $`h_{\rho N}^i`$ and $`f_\rho `$ is undetermined from the diagram of Figure 7 alone. Nevertheless, we can fix the overall phase using two phenomenological inputs. Parity violating experiments in the p-p system constrain the sign of the combination $`g_{\rho N}h_{\rho N}^i`$ . In particular, the scale of the longitudinal analyzing power, $`A_L`$, is set by the combination of constants
$$A_Lg_{\rho NN}(2+\kappa _V)[h_\rho ^0+h_\rho ^1+h_\rho ^2/\sqrt{6}]+g_{\omega NN}(2+\kappa _S)[h_\omega ^0+h_\omega ^1],$$
(94)
where the constant of proportionality is positive, $`\kappa _V=3.7`$ and $`\kappa _S=0.12`$. Using the standard values for the strong $`VNN`$ couplings, one finds that $`A_L`$ has roughly the same sensitivity to each of the $`h_V^i`$ appearing in Eq. (94) (modulo the $`1/\sqrt{6}`$ coefficient of $`h_\rho ^2`$). From the 45 MeV experiment performed at SIN , for example, one obtains the approximate constraint
$$h_\rho ^0+h_\rho ^1+h_\rho ^2/\sqrt{6}+h_\omega ^0+h_\omega ^128\pm 4,$$
(95)
where the $`h_V^i`$ have are expressed in units of $`g_\pi `$ and where a positive sign has been assumed for $`g_{VNN}`$. Given this constraint, it is very unlikely that the product $`(h_\rho ^0+h_\rho ^2/\sqrt{6})g_{\rho NN}>0`$ unless the corresponding products involving $`h_\rho ^1`$ and $`h_\omega ^{0,1}`$ in Eq. (94) obtain anomalously large, negative values. In fact, a fit to hadronic and nuclear PV observables in Ref. strongly favors a phase difference between the strong and weak $`VNN`$ couplings.
Experimentally, one also knows the isovector nucleon charge radius
$$r^2_{EXP}^{T=1}=6\frac{dF_1(q^2)}{dq^2}|_{q^2=0}>0,$$
(96)
where
$$<p^{}|j_\mu ^{T=1}(0)|p>=e\overline{u}(p^{})[F_1(q^2)+\frac{i\sigma _{\mu \nu }q^\nu }{2m_N}F_2(q^2)]u(p).$$
(97)
One may reasonably approximate the $`\rho ^0`$ contribution to $`r^2^{T=1}`$ using VMD . The calculation is the same as above but with the weak hadronic coupling replaced by the strong coupling. The result is
$$F_1^{\rho ^0}(q^2)=\frac{g_{\rho NN}}{f_\rho }\frac{q^2}{q^2m_\rho ^2}.$$
(98)
Then we have
$$\frac{dF_1^{VMD}(q^2)}{dq^2}|_{q^2=0}=\frac{g_{\rho NN}}{f_\rho m_\rho ^2}.$$
(99)
Comparing Eqs. (96) and (99), and noting that the $`\rho ^0`$ generates a positive contribution to $`r^2^{T=1}`$ , we arrive at $`g_{\rho NN}/f_\rho <0`$. Combining this result with $`g_{\rho NN}h_\rho ^i<0`$ as favored by the $`\stackrel{}{p}p`$ experiments we obtain the relative sign between $`h_\rho ^i`$ and $`f_\rho `$: $`h_\rho ^i/f_\rho >0`$. Accordingly we determine the relative signs for PV $`\omega ,\varphi `$-nucleon coupling constants.
## VI The scale of $`R_A`$
Expressions for the anapole contributions to $`R_A^{T=0}`$ and $`R_A^{T=1}`$ in terms of the $`a_{s,v}`$ appear in Eq. (10). We may now use these expressions, along with the results in Eqs. (78-84) and (92-93), to obtain a numerical estimate for the $`R_A^{T=0,1}|_{\text{anapole}}`$. To do so, we use the global fit value for the weak mixing angle in the on-shell scheme, $`\mathrm{sin}^2\theta _W=0.2230`$ , $`g_A=1.267\pm 0.004`$ , $`f_\rho =5.26`$ , $`f_\omega =17,f_\varphi =13`$ , $`\alpha =F/(D+F)=0.36`$, $`\mu =\mathrm{\Lambda }_\chi `$. We express all the PV coupling constants in units of $`g_\pi =3.8\times 10^8`$ as is traditionally done . We obtain
$`R_A^{T=0}|_{\text{anapole}}`$ $`=`$ $`10^2\{0.17h_\pi +h_A^10.0036h_{n\mathrm{\Sigma }^{}K}`$ (102)
$`0.033(h_V^{n\mathrm{\Sigma }^{}K^+}+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{\sqrt{2}}})+0.2(h_A^{pK}+h_A^{nK})0.006h_{p\mathrm{\Lambda }K}`$
$`+0.088h_V^{p\mathrm{\Lambda }K^+}0.26|h_\rho ^1|0.08|h_\omega ^0|0.05|h_\varphi ^0|\}`$
$`R_A^{T=1}|_{\text{anapole}}`$ $`=`$ $`10^2\{h_A^20.6(h_V^0+{\displaystyle \frac{4}{3}}h_V^2)0.0012h_{n\mathrm{\Sigma }^{}K}0.033(h_V^{n\mathrm{\Sigma }^{}K^+}`$ (105)
$`+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{\sqrt{2}}})+0.2(h_A^{pK}h_A^{nK})0.006h_{p\mathrm{\Lambda }K}+0.088h_V^{p\mathrm{\Lambda }K^+}`$
$`0.26(|h_\rho ^0|+{\displaystyle \frac{|h_\rho ^2|}{\sqrt{6}}})0.087|h_\omega ^1|+0.05|h_\varphi ^1|\},`$
where we have set the phase of the vector meson contributions as discussed above, and used the relations in Eq. (70).
The expressions in Eqs. (102-105) illustrate the sensitivity of the radiative corrections to the various PV hadronic couplings. As expected on general grounds, the overall scale of $`R_A^{T=0,1}`$ is at about the one percent level \[see Eq. (13)\]. In terms of the conventional PV couplings, $`R_A^{T=0}`$ is most sensitive to $`h_\pi `$ and $`h_\rho ^1`$, while $`R_A^{T=1}`$ is most strongly influenced by $`h_\rho ^0+h_\rho ^2/\sqrt{6}`$. The corrections also display strong dependences on the couplings $`h_{V,A}^i`$ not included in the standard analysis of nuclear and hadronic PV. In particular, the couplings $`h_A^2`$ and $`h_V^0+4h_V^2/3`$ are weighted heavily in $`R_A^{T=1}`$. In general, the sensitivity to the PV $`NYK`$ couplings is considerably weaker than the sensitivity to the $`NN\pi `$ and $`NN\rho `$ couplings.
In order to make an estimate of $`R_A^{T=0,1}`$, we require inputs for the PV couplings. To that end, we use the “best values” for $`h_\pi `$, $`h_\rho ^i`$, and $`h_\omega ^i`$ given in Ref. . These values are consistent with the fit of Ref. . For the $`h_\varphi ^i`$ we use the “best values” of Ref. . The analyses given in Refs. , together with experimental input, also allow for the standard couplings to take on a range of values. For example, the ranges for the $`h_\omega ^i`$ given in Refs. correspond to
$$33h_\omega ^0+h_\omega ^113.$$
(106)
In order to maintain consistency with the experimental constraint of Eq. (95), one then requires
$$0h_\rho ^0+h_\rho ^1+h_\rho ^2/\sqrt{6}45.$$
(107)
We adopt this range even though it is smaller than the range given in Ref. . Indeed, allowing the $`h_\rho ^i`$ to assume the full ranges given in Ref. would require the $`h_\omega ^i`$ to vary outside their corresponding theoretical “reasonable ranges” if the constraint of Eq. (95) is to be satisfied. Since one expects $`|h_\rho ^1|<<|h_\rho ^{0,2}|`$ , we have a reasonable range of values for the important isoscalar $`\rho `$ contribution in Eq. (105), and the rather broad range of values allowed for the $`h_\rho ^i`$ contributes significantly to our estimated uncertainty in $`R_A^{T=1}`$. For $`h_{\omega ,\varphi }^i`$, we use the ranges of Refs. Allowing the $`h_\omega ^i`$ to assume positive values would require a sign change on the correspnding terms in Eqs. (102,105).. In contrast to the situation with the $`h_\rho ^0`$ contribution, however, the variation in the $`h_{\omega ,\varphi }^i`$ over their “reasonable ranges” has negligible impact on our estimated theoretical uncertainty.
Estimating values for the Yukawa couplings $`h_{NYK}`$ and for the $`h_{V,A}`$ is more problematic—to date, no calculation on the level of Ref. has been performed for such couplings. Estimates for $`h_{V,A}`$, based on dimensional and factorization arguments, were given in Ref. and generally yielded values for $`h_{V,A}`$ in the non-strange sector on the order of $`g_\pi `$. For our central values, then, we take $`h_{V,A}^i=g_\pi `$, resulting in roughly 1% contributions from the PV vector and axial vector interactions. Without performing a detailed calculation as in Ref. , one might also attempt to determine reasonable ranges for these parameters by looking to phenomenology. To that end, the authors of Ref. considered analogies between the axial vector PV operators of Eq. (30-33) and contact operators needed to explain the size of $`\mathrm{\Delta }I=1/2`$ hyperon P-wave decay amplitudes. From this analogy, these authors conclude that $`|h_A^i|10g_\pi `$ may be reasonable. However, whether such large ranges are consistent with nuclear PV data remains to be determined. In the absence of such an analysis, which goes beyond the scope of the present work, we adopt the range $`10g_\pi h_A^i10g_\pi `$ suggested in Ref. . The corresponding uncertainties in the $`R_A^{T=0,1}`$ are roughly $`\pm 10\%`$.
The implications of phenomenology for the $`h_V^i`$ are even less clear than for the $`h_A^i`$. However, we note that large values $`h_V^i\pm 10g_\pi `$ do not appear to be ruled out by hadronic and nuclear PV data. At tree-level, for example, the vector terms in $`_{\mathrm{\Delta }T=0,1,2}^{\pi N}`$ do not contribute to the PV NN interaction through the one $`\pi `$-exchange amplitudes of Figure 8a. It is straightforward to show that the corresponding amplitude vanishes for on-shell nucleons <sup>§</sup><sup>§</sup>§The on-shell approximation is generally used in deriving the PV NN potential from Feynman diagrams.. Thus, at this level, purely hadronic PV processes are insensitive to the $`h_V^i`$ and provide no constraints on these couplings. In PV electromagnetic processes, however, the $`h_V^i`$ do contribute through PV two-body currents, such as those shown in Figure 8b. Nevertheless, one expects the impact of PV two-body currents to be considerably weaker than that of the PV NN potential. The PV $`\gamma `$-decay of <sup>18</sup>F, for example, is dominated by the mixing of a nearly-degenerate pair of $`(J^\pi ,T)=(0^{},0)`$ and $`(0^+,1)`$ states. The small energy denominator associated with this parity-mixing enhances the relative importance of the PV NN potential by roughly two orders of magnitude over the generic situation with typical nuclear level spacings. By contrast, the PV two-body currents do not participate in parity-mixing and receive no such enhancements. A similar situation holds for PV electromagnetic processes in other nuclei of interest. Hence, we expect the PV $`\gamma `$-decays of light nuclei to be relatively insensitive to the $`h_V^i`$, even if the latter are on the order of $`10g_\pi `$. Consequently, we rather generously take $`10g_\pi h_V^0+4h_V^2/310g_\pi `$, yielding a $`\pm 7\%`$ contribution to the uncertainty in $`R_A^{T=1}`$. Allowing similarly large ranges for the PV $`NYK`$ couplings has a negligible impact on the uncertainty in the $`R_A^{T=0,1}`$.
With these input values for the PV couplings, we arrive at the anapole contributions to $`R_A^{T=0,1}`$ shown in Table I. The latter must be added to the one-quark Standard Model contributions, also shown in Table I. We compute the one-quark corrections using the on-shell parameters given in Refs. . We emphasize that the quoted values for the $`R_A^{T=0,1}`$ are renormalization scheme-dependent. The relative size of the isovector one-quark corrections are smaller, for example, in the $`\overline{\text{MS}}`$ scheme, where one has $`R_A^{T=1}(\text{SM})=0.18`$ and $`R_A^{T=0}(\text{SM})=0.07`$. The corresponding tree-level amplitude, however, is also smaller by a factor of $`1.44`$ than the on-shell tree-level amplitude. A reader working in the $`\overline{\text{MS}}`$ scheme should, therefore, take care to adjust the tree-level amplitude and SM radiative corrections appropriately from the on-shell values used here. Moreover, the anapole contributions to the $`R_A^{(T)}`$ will be a factor of 1.44 larger in the $`\overline{\text{MS}}`$ scheme since the tree-level amplitude is correspondingly smallerNote that $`R_A^{T=0}`$ gives the ratio of the isoscalar, axial vector amplitude to the tree-level isovector, axial vector amplitude. The sign of $`R_A^{T=0}`$ as defined here is opposite that of Ref. ..
Adding the one-quark and anapole contributions yields a large, negative value for $`R_A^{T=1}`$. This result contains considerable theoretical uncertainty, mostly due to our liberal assignment of reasonable ranges to the $`h_{V,A}`$. Even with this generous theoretical uncertainty, however, $`R_A^{T=1}`$ is still roughly a factor of two away from the apparent SAMPLE result. Compared with the one-quark SM contribution, the many-quark anapole contribution is relatively small – though it does push the total in the right direction. The isoscalar correction, $`R_A^{T=0}`$, is considerably smaller in magnitude than $`R_A^{T=1}`$ yet retains a sizeable theoretical uncertainty.
## VII Conclusions
In view of the preliminary SAMPLE result for PV quasielastic electron scattering from <sup>2</sup>H, we have up-dated our previous calculation of the axial vector radiative corrections $`R_A^{T=0,1}`$. Using the framework of HBChPT, we have computed all many-quark anapole contributions through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. We include new one-loop contributions involving the PV vector couplings, $`h_V^i`$ and estimate the scale of the analytic, low-energy constants using resonance saturation. We fix the sign of the latter using the phenomenology of PV $`\stackrel{}{p}p`$ scattering and of nucleon EM form factors. We also show that large classes of loops involving decuplet intermediate states, magnetic insertions, and PV EM insertions vanish through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. Finally, we extend the previous analyses to include SU(3) symmetry, and determine that the impact of kaon loops is generally negligible. In the end, we find that $`R_A^{T=1}`$—though large and negative—is still a factor of two or so away from the suggestion that $`R_A^{T=1}1`$ from the SAMPLE experiment. Even allowing for considerable theoretical uncertainty – dominated by the PV couplings $`h_{V,A}^i`$ – there remains a sizable gap between our result and the preliminary experimental value.
There exist a number of possible additional contributions to $`R_A^{T=0,1}`$ not considered here which may ultimately account for the apparent experimental result. The most obvious include higher-order chiral corrections. This appears, however, to be an unlikely source of large contributions. On general grounds, we expect the size of the $`𝒪(1/\mathrm{\Lambda }_\chi ^3)`$ contributions to be suppressed by $`m/\mathrm{\Lambda }_\chi `$ relative to those considered here, where $`m`$ denotes a pseudoscalar mass. For kaon loops, this suppression factor is only $`1/2`$; however, at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ kaon loops generate at most a few percent contribution to $`R_A^{T=0,1}`$. The suppression factor for the next order pionic contributions is closer to 0.1. Hence, it would be surprising if the next order in the chiral expansion could close the factor of two gap with experiment.
More promising sources of sizeable contributions include $`Z\gamma `$ box graph contributions, where the full tower of hadronic intermediate states is included, as well as parity-mixing in the deuteron wavefunction. At a more speculative level, one might also consider contributions from physics beyond the Standard Model. For example, the presence of an additional, relatively light neutral gauge boson might modify the SM $`V(e)\times A(q)`$ amplitudes and contribute to $`R_A^{T=1}`$. A popular class of $`Z^{}`$ models are generated by E<sub>6</sub> symmetry . The contribution of an extra, neutral weak E<sub>6</sub> gauge boson $`Z^{}`$ is given by
$$R_A^{T=1}(\text{new})=\frac{4}{5}\frac{1}{14\mathrm{sin}^2\theta _W}\mathrm{sin}^2\varphi \frac{G_\varphi ^{}}{G_\mu },$$
(108)
where $`\varphi `$ is a mixing angle which governs the structure of an additional U(1) group in E<sub>6</sub> theories and $`G_\varphi ^{}`$ is the Fermi constant associated with the new U(1) group . Note that this contribution has the wrong sign to account for the large negative value of $`R_A^{T=1}`$.
Alternatively, one might consider new tree-level interactions generated by supersymmetric extensions of the SM. Such interactions arise when R-parity, or equivalently, $`BL`$, is not conserved ($`B`$ and $`L`$ denote baryon and lepton number, respectively). The contribution from R-parity violating SUSY interactions is given by
$$R_A^{T=1}(\text{new})=\left(\frac{1}{14x}\right)\left[\mathrm{\Delta }_{11k}^{}(\stackrel{~}{d}_R^k)\mathrm{\Delta }_{1j1}^{}(\stackrel{~}{q}_L^j)\mathrm{\Delta }_{12k}(\stackrel{~}{e}_R^k)(14x+4\lambda _x)\right],$$
(109)
where $`x=\mathrm{sin}^2\theta _W`$,
$$\lambda _x=\frac{x(1x)}{12x}\left(\frac{1}{1\mathrm{\Delta }r}\right)0.3,$$
(110)
$`\mathrm{\Delta }r`$ is a radiative correction, and where
$$\mathrm{\Delta }_{ijk}(\stackrel{~}{f})=\frac{1}{4\sqrt{2}}\frac{|\lambda _{ijk}|^2}{G_\mu M_{\stackrel{~}{f}}^2},$$
(111)
with $`\stackrel{~}{f}`$ denoting the superpartner of fermion $`f`$ and $`i,j,k`$ labeling fermion generations. The terms having a prime are semileptonic whereas the un-primed terms are purely leptonic. In principle, the correction in Eq. (109) could generate a negative contribution to $`R_A^{T=1}`$. However, the various other electroweak data constrain the terms appearing in this expression. For example, relations between $`G_\mu `$ and other SM parameters require
$$0.0023\mathrm{\Delta }_{12k}(\stackrel{~}{e}_R^k)0.0028,$$
(112)
at 90 % C.L., so that the first term in Eq. (109) cannot provide the large negative contribution needed to explain the SAMPLE result. Similarly, assuming only the semileptonic R-parity violating interactions modify the weak charge of nuclei, the recent determination of the cesium weak charge by the Boulder group implies that
$$0.00262.6\mathrm{\Delta }_{11k}^{}(\stackrel{~}{d}_R^k)2.9\mathrm{\Delta }_{1j1}^{}(\stackrel{~}{q}_L^j)0.015,$$
(113)
at 95 % C.L. (for $`m_H=300`$ GeV). Thus, it appears unlikely that the second term in Eq. (109) could enhance $`R_A^{T=1}`$ by a factor of two.
In short, two of the most popular new physics scenarios having implications for low-energy phenomenology appear unlikely to enhance $`R_A^{T=1}`$ significantly. Thus, if more conventional hadronic and nuclear processes cannot account for the SAMPLE result, one may be forced to consider more exotic alternatives.
## Acknowledgement
We wish to thank D. Beck, E.J. Beise, and R. McKeown for useful discussions. This work was supported in part under U.S. Department of Energy contract #DE-AC05-84ER40150, the National Science Foundation and a National Science Foundation Young Investigator Award.
## A Formalism
In this section we first review the general parity and CP conserving Lagrangians including $`N,\pi ,\mathrm{\Delta },\gamma `$ in the relativistic form. We follow standard conventions and introduce
$$\mathrm{\Sigma }=\xi ^2,\xi =e^{\frac{i\pi }{F_\pi }},\pi =\frac{1}{2}\pi ^a\tau ^a$$
(A1)
with $`F_\pi =92.4`$ MeV being the pion decay constant. The chiral vector and axial vector currents are given by
$`A_\mu `$ $`=`$ $`{\displaystyle \frac{i}{2}}(\xi D_\mu \xi ^{}\xi ^{}D_\mu \xi )={\displaystyle \frac{D_\mu \pi }{F_\pi }}+O(\pi ^3)`$ (A2)
$`V_\mu `$ $`=`$ $`{\displaystyle \frac{1}{2}}(\xi D_\mu \xi ^{}+\xi ^{}D_\mu \xi ).`$ (A3)
and we require also the gauge and chiral covariant derivativs
$`D_\mu \pi `$ $`=`$ $`_\mu \pi ie𝒜_\mu [Q,\pi ]`$ (A4)
$`𝒟_\mu `$ $`=`$ $`D_\mu +V_\mu ,`$ (A5)
with
$$Q=\left(\begin{array}{cc}\frac{2}{3}\hfill & 0\hfill \\ 0\hfill & \frac{1}{3}\hfill \end{array}\right)$$
(A6)
and $`𝒜_\mu `$ being the photon field. The chiral field strength tensors are
$$F_{\mu \nu }^\pm =\frac{1}{2}(_\mu 𝒜_\nu _\nu 𝒜_\mu )(\xi Q^{}\xi ^{}\pm \xi ^{}Q^{}\xi )$$
(A7)
with
$$Q^{}=\left(\begin{array}{cc}1\hfill & 0\hfill \\ 0\hfill & 0\hfill \end{array}\right)$$
(A8)
acting in the space of baryon isodoublets.
For the moment, we restrict our attention to SU(2) flavor space and consider just $`\pi `$, $`N`$, and $`\mathrm{\Delta }`$ degrees of freedom. We represent the nucleon as a two component isodoublet field, while for the $`\mathrm{\Delta }`$, we use the isospurion formalism, treating the $`\mathrm{\Delta }`$ field $`T_\mu ^i(x)`$ as a vector spinor in both spin and isospin space with the constraint $`\tau ^iT_\mu ^i(x)=0`$. The components of this field are
$$T_\mu ^3=\sqrt{\frac{2}{3}}\left(\begin{array}{c}\mathrm{\Delta }^+\hfill \\ \mathrm{\Delta }^0\hfill \end{array}\right)_\mu ,T_\mu ^+=\left(\begin{array}{c}\mathrm{\Delta }^{++}\hfill \\ \mathrm{\Delta }^+/\sqrt{3}\hfill \end{array}\right)_\mu ,T_\mu ^{}=\left(\begin{array}{c}\mathrm{\Delta }^0/\sqrt{3}\hfill \\ \mathrm{\Delta }^{}\hfill \end{array}\right)_\mu .$$
(A9)
The field $`T_\mu ^i`$ also satisfies the constraints for the ordinary Schwinger-Rarita spin-$`\frac{3}{2}`$ field,
$$\gamma ^\mu T_\mu ^i=0\text{and}p^\mu T_\mu ^i=0.$$
(A10)
We eventually convert to the heavy baryon expansion, in which case the latter constraint becomes $`v^\mu T_\mu ^i=0`$ with $`v_\mu `$ the heavy baryon velocity.
It is useful to review the spacetime and chiral transformation properties of the various fields. Under a chiral transformation,
$`\xi L\xi U^{}=U\xi R^{}`$ (A11)
$`A_\mu UA_\mu U^{}`$ (A12)
$`𝒟_\mu U𝒟_\mu U^{},`$ (A13)
and
$$NUN,T_\mu UT_\mu ,\mathrm{\Sigma }L\mathrm{\Sigma }R^{}\text{etc}.$$
(A14)
In the $`SU(2)`$ sector parity violating effects are conveniently described by introducing the operators :
$$X_L^a=\xi ^{}\tau ^a\xi ,X_R^a=\xi \tau ^a\xi ^{},X_\pm ^a=X_L^a\pm X_R^a.$$
(A15)
which transform as
$$X_{L,R}^aU\stackrel{~}{X_{L,R}^a}U^{},$$
(A16)
with the index a rotating like a vector of $`SU(2)_L`$ and $`SU(2)_R`$ respectively.
The P and CP transformation properties of these fields are shown in Table 2.
Finally, we note that in the Lagrangians of Section III, one has the following definitions:
$`𝒟_\mu ^{ij}`$ $`=`$ $`\delta ^{ij}𝒟_\mu 2iϵ^{ijk}V_\mu ^k`$ (A17)
$`\omega _\mu ^i`$ $`=`$ $`Tr[\tau ^iA_\mu ]`$ (A18)
$`A_\mu ^{ij}`$ $`=`$ $`\xi _{3/2}^{ik}A_\mu \xi _{3/2}^{kj},`$ (A19)
where $`\xi _{3/2}^{ij}=\frac{2}{3}\delta ^{ij}\frac{i}{3}ϵ_{ijk}\tau ^k`$ is the isospin $`3/2`$ projection operator.
## B The $`SU(3)`$ parity violating and CP conserving Lagrangian
In this Appendix we list the parity violating and CP conserving $`SU(3)`$ Lagrangian for the pseudosclar meson octet and baryon octet. We are interested in the diagonal case of the parity violating electron nucleon scattering. Hence, we include only those interaction terms that ensure strangeness and charge conservation at each vertex. In the following we use $`\xi =e^{\frac{i\pi }{F_\pi }},\pi =\frac{1}{2}\pi ^a\lambda ^a`$, $`X_L^a=\xi ^{}\lambda ^a\xi `$, $`X_R^a=\xi \lambda ^a\xi ^{}`$, $`X_\pm ^a=X_L^a\pm X_R^a`$, $`[A,B]_\pm =AB\pm BA`$.
We classify the parity violating Lagrangian according to isospin violation $`\mathrm{\Delta }T=0,1,2`$, which arises from the operators of $`X_L^a,X_R^a`$, their combinations and products. The $`\mathrm{\Delta }T=2`$ piece comes from the operators $`^{ab}\{X_L^a𝒪X_L^b\pm (LR)\}`$ with $`𝒪=N,\overline{N},A_\mu `$ and
$$^{ab}=\frac{1}{3}\left(\begin{array}{ccc}1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 2\hfill \end{array}\right),$$
(B1)
where $`a,b=1,2,3`$. Several operators contribute to the $`\mathrm{\Delta }T=1`$ part, like $`X_\pm ^3,f^{3ab}\{X_L^a𝒪X_L^b\pm (LR)\},d^{3ab}\{X_L^a𝒪X_L^b\pm (LR)\}`$ where $`f^{abc},d^{abc}`$ are the antisymmetric and symmetric structure constants of $`SU(3)`$ algebra. With the requirement that the final Lagrangian be hermitian, parity-violating and CP-conserving, the operator with $`f^{3ab}`$ vanishs. For the $`\mathrm{\Delta }T=0`$ part relevant operators are $`\mathrm{𝟏},X_\pm ^8,f^{8ab}\{X_L^a𝒪X_L^b\pm (LR)\},d^{8ab}\{X_L^a𝒪X_L^b\pm (LR)\},\delta ^{ab}\{X_L^a𝒪X_L^b\pm (LR)\}`$. For the same reason the $`f^{8ab}`$ structure does not contribute. Note the matrix identity $`\lambda ^a\lambda ^b\lambda ^a=4(C_2(\mathrm{𝟑})\frac{1}{2}C_2(\mathrm{𝟖}))\lambda ^b`$, where $`C_2(\mathrm{𝟑}),C_2(\mathrm{𝟖})`$ are the Casimir invariants of the basic and adjoint representations of $`SU(3)`$ group respectively. Hence, the operator containing $`\delta ^{ab}`$ is identical to the unit operator.
Based on these considerations, we obtain
$`_{\mathrm{\Delta }T=0}^{\text{PV}}=h_3F_\pi Tr\overline{N}[X_{}^8,N]_++h_4F_\pi Tr\overline{N}[X_{}^8,N]_{}+v_1Tr\overline{N}\gamma ^\mu [A_\mu ,N]_+`$ (B2)
$`+v_2Tr\overline{N}\gamma ^\mu [A_\mu ,N]_{}+{\displaystyle \frac{v_7}{2}}Tr\overline{N}\gamma ^\mu A_\mu NX_+^8+{\displaystyle \frac{v_8}{2}}Tr\overline{N}\gamma ^\mu X_+^8NA_\mu `$ (B3)
$`+{\displaystyle \frac{v_9}{2}}Tr\overline{N}\gamma ^\mu [X_+^8,A_\mu ]_+N+{\displaystyle \frac{v_{10}}{2}}Tr\overline{N}\gamma ^\mu N[X_+^8,A_\mu ]_++a_5Tr\overline{N}\gamma ^\mu \gamma _5A_\mu NX_{}^8`$ (B4)
$`+a_6Tr\overline{N}\gamma ^\mu \gamma _5X_{}^8NA_\mu +a_7Tr\overline{N}\gamma ^\mu \gamma _5[X_{}^8,A_\mu ]_+N+a_8Tr\overline{N}\gamma ^\mu \gamma _5N[X_{}^8,A_\mu ]_+`$ (B5)
$`+\sqrt{3}v_{11}d^{8ab}Tr\{\overline{N}\gamma ^\mu NX_L^aA_\mu X_L^b+(LR)\}`$ (B6)
$`+\sqrt{3}v_{12}d^{8ab}Tr\{\overline{N}\gamma ^\mu X_L^aA_\mu X_L^bN+(LR)\}`$ (B7)
$`+\sqrt{3}v_{13}d^{8ab}Tr\{\overline{N}\gamma ^\mu X_L^aN[X_L^b,A_\mu ]_++(LR)\}`$ (B8)
$`+\sqrt{3}v_{14}d^{8ab}Tr\{\overline{N}\gamma ^\mu [X_L^a,A_\mu ]_+NX_L^b+(LR)\}`$ (B9)
$`+\sqrt{3}a_9d^{8ab}Tr\{\overline{N}\gamma ^\mu \gamma _5NX_L^aA_\mu X_L^b(LR)\}`$ (B10)
$`+\sqrt{3}a_{10}d^{8ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aA_\mu X_L^bN(LR)\}`$ (B11)
$`+\sqrt{3}a_{11}d^{8ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aN[X_L^b,A_\mu ]_+(LR)\}`$ (B12)
$`+\sqrt{3}a_{12}d^{8ab}Tr\{\overline{N}\gamma ^\mu \gamma _5[X_L^a,A_\mu ]_+NX_L^b(LR)\}`$ $`,`$ (B13)
$`_{\mathrm{\Delta }T=1}^{\text{PV}}=h_1F_\pi Tr\overline{N}[X_{}^3,N]_++h_2F_\pi Tr\overline{N}[X_{}^3,N]_{}+{\displaystyle \frac{v_3}{2}}Tr\overline{N}\gamma ^\mu A_\mu NX_+^3`$ (B14)
$`+{\displaystyle \frac{v_4}{2}}Tr\overline{N}\gamma ^\mu X_+^3NA_\mu +{\displaystyle \frac{v_5}{2}}Tr\overline{N}\gamma ^\mu [X_+^3,A_\mu ]_+N+{\displaystyle \frac{v_6}{2}}Tr\overline{N}\gamma ^\mu N[X_+^3,A_\mu ]_+`$ (B15)
$`+a_1Tr\overline{N}\gamma ^\mu \gamma _5A_\mu NX_{}^3+a_2Tr\overline{N}\gamma ^\mu \gamma _5X_{}^3NA_\mu +a_3Tr\overline{N}\gamma ^\mu \gamma _5[X_{}^3,A_\mu ]_+N`$ (B16)
$`+a_4Tr\overline{N}\gamma ^\mu \gamma _5N[X_{}^3,A_\mu ]_++v_{15}d^{3ab}Tr\{\overline{N}\gamma ^\mu NX_L^aA_\mu X_L^b+(LR)\}`$ (B17)
$`+v_{16}d^{3ab}Tr\{\overline{N}\gamma ^\mu X_L^aA_\mu X_L^bN+(LR)\}`$ (B18)
$`+v_{17}d^{3ab}Tr\{\overline{N}\gamma ^\mu X_L^aN[X_L^b,A_\mu ]_++(LR)\}`$ (B19)
$`+v_{18}d^{3ab}Tr\{\overline{N}\gamma ^\mu [X_L^a,A_\mu ]_+NX_L^b+(LR)\}`$ (B20)
$`+a_{13}d^{3ab}Tr\{\overline{N}\gamma ^\mu \gamma _5NX_L^aA_\mu X_L^b(LR)\}`$ (B21)
$`+a_{14}d^{3ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aA_\mu X_L^bN(LR)\}`$ (B22)
$`+a_{15}d^{3ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aN[X_L^b,A_\mu ]_+(LR)\}`$ (B23)
$`+a_{16}d^{3ab}Tr\{\overline{N}\gamma ^\mu \gamma _5[X_L^a,A_\mu ]_+NX_L^b(LR)\}`$ $`,`$ (B24)
$`_{\mathrm{\Delta }T=2}^{\text{PV}}={\displaystyle \frac{v_{19}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu NX_L^aA_\mu X_L^b+(LR)\}`$ (B25)
$`+{\displaystyle \frac{v_{20}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu X_L^aA_\mu X_L^bN+(LR)\}`$ (B26)
$`+{\displaystyle \frac{v_{21}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu X_L^aN[X_L^b,A_\mu ]_++(LR)\}`$ (B27)
$`+{\displaystyle \frac{v_{22}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu [X_L^a,A_\mu ]_+NX_L^b+(LR)\}`$ (B28)
$`+{\displaystyle \frac{a_{17}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu \gamma _5NX_L^aA_\mu X_L^b(LR)\}`$ (B29)
$`+{\displaystyle \frac{a_{18}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aA_\mu X_L^bN(LR)\}`$ (B30)
$`+{\displaystyle \frac{a_{19}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu \gamma _5X_L^aN[X_L^b,A_\mu ]_+(LR)\}`$ (B31)
$`+{\displaystyle \frac{a_{20}}{2}}^{ab}Tr\{\overline{N}\gamma ^\mu \gamma _5[X_L^a,A_\mu ]_+NX_L^b(LR)\}`$ $`.`$ (B32)
These Lagrangians contain 4 Yukawa couplings, 20 axial vector couplings and 22 vector couplings, all of which should be fixed from the experimental data or from model calculations. In reality, however, we have only limited information which constrains a few of them. It is useful to expand the above Lagrangians to the order involving the minimum number of Goldstone bosons and to collect those vertices needed in the calculation of $`R_A`$:
$`_{Yukawa}^{1\pi }=2\sqrt{2}i(h_1+h_2)(\overline{p}n\pi ^+\overline{n}p\pi ^{})`$ (B33)
$`+i[h_1h_2+\sqrt{3}(h_3h_4)](\overline{p}\mathrm{\Sigma }^0K^+\overline{\mathrm{\Sigma }^0}pK^{})`$ (B34)
$`+\sqrt{2}i[h_1h_2+\sqrt{3}(h_3h_4)](\overline{n}\mathrm{\Sigma }^{}K^+\overline{\mathrm{\Sigma }}^{}nK^{})`$ (B35)
$`i[{\displaystyle \frac{h_1}{\sqrt{3}}}+\sqrt{3}h_2+h_3+3h_4](\overline{p}\mathrm{\Lambda }K^+\overline{\mathrm{\Lambda }}pK^{})+\mathrm{}`$ $`.`$ (B36)
$`_V^{1\pi }={\displaystyle \frac{h_V^{pn\pi ^+}}{F_\pi }}\overline{p}\gamma ^\mu nD_\mu \pi ^+{\displaystyle \frac{h_V^{p\mathrm{\Sigma }^0K^+}}{F_\pi }}\overline{p}\gamma ^\mu \mathrm{\Sigma }^0D_\mu K^+`$ (B37)
$`{\displaystyle \frac{h_V^{n\mathrm{\Sigma }^{}K^+}}{F_\pi }}\overline{n}\gamma ^\mu \mathrm{\Sigma }^{}D_\mu K^+{\displaystyle \frac{h_V^{p\mathrm{\Lambda }K^+}}{F_\pi }}\overline{p}\gamma ^\mu \mathrm{\Lambda }D_\mu K^++h.c.+\mathrm{}`$ $`,`$ (B38)
where
$`h_V^{pn\pi ^+}={\displaystyle \frac{v_1+v_2}{\sqrt{2}}}+{\displaystyle \frac{4\sqrt{2}}{3}}(v_{14}v_{12})+{\displaystyle \frac{\sqrt{6}}{3}}(v_7+v_9)+{\displaystyle \frac{\sqrt{2}}{3}}v_{20}`$ (B39)
$`h_V^{p\mathrm{\Sigma }^0K^+}={\displaystyle \frac{1}{2}}(v_1v_2+v_4+v_6)+{\displaystyle \frac{v_8v_{10}}{2\sqrt{3}}}+{\displaystyle \frac{2}{3}}(v_{11}v_{13}v_{15}v_{21}{\displaystyle \frac{1}{2}}v_{17})+2v_{18}`$ (B40)
$`h_V^{n\mathrm{\Sigma }^{}K^+}={\displaystyle \frac{1}{\sqrt{2}}}(v_1v_2+v_6v_4)+{\displaystyle \frac{1}{\sqrt{6}}}(v_8v_{10})+{\displaystyle \frac{\sqrt{2}}{3}}(v_{17}+v_{21})+{\displaystyle \frac{2\sqrt{2}}{3}}(v_{11}v_{13}v_{15})`$ (B41)
$`h_V^{p\mathrm{\Lambda }K^+}={\displaystyle \frac{1}{\sqrt{3}}}({\displaystyle \frac{v_1}{2}}+{\displaystyle \frac{2}{3}}v_{11}{\displaystyle \frac{4}{3}}v_{12}+{\displaystyle \frac{16}{3}}v_{13}{\displaystyle \frac{2}{3}}v_{14}{\displaystyle \frac{2}{3}}v_{15}+{\displaystyle \frac{4}{3}}v_{16}{\displaystyle \frac{17}{3}}v_{17}`$ (B42)
$`+{\displaystyle \frac{4}{3}}v_{18}{\displaystyle \frac{3}{2}}v_2+{\displaystyle \frac{v_4}{2}}v_5+{\displaystyle \frac{v_6}{2}})+{\displaystyle \frac{v_8v_{10}}{6}}+{\displaystyle \frac{2v_7+v_9}{3}}`$ $`.`$ (B43)
$`_A^{2\pi }=i{\displaystyle \frac{h_A^{p\pi }}{f_\pi ^2}}\overline{p}\gamma ^\mu \gamma _5p(\pi ^+D_\mu \pi ^{}\pi ^{}D_\mu \pi ^+)`$ (B44)
$`i{\displaystyle \frac{h_A^{pK}}{f_\pi ^2}}\overline{p}\gamma ^\mu \gamma _5p(K^+D_\mu K^{}K^{}D_\mu K^+)`$ (B45)
$`i{\displaystyle \frac{h_A^{n\pi }}{f_\pi ^2}}\overline{n}\gamma ^\mu \gamma _5n(\pi ^+D_\mu \pi ^{}\pi ^{}D_\mu \pi ^+)`$ (B46)
$`i{\displaystyle \frac{h_A^{nK}}{f_\pi ^2}}\overline{n}\gamma ^\mu \gamma _5n(K^+D_\mu K^{}K^{}D_\mu K^+)+\mathrm{}`$ $`,`$ (B47)
where
$`h_A^{p\pi }=2a_3{\displaystyle \frac{4}{3}}a_{16}+{\displaystyle \frac{2}{3}}a_{14}a_{18}+2a_{13}`$ (B48)
$`h_A^{pK}=a_3a_4+\sqrt{3}(a_7a_8)+a_9+a_{10}+a_{11}+a_{12}+{\displaystyle \frac{1}{3}}(a_{16}+a_{14}a_{18}+a_{13}a_{19})`$ (B49)
$`h_A^{n\pi }=2a_3{\displaystyle \frac{4}{3}}a_{16}+{\displaystyle \frac{2}{3}}a_{14}+a_{18}+2a_{13}`$ (B50)
$`h_A^{nK}=a_4+\sqrt{3}a_8+a_9+{\displaystyle \frac{5}{2}}a_{10}2a_{11}a_{15}+a_{14}+{\displaystyle \frac{1}{3}}(a_{18}+a_{19}+a_{13})`$ (B51)
Note only $`a_{18}`$ contributes to $`R_A`$ in the parity violating two pions vertices. In the two kaons vertices $`a_{34},a_{78},a_{1019}`$ all lead to nonzero contribution to $`R_A`$.
## C $`\mathrm{\Delta }`$ Intermediate States and EM insertions
As noted in Section IV, the amplitudes of Figures 4-6 vanish through $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. Below, we briefly summarize the reasons behind this result.
### 1 PV $`\pi \mathrm{\Delta }N`$ contribution
In the case where the $`\mathrm{\Delta }`$ enters as an intermediate state we have the Feynman diagrams shown in Figure 4. Since the final and initial states are both nucleons, the two-pion parity violating vertices in Eqs. (36)-(45) arise first at two-loop order and contribute to the nucleon anapole moment at the order of $`O(1/\mathrm{\Lambda }_\chi ^3)`$. Although the PV $`N\mathrm{\Delta }\pi `$ interactions nominally contribute at lower order, in this case such contributions vanish up to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. The reason is as follows. Each of the parity violating and CP conserving single pion vertices has the same Lorentz structure—$`i\gamma _5`$. In the heavy baryon expansion, the relevant vertices are obtained by the substitution $`P_+i\gamma _5P_+`$, which vanishes. The leading nonzero contribution arises at first order in the $`1/m_N`$ expansion. Consequently, its contribution to the nucleon anapole moment appears only at $`𝒪(1/\mathrm{\Lambda }_\chi ^2m_N)`$, and since in this work we truncate at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$, the PV $`\pi \mathrm{\Delta }N`$ vertices do not contribute.
### 2 Magnetic moment insertions
The nucleon has a large isovector magnetic moment. We thus consider associated possible PV chiral loop corrections which lead to a nucleon anapole moment. The relevant diagrams are shown in Figure 5. At $`O(1/\mathrm{\Lambda }_\chi ^2)`$ there are only four relevant diagrams Figures 5a-d. Since the magnetic moment is of $`O(1/\mathrm{\Lambda }_\chi )`$ and the strong pion baryon vertex is of $`O(1/F_\pi )`$, the remaining PV vertex must be a Yukawa coupling if the loop is to contribute at $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$ or lower. For the nucleon magnetic moment insertion we have, for example,
$$iM_{5a}+iM_{5b}=iϵ^{\mu \nu \alpha \beta }\epsilon _\mu q_\nu v_\alpha \frac{\sqrt{2}g_Ae\mu _Nh_{\pi N}}{m_NF_\pi }[S_\beta ,S_\sigma ]_+\frac{d^Dk}{(2\pi )^D}\frac{k^\sigma }{vk}\frac{1}{v(q+k)}\frac{1}{k^2m_\pi ^2+iϵ},$$
(C1)
where $`e_\mu `$ is the photon polarization vector and $`\mu _N`$ is the nucleon magnetic moment. The denominator of the integrand in (C1) is nearly the same as for $`M_{3e}`$. The numerator contains a single factor of $`Sk`$. Hence, Figures 5a and 5b vanish for the same reason as does $`M_{3e}`$.
For the nucleon delta transition magnetic moment insertion we have
$`iM_{5c}+iM_{5d}`$ $`=`$ $`{\displaystyle \frac{2}{\sqrt{3}}}{\displaystyle \frac{g_{\pi N\mathrm{\Delta }}e\mu _{\mathrm{\Delta }N}h_{\pi N}}{m_NF_\pi }}(q_\sigma ϵ_\nu ϵ_\sigma q_\nu )[P_{3/2}^{\mu \nu }S^\sigma +S^\sigma P_{3/2}^{\nu \mu }]`$ (C3)
$`\times {\displaystyle }{\displaystyle \frac{d^Dk}{(2\pi )^D}}{\displaystyle \frac{k_\mu }{vk}}{\displaystyle \frac{1}{v(q+k)}}{\displaystyle \frac{1}{k^2m_\pi ^2+iϵ}},`$
where $`\mu _{\mathrm{\Delta }N}`$ is the nucleon delta transition magnetic moment and $`P_{3/2}^{\mu \nu }=g^{\mu \nu }v^\mu v^\nu +\frac{4}{3}S^\mu S^\nu `$ is the spin $`\frac{3}{2}`$ projection operator in the heavy baryon chiral perturbation framework. Since the integrand is the same as in $`M_{3e}`$ the integral is proportional to $`v_\mu `$. Moreover, $`v_\mu P_{3/2}^{\mu \nu }=0`$, so that $`M_{5c}+M_{5d}=0`$. Finally, the $`\mathrm{\Delta }`$ magnetic insertions of Figures 5e-h require the PV $`N\mathrm{\Delta }\pi `$ vertex, which starts off at $`𝒪(1/m_NF_\pi )`$. Thus, the latter do not contribute up to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$. In short, none of the magnetic insertions contribute at the order to which we work in this analysis.
### 3 PV electromagnetic insertions
Another possible source to the nucleon anapole moment arises from the PV magnetic moment like insertions as shown in Figure 6. All three PV $`\gamma NN`$ vertices $`c_{13}`$ in Eq. (34) and PV $`\gamma \mathrm{\Delta }N`$ vertices $`d_{46}`$ in Eq. (49) start off with one pion, so they are of order $`𝒪(1/\mathrm{\Lambda }_\chi F_\pi )`$. Vertices $`d_{7,8}`$ have two pions and are order $`𝒪(1/\mathrm{\Lambda }_\chi F_\pi ^2)`$. The corresponding contributions to the nucleon anapole moment appear at order of $`𝒪(1/\mathrm{\Lambda }_\chi ^3)`$ or $`𝒪(1/\mathrm{\Lambda }_\chi ^4)`$, respectively. The leading PV $`\gamma \mathrm{\Delta }N`$ vertices $`d_{13}`$ do not have pions and are of the order $`𝒪(1/\mathrm{\Lambda }_\chi )`$. In our case, however, the final and initial states are both nucleons. The $`\mathrm{\Delta }`$ can appear as the intermediate state inside the chiral loop, which leads to an additional factor $`1/F_\pi ^2`$ from two strong vertices. In the end $`d_{13}`$ contributes to the nucleon anapole moment at $`𝒪(1/\mathrm{\Lambda }_\chi ^3)`$. Finally, the PV $`\gamma \mathrm{\Delta }\mathrm{\Delta }`$ vertices contain one $`\pi `$. Since the $`\mathrm{\Delta }`$ can only appear as an intermediate state, this vertex contributes at two-loop order and is of higher-order in chiral counting than we consider here (the corresponding diagrams are not shown). Thus, to $`𝒪(1/\mathrm{\Lambda }_\chi ^2)`$, the PV electromagnetic insertions do not contribute.
Figure Captions
FIG 1. Axial vector $`\gamma NN`$ coupling, generated by PV hadronic interactions
Figure 2. Feynman diagrams for polarized electron nucleon scattering. Figure 2a gives tree-level $`Z^0`$-exchange amplitude, while FIG. 2b gives the anapole moment contribution. The dark circle indicates an axial vector coupling.
Figure 3. Meson-nucleon intermediate state contributions to the nucleon anapole moment. The shaded circle denotes the PV vertex. The solid, dashed and curly lines correspond to the nucleon, pion and photon respectively. For the $`SU(3)`$ case the intermediate states can also be hyperons and kaons.
Figure 4. The contribution to the nucleon anapole moment from PV $`\pi \mathrm{\Delta }N`$ vertices. The double line is the $`\mathrm{\Delta }`$ intermediate state.
Figure 5. Anapole moment contributions generated by insertions of the baryon magnetic moment operator, denoted by the cross, and the PV hadronic couplings, denoted by the shaded circle.
Figure 6. PV electromagnetic insertions, denoted by the overlapping cross and shaded circle.
Figure 7. Vector meson contribution to the anapole moment. Shaded circle indicates PV hadronic coupling.
Figure 8. Contributions to (a) PV NN interaction and (b) PV two-body current generated by the vector terms in Eqs. (27-33).
|
warning/0002/cond-mat0002313.html
|
ar5iv
|
text
|
# Steady-state mode III cracks in a viscoelastic lattice model
## I Introduction
Recent experiments on dynamical fracture (Fineberg, et al., 1991, 1992) have shown that cracks in brittle materials become unstable above a critical velocity. This instability is associated with the formation of a roughened fracture surface, with an increased dissipation of elastic energy, and with complicated velocity dynamics - for a review, see (Fineberg and Marder, 1999). Although there are hints of such an instability in the traditional continuum formulations of ideally brittle cracks (Yoffe, 1951), a systematic treatment does not appear possible. Indeed, recent attempts (Langer and Lobkovsky, 1998) to utilize a cohesive stress modification (Barenblatt, 1959) of the continuum elastic equations in order to address this problem have been rather unsuccessful.
In the absence of a compelling continuum approach, one must deal in some manner with discrete dynamics on the “atomic scale”. In this regard, Slepyan (Slepyan, 1981, 1982; Kulamekhtova, 1984) pioneered the idea of studying lattice models in which atoms interact (via piece-wise linear springs) with their neighbors in a predetermined lattice geometry. This conceptual framework was further developed by Marder and collaborators (Marder and Gross, 1995; Marder and Liu, 1993). Here, one can directly obtain the relationship between the crack tip velocity and the imposed driving and furthermore one can search for conditions which do not allow stable steady crack motion. Clearly, lattice models are not fully realistic, especially at displacements that are not small compared to the underlying lattice spacing; to be more realistic, one must resort to molecular dynamics simulations including all possible atomic interactions (Abraham, et al., 1994; Zhou, et al., 1997; Gumbsch, et al., 1997; Holland and Marder, 1997, 1998; Omeltchenko, et al., 1997). However, the analytic tractability of these models as well as indications (Marder and Gross, 1995; Pla, et al., 1998; Sander and Ghasias, 1999) that they do contain the essential mechanism responsible for the observed dynamical instability make them well worthy of serious attention.
Slepyan recognized that in an ideally brittle material (where each spring is linear until a displacement at which it completely breaks) one can use Fourier methods to find the lattice analog of a traveling wave solution. In this solution, each point on the lattice (at the same transverse coordinate) undergoes the same time history of motion as any other, albeit with some time delay. Once one obtains the solution, one must check that all springs assumed to be linear have displacements that are in fact below the breaking threshold. The violation of this assumption by the steadily propagating solution signals the onset of more complex dynamical behavior, at least in qualitative accord to what is seen experimentally (Fineberg and Marder, 1999; Marder and Gross, 1995). Using the Wiener-Hopf technique, Slepyan solved for the velocity-driving curve in the limit where the width of the lattice transverse to the crack direction goes to infinity.
One important consideration in all models of brittle fracture concerns dissipation mechanisms. From our perspective, it makes sense to put in dissipation at the lattice scale in such a way so as not dominate the large-scale continuum elastic field. If in addition one demands an equation that is local in time, one is led (Langer, 1992; Pla, et al., 1998) to the introduction of a Kelvin viscosity which dissipates energy proportional to the rate of change of spring lengths. In the naive continuum limit, this gives rise to a third derivative (two space, one time) term. That means that if the viscosity is chosen to be O(1) on the lattice scale, it will scale to zero as far as the macroscopic dynamics is concerned. We saw this explicitly in a previous set of papers (Kessler and Levine, 1998; Kessler, 2000) which solved this problem for finite width lattices and considered the nature of the solution as the width became large. Nevertheless, the viscosity can have a considerable effect on aspects of the solution which depend on the microscopic details, namely the crack speed (for a fixed stress intensity factor) and the self-consistency of the traveling wave ansatz.
The rest of this paper is organized as follows. In the next section, we formally define the model, employing Slepyan’s idea (Slepyan, 1981; Kulamekhtova, 1984) of replacing the driving at some external boundary with some local forcing on the crack surface. This allows for the solution in Section III of the strictly infinite lattice model; the connection to a finite problem with some displacement driving is made by appealing to the universality of the microscopic crack solution given a fixed stress intensity factor. In the following section, we discuss the numerical evaluation of the velocity curve from its formal expression. Following that, we turn to the aforementioned self-consistency condition and study the effect of dissipation on the critical velocity. We conclude with some observations in the final section.
## II Square lattice model
We wish to study mode III cracks and the effects thereupon of Kelvin viscosity. Our model thus consists of a square lattice of mass points undergoing (scalar) displacements out of the plane. The lattice extends infinitely long in both the $`x`$ (along-crack) and $`y`$ (transverse) directions. The lattice points are connected to their nearest neighbors by ideally brittle springs which behave linearly with spring constant 1 until some threshold elongation $`2ϵ`$ at which point they irreversibly crack. All the (uncracked) springs have a viscous damping $`\eta `$. Let us assume that the crack corresponds to a sequential breaking of the vertical bonds between the mass points at rows $`y=1`$ and $`y=0`$. The equation of motion for the masses in rows $`y>1`$ reads
$$\ddot{u}(x,y)=\left(1+\eta \frac{d}{dt}\right)\left(u(x+1,y)+u(x1,y)+u(x,y+1)+u(x,y1)4u(x,y)\right).$$
(1)
We allow the vertical bonds along the crack surface to have a different spring constant $`k`$ and damping parameter $`\stackrel{~}{\eta }`$, to model the effect of having weak links along the crack surface. The equation for $`y=1`$ then reads, using the assumed symmetry $`u(x,y)=u(x,1y)`$,
$`\ddot{u}(x,y)`$ $`=`$ $`\left(1+\eta {\displaystyle \frac{}{t}}\right)\left(u(x+1,1)+u(x1,1)+u(x,2)3u(x,1)\right)`$ (2)
$`+\theta (ϵu(x,1))\left(k+\stackrel{~}{\eta }{\displaystyle \frac{}{t}}\right)\left(2u(x,1)\right).`$
It is easily verified that the equations for $`y0`$ are consistent with the above symmetry. We can, without loss of generality choose $`ϵ=1/2`$ (so that the threshold elongation is unity). Note that in these units, the elastic wave speed is unity, so all velocities are dimensionless, expressed as fractions of the wave speed.
We are interested in steady-state cracks, described by the Slepyan traveling wave ansatz,
$$u(x,y,t)=u(xvt,y)$$
(3)
which implies that every mass point in a given row undergoes the same time history, translated in time. We choose the origin of time such that $`u(0,1)=ϵ`$, so that it represents the moment of cracking of the spring in column 0. If we define $`\tau =xvt`$, the equations of motion for $`y>1`$ become
$`(1\eta v{\displaystyle \frac{}{\tau }})(u(\tau +1,y)`$ $`+`$ $`u(\tau 1,y)+u(\tau ,y1)`$ (4)
$`+`$ $`u(\tau ,y1)4u(\tau ,y))v^2{\displaystyle \frac{^2u(\tau ,y)}{\tau ^2}}=0.`$
For $`y=1`$, we separate out the terms proportional to $`\theta (\tau )`$, giving
$`(1\eta v{\displaystyle \frac{}{\tau }})(u(\tau +1,1)+u(\tau 1,1)`$ $`+`$ $`u(\tau ,2)3u(\tau ,1))`$
$`+\left(k\stackrel{~}{\eta }v{\displaystyle \frac{}{\tau }}\right)\left(2u(\tau ,1)\right)`$ $``$ $`v^2{\displaystyle \frac{^2u(\tau ,1)}{\tau ^2}}=\sigma (\tau ).`$ (5)
The driving term for this equation has the form
$$\sigma (\tau )=\theta (\tau )\left[F_0+2\left(k\stackrel{~}{\eta }v\frac{}{\tau }\right)u(\tau ,1)\right].$$
(6)
In this expression, we have inserted an external force $`F_0`$ which acts on the crack boundaries. This force is an artifice which serves as an inhomogeneous source term, in place of introducing the driving through the boundary condition on the top and bottom surfaces. We will see later how to choose this function.
We Fourier transform with respect to $`\tau `$ (see the appendix for our conventions regarding Fourier transforms and Fourier integrals). If we assume that $`u(q,y)=\chi _q[\xi (q)]^{y1}`$ for $`y1`$, we obtain from Eqn. (4) the dispersion relationship
$$(1+iq\eta v)(e^{iq}+e^{iq}+\xi +\frac{1}{\xi }4)+q^2v^2=0$$
(7)
This can be rewritten as
$$(\xi ^{1/2}\frac{1}{\xi ^{1/2}})^2=4\mathrm{sin}^2\frac{q}{2}\frac{q^2v^2}{1+iq\eta v}h^2$$
(8)
which therefore implies that $`\xi ^{1/2}=\pm h/2\pm \sqrt{(h/2)^2+1}`$. If we define
$$r^2h^2+4$$
then we can write
$$\xi =\frac{(rh)^2}{4}=1\frac{h(hr)}{2}=\frac{rh}{r+h}$$
(9)
Note that we have must choose the sign that corresponds to a decaying solution in the $`y`$ direction, i.e. $`|\xi |<1`$.
If we substitute this ansatz into Eqn. (II), we get
$$\left(2(k+iq\stackrel{~}{\eta }v)+(1+iq\eta v)(1\frac{1}{\xi })\right)\chi =\sigma ^F$$
(10)
where $`\sigma ^F`$ is the Fourier transform of $`\sigma `$. Noting that $`1\frac{1}{\xi }=\frac{r(r+h)}{2}`$, we obtain
$$\chi _q=\frac{2\sigma ^F}{N(q)}$$
(11)
with
$$N(q)=h(r+h)(1+i\eta vq)+4(k+i\stackrel{~}{\eta }vq).$$
(12)
This equation is implicit, since the driving term $`\sigma ^F`$ on the right hand side still depends on the displacement field $`u(\tau ,1)`$. In the next section, we solve this equation via the Wiener-Hopf technique.
## III Wiener-Hopf solution
Let us define $`(\tau )=2u(\tau ,1)`$, the bond elongation between rows $`y=0`$ and $`y=1`$. In the appendix, we define $`^+(q)`$ and $`^{}(q)`$ as the decomposition of $`(\tau )`$ into terms which are analytic in the upper and lower half planes respectively. This allows us to write Eq. (6) as
$$\sigma ^F(q)=F_0^{}(q)+(k+i\stackrel{~}{\eta }vq)^{}(q)\stackrel{~}{\eta }v(0)$$
(13)
where we have employed a similar breakup for the driving term $`F_0`$, and where we have utilized Eq. (62).
Substituting the formula for $`\sigma ^F`$ into Eqn. (10) for $`\chi =(^++^{})/2`$ leads to a closed form equation for $``$; namely
$$^++^{}(1\frac{4(k+i\stackrel{~}{\eta }vq)}{N(q)})=\frac{4F_0^{}}{N(q)}\frac{4\stackrel{~}{\eta }v(0)}{N(q)}.$$
(14)
Defining $`S`$ by
$$S1\frac{4(k+i\stackrel{~}{\eta }vq)}{N(q)}=\frac{h(1+i\eta vq)}{h(1+i\eta vq)+(rh)(k+i\stackrel{~}{\eta }vq)},$$
(15)
we can rewrite Eq. (14) as
$$^++S^{}+\frac{\stackrel{~}{\eta }v}{k+iq\stackrel{~}{\eta }v}(1S)(0)=\frac{(1S)F_0^{}}{k+iq\stackrel{~}{\eta }v}.$$
(16)
We will solve this equation using the Wiener-Hopf technique. The key is to factor $`S`$ into a product of two pieces, $`S(q)=S^+(q)S^{}(q)`$, each of which is regular in the upper ($`S^+`$) or lower ($`S^{}`$) half plane. Doing so, we get
$$\frac{^+}{S^+}+S^{}^{}+\frac{\eta v}{k+iq\stackrel{~}{\eta }v}(\frac{1}{S^+}S^{})(0)=\frac{1S}{S^+}\frac{F_0^{}}{k+iq\stackrel{~}{\eta }v}.$$
(17)
We then have to deal with the extra singularity at $`q=\frac{ik}{\eta v}`$. We do this by a simple subtraction
$$\frac{^+}{S^+}+S^{}^{}+\frac{\stackrel{~}{\eta }v(0)}{k+iq\stackrel{~}{\eta }v}(\frac{1}{S^+}\frac{1}{S^+}|_{q=\frac{ik}{\stackrel{~}{\eta }v}}+\frac{1}{S^+}|_{q=\frac{ik}{\stackrel{~}{\eta }v}}S^{})=\frac{1S}{S^+}\frac{F_0^{}}{k+iq\stackrel{~}{\eta }v}.$$
(18)
We have now to choose $`F_0`$, our external forcing. In taking the width to infinity, in essence we are solving for the “inner” solution in the sense of boundary-layer theory; i.e., only on the scale of the lattice spacing. The external forcing in the finite width problem only acts on the large scale, and does not vary on the lattice scale. Thus, $`F_0`$ must have support only at $`q=0`$. Following Slepyan, then, we choose $`F_0`$ such that
$$\frac{1S}{S^+(k+iq\stackrel{~}{\eta }v)}F_0^{}=2\pi B\delta (q)=B\left(\frac{1}{iq+0}+\frac{1}{iq+0}\right),$$
(19)
where $`B`$ is a constant. This then gives us the the two separate equations
$`^+`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\eta }v(0)}{k+iq\stackrel{~}{\eta }v}}\left(1{\displaystyle \frac{S^+}{S^+|_{q=\frac{ik}{\stackrel{~}{\eta }v}}}}\right)+{\displaystyle \frac{BS^+}{iq+0}},`$ (20a)
$`^{}`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{\eta }v(0)}{k+iq\stackrel{~}{\eta }v}}\left({\displaystyle \frac{1}{S^{}S^+|_{q=\frac{ik}{\eta v}}}}1\right)+{\displaystyle \frac{B}{(iq+0)S^{}}}.`$ (20b)
As shown in Eqs. (59) and (61) we can solve for $`(0)`$ from Eqs. (20a) or (20b) by multiplying the first by $`iq+\mathrm{}`$ or the second by $`iq+\mathrm{}`$. This yields
$$(0)=BS^+|_{q=\frac{ik}{\stackrel{~}{\eta }v}}.$$
(21)
Substituting this back into Eqns. (20), and defining $`\eta _k\stackrel{~}{\eta }/k`$, we find
$`^+`$ $`=`$ $`{\displaystyle \frac{BS^+}{(iq+0)(1+iq\eta _kv)}}{\displaystyle \frac{\eta _kvB}{1+iq\eta _kv}}S^+|_{q=\frac{i}{\eta _kv}},`$ (22a)
$`^{}`$ $`=`$ $`{\displaystyle \frac{B}{(iq+0)(1+iq\eta _kv)S^{}}}+{\displaystyle \frac{\eta _kvB}{1+iq\eta _kv}}S^+|_{q=\frac{i}{\eta _kv}}S^{}`$ (22b)
These equations directly determine the displacements given the strength of the external driving, $`B`$. To get a physical handle on the meaning of this constant, we note that Eqns. (20) can be directly solved for the leading behavior of $`^\pm `$ as $`q`$ goes to $`0`$, using the fact that in this limit,
$$S^\pm \kappa ^{\pm 1}\sqrt{A}(q\pm i0)^{1/2}$$
(23)
where $`A=\frac{\sqrt{1v^2}}{2k}`$, and $`\kappa `$ is a constant, reflecting the arbitrariness in how we perform the decomposition of $`S`$. We find
$`_+\kappa \sqrt{A}{\displaystyle \frac{i^{1/2}B}{(iq+0)^{1/2}}}`$ (24a)
$`_{}{\displaystyle \frac{\kappa }{\sqrt{A}}}{\displaystyle \frac{i^{1/2}B}{(iq+0)^{3/2}}}`$ (24b)
This is just the expected (Fourier transform of) the singular solution of continuum elasticity as one approaches the crack tip. Thus, as in all boundary-layer problems, the large-distance limit of the “inner” solution matches on to the short-distance limit of the “outer” solution, confirming the justification of the choice of $`F_0`$ above. We need to relate the strength of the short-distance singularity of the continuum solution to the driving displacement $`\mathrm{\Delta }`$. This can be done directly by solving the continuum elastic equations. It is easier, however, to follow Slepyan and derive this relation by energy considerations. One can calculate (Slepyan, 1981) the flux of energy into the boundary-layer, or “process” zone, obtaining
$$T=\frac{ik\kappa ^2B^2}{2}v$$
(25)
Now consider the flux of energy $`T_0`$ being used to break bonds, which is simply related to $`(0)`$
$$T_0=\frac{k\left(BS^+|_{q=\frac{i}{\eta _kv}}\right)^2}{2}v.$$
(26)
Then, using the idea that the Griffith’s displacement is determined by exactly the condition that all the energy flux is needed just to break the bonds, we have
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\sqrt{\frac{T}{T_0}}=\frac{i^{1/2}\kappa }{S^+|_{q=\frac{i}{\eta _kv}}}$$
(27)
Notice that the factor of $`\kappa `$ guarantees that the result is invariant with respect to how the decomposition of $`S`$ is performed.
In order to make contact with other treatments, we consider the standard case $`k=1`$ and $`\eta _k=\stackrel{~}{\eta }=\eta `$, in which all bonds are equivalent. In this case, the function $`S`$ has a particularly simple form,
$$S(q)=\frac{h}{r}=\frac{H}{R}$$
(28)
with $`H^2=4(1+iq\eta v)\mathrm{sin}^2\frac{q}{2}q^2v^2`$ and $`R^2=4(1+iq\eta v)(\mathrm{sin}^2\frac{q}{2}+1)q^2v^2`$. We can represent these functions in term of the products over all their roots
$`H`$ $`=`$ $`\sqrt{1v^2}(q+i0)^{1/2}(qi0)^{1/2}{\displaystyle \underset{n,m}{}}\left(1{\displaystyle \frac{q}{Q_n^+}}\right)^{1/2}\left(1{\displaystyle \frac{q}{Q_m^{}}}\right)^{1/2}`$ (29)
$`R`$ $`=`$ $`2{\displaystyle \underset{n,m}{}}\left(1{\displaystyle \frac{q}{q_n^+}}\right)^{1/2}\left(1{\displaystyle \frac{q}{q_m^{}}}\right)^{1/2}`$ (30)
where $`Q_n^+(q_n^+)`$ are the (nonzero) roots of $`H^2`$($`R^2`$) in the lower half plane and $`Q_n^{}(q_n^{})`$ are corresponding roots of $`H^2`$($`R^2`$) in the upper half plane. This decomposition allows us to write down explicitly the factors $`S^+`$, $`S^{}`$:
$`S^+(q)={\displaystyle \frac{(1v^2)^{1/4}}{\sqrt{2}}}(q+i0)^{1/2}{\displaystyle \underset{n,m}{}}\left({\displaystyle \frac{1\frac{q}{Q_n^+}}{1\frac{q}{q_m^+}}}\right)^{1/2}`$ (31a)
$`S^{}(q)={\displaystyle \frac{(1v^2)^{1/4}}{\sqrt{2}}}(qi0)^{1/2}{\displaystyle \underset{n,m}{}}\left({\displaystyle \frac{1\frac{q}{Q_n^{}}}{1\frac{q}{q_m^{}}}}\right)^{1/2}`$ (31b)
Examining the small $`q`$ behavior, we find the $`\kappa =1`$, so that substitution of $`S^+`$ into Eqn. (27) gives
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\frac{(2\eta v)^{1/2}}{(1v^2)^{1/4}}\underset{n,m}{}\left(\frac{1\frac{i}{\eta vq_n^+}}{1\frac{i}{\eta vQ_m^+}}\right)^{1/2}$$
(32)
If we explicitly take out of the product in Eq. (32) the one imaginary $`q^+`$, which we will denote as $`q_0^+`$, we get
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\frac{1}{(1v^2)^{1/4}}\sqrt{\frac{2(1+iv\eta q_0^+)}{iq_0^+}}\underset{n0,m}{}\left(\frac{1+iv\eta q_n^+}{1+iv\eta Q_m^+}\frac{Q_m^+}{q_n^+}\right)^{1/2}$$
(33)
This formula was obtained by Kessler (2000), who started from a finite lattice of transverse size $`N_y`$ and considered the $`N_y\mathrm{}`$ limit. Note that this can be done for $`k=1`$ on a square lattice, (the case being considered here), but not for general $`k`$ or for modes I or III on a triangular lattice; the method here works for these cases as well (Pechenik, et al., 2000).
Next, it is easy to see that in the limit $`\eta 0^+`$, we have $`S|_{q=\frac{i}{\eta v}}=1`$ and that therefore
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\left(i\frac{S^{}}{S^+}|_{q=\frac{i}{\eta v}}\right)^{1/2}=i^{1/2}\underset{n,m}{}\left(\frac{q_m^{}Q_n^+}{q_m^+Q_n^{}}\right)^{1/4}.$$
(34)
Furthermore, in this limit $`q_0^+=q_0^{}`$. All other complex roots cancel out, leaving only real roots. Hence,
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\underset{overrealQ_n,q_m}{}\left(\frac{q_m^{}Q_n^+}{q_m^+Q_n^{}}\right)^{1/4},$$
(35)
a result originally obtained by Slepyan. Note that in our treatment, the presence of one of these real roots in either the numerator or the denominator of the product expression depends entirely on the half-plane from where it came as we took the limit $`\eta 0^+`$. To see what this condition implies, let us denote by $`H^2`$(or $`R^2`$) as $`f`$; then the small imaginary part of a specific real root, $`i\delta `$, satisfies the equation
$$i\delta \frac{f}{q}|_{\eta =0,q=q_{real}}+\eta \frac{f}{\eta }|_{\eta =0,q=q_{real}}=0,$$
(36)
which gives as the determining factor
$$\frac{\delta }{\eta }=\frac{q^3v^3}{\frac{}{q}(4\mathrm{sin}^2\frac{q}{2}v^2q^2)}|_{q=q_{real}}.$$
(37)
This condition is exactly what appears in Slepyan’s work.
## IV Numerical calculations
In this section we derive an integral form for the basic result Eq. (32). We then present a numerical procedure for finding the displacement-velocity curve. The basic notion is to utilize Eq. (VII) to find $`S^+`$. It is more convenient for the numerical work to isolate explicitly the singularity at $`q=0`$, defining $`K(q)`$ by
$$S^2(q)=\frac{A^2\varphi ^2(q+i0)(qi0)}{q^2+\varphi ^2}K(q)$$
(38)
where $`\varphi `$ is an arbitrary positive constant. Because of the factor $`A^2q^2`$ in the denominator, $`K(0)=1`$; the factor $`(q^2+\varphi ^2)/\varphi ^2`$ does not alter the desired asymptotic behavior at infinity: $`Kconst`$ as $`q\pm \mathrm{}`$. Applying Eq. (VII), we get for $`Imq>0`$
$$S^+=\mathrm{exp}\frac{1}{4\pi i}_{\mathrm{}}^+\mathrm{}\frac{\mathrm{ln}S^2(\gamma )}{\gamma q}𝑑\gamma =\left(A\varphi \frac{q+i0}{q+i\varphi }\right)^{\frac{1}{2}}\left(K^+\right)^{\frac{1}{2}}$$
(39)
and for $`Imq<0`$
$$\frac{1}{S^{}}=\mathrm{exp}\frac{1}{4\pi i}_{\mathrm{}}^+\mathrm{}\frac{\mathrm{ln}S^2(\gamma )}{\gamma q}𝑑\gamma =\left(\frac{1}{A\varphi }\frac{qi\varphi }{qi0}\right)^{\frac{1}{2}}\left(K^{}\right)^{\frac{1}{2}}$$
(40)
with
$$\left(K^\pm \right)^{\pm \frac{1}{2}}=\mathrm{exp}\frac{1}{4\pi i}_{\mathrm{}}^+\mathrm{}\frac{\mathrm{ln}K(\gamma )}{\gamma q}𝑑\gamma $$
(41)
where the upper sign is for $`Imq>0`$ and the lower sign for $`Imq<0`$.
There is one point regarding these expressions that must be clarified. We must make the correct choice of the branch cuts for the square roots in $`h`$ and $`r`$ appearing in $`S`$. It is convenient to express things in terms of $`S^{}(q)h/r=H/R`$, the simple $`k=1`$ and $`\eta _k=\eta `$, form of $`S`$, so that from Eqn. (15)
$$S=\frac{S^{}(1+i\eta vq)}{S^{}(1+i\eta vq)+(1S^{})k(1+i\eta _kvq)},$$
(42)
and from Eqn. (9)
$$\xi =\frac{1S^{}}{1+S^{}}.$$
(43)
We must choose $`S^{}`$ such that $`|\xi |<1`$ as noted below Eqn. (9). It is easy to see from the above expression for $`\xi `$ that it is sufficient to choose $`S^{}`$ to lie in the first and fourth quadrants to guarantee this. For $`S^{}`$, we can then generate $`S`$ directly via Eqn. (42), and from this $`K`$.
As $`K(0)=1`$, there is no pole in the integral (41) at $`\gamma =0`$ when $`q=0`$. This means that $`K^+(0)=(K^{}(0))^1`$ and the asymptotic expressions for $`S^\pm `$ are
$$S^+|_{q0}=\sqrt{\frac{AK^+(0)}{i}}(q+i0)^{\frac{1}{2}};S^{}|_{q0}=\sqrt{\frac{iA}{K^+(0)}}(qi0)^{\frac{1}{2}}.$$
(44)
Comparing this to Eqn. (23), we have
$$\kappa =\sqrt{iK^+(0)},$$
(45)
so that
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\sqrt{\frac{1+\varphi \eta _kv}{A\varphi }\frac{K^+(0)}{K^+(\frac{i}{\eta _kv})}}$$
(46)
Explicitly writing out the integrals, we get
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\sqrt{\frac{1+\varphi \eta _kv}{A\varphi }}\mathrm{exp}\frac{1}{4\pi i}_{\mathrm{}}^+\mathrm{}\frac{\mathrm{ln}K(\gamma )}{\gamma (1+i\eta _kv\gamma )}𝑑\gamma $$
(47)
The extra degree of convergence of the integrand at infinity is very desirable, as it helps control the fact that $`\mathrm{ln}K(\xi )`$ is a rapidly oscillating function. Also, the fact that the answer must be independent of $`\varphi `$ provides a check on the numerical routines. A change of variables $`\gamma \gamma `$ allows the integral to be rewritten as
$$\frac{\mathrm{\Delta }}{\mathrm{\Delta }_G}=\sqrt{\frac{1+\varphi \eta _kv}{A\varphi }}\mathrm{exp}\frac{1}{4\pi }_0^+\mathrm{}Im\frac{\mathrm{ln}K(\gamma )}{\gamma (1+i\eta _kv\gamma )}𝑑\gamma $$
(48)
To actually compute this integral, we divided the region of integration into two parts from $`0`$ to $`1`$ and from $`1`$ to $`+\mathrm{}`$. The second integral was further transformed so as to have in the numerator the factor $`\mathrm{ln}A^2\varphi ^2K(\xi )`$ which goes to zero at infinity; the subtracted integral with the numerator $`\mathrm{ln}1/(A\varphi )^2`$ was performed analytically. After all these manipulations, the integrals were successfully computed by standard mathematical library subroutines.
Results from our numerical calculations for $`k=1`$ and $`\eta _k=\stackrel{~}{\eta }=\eta `$, are presented in Figs. 1, 2, which reproduce, of course, those of (Kessler, 2000). At small $`\eta `$, there are large oscillations in the $`v`$ vs $`\mathrm{\Delta }`$ curve, as found initially by Marder and Gross for lattices with small transverse sizes. The curves all hit the $`v=0`$ line at the point which marks the end of the band of lattice-trapped static cracks (Kessler and Levine, 1998). For larger damping, the oscillations disappear and the moving crack branch bifurcates smoothly (in the backward direction) from this point, eventually turning around and becoming stable. In Fig. 3, we contrast the standard result to that for smaller $`k`$, keeping $`\eta _k=\eta `$. We see that the $`v=0`$ limit has moved closer to unity, in accord with the finding in (Kessler and Levine, 1998) that the window of arrested cracks narrows as $`k`$ decreases (note for purposes of comparison that our $`k`$ is a factor of 2 smaller than that defined in (Kessler and Levine, 1998). The velocity at the minimum $`\mathrm{\Delta }/\mathrm{\Delta }_G`$ however is hardly affected, so the minimum stable velocity is essentially unchanged. Past this point, the velocity is higher at smaller $`k`$, even when the change in $`\mathrm{\Delta }_G`$ is factored in. This is further evidence of how the microscopic details influence the velocity versus driving displacement curve.
## V Consistency of the solution
As mentioned in the introduction, cracks become unstable above a critical velocity. In the framework of the model being considered here, one might assume that this velocity is associated with the solution as found by the Wiener-Hopf method becoming inconsistent. As we shall see, there are two kinds of inconsistencies. The first sets in for small $`v`$, for small enough $`\eta `$. Here the vertical bond between $`y=0`$ and $`y=1`$ first achieves critical extension prematurely, at some positive $`\tau `$, contradicting the assumption that the bond breaks at $`\tau =0`$. In the second kind of inconsistency, a horizontal bond achieves critical extension. If the model is defined so that only the vertical bonds between $`y=0`$ and $`y=1`$ are breakable, then this does not present a problem. If one the other hand, all the springs are chosen identical, so that the horizontal springs also break at the same critical extension as the vertical ones, then the solution is indeed inconsistent at this point. This is what occurs at large enough velocity, for all $`\eta `$.
To proceed, we must calculate the bond lengths, back transforming our Fourier space solutions so as to obtain the physical displacements. We find the time dependence of the function $`(\tau )`$. From Eqs. (20a), (20b) and (21), we have
$$^+=\frac{(0)}{(iq+0)(1+iq\eta _kv)}\frac{S^+}{S^+(q)|_{q=\frac{i}{\eta _kv}}}\frac{\eta _kv(0)}{1+iq\eta _kv};$$
(49)
$$^{}=\frac{(0)}{S^+|_{q=\frac{i}{\eta _kv}}S^{}(q)}\frac{1}{(iq+0)(1+iq\eta _kv)}+\frac{\eta _kv(0)}{1+iq\eta _kv}.$$
(50)
Performing the inverse Fourier transform, we get
$$(\tau )=(0)_{\mathrm{}}^+\mathrm{}\frac{\sqrt{A\varphi }e^{iq\tau }}{\sqrt{(iq0)(iq\varphi )}(1+iq\eta _kv)}\frac{(K^+)^{\frac{1}{2}}}{S^+|_{q=\frac{i}{\eta _kv}}}\frac{dq}{2\pi }$$
(51)
for $`\tau >0`$ and
$$(\tau )=(0)_{\mathrm{}}^+\mathrm{}\frac{e^{iq\tau }}{\sqrt{A\varphi }(iq+0)(1+iq\eta _kv)}\left(\frac{iq+\varphi }{iq+0}\right)^{\frac{1}{2}}\frac{(K^{})^{\frac{1}{2}}}{S^+|_{q=\frac{i}{\eta _kv}}}\frac{dq}{2\pi }+(0)e^{\frac{\tau }{\eta _kv}}$$
(52)
for $`\tau <0`$. Here we have used the expressions for $`S^\pm `$ derived in the last section.
Clearly, these integrals must be done numerically. To proceed, we transform the pieces containing the factors $`K^\pm `$ by dividing both the numerator and denominator by $`\left(K^+(0)\right)^{1/2}`$. In the denominator, we get $`\mathrm{\Delta }_G/\mathrm{\Delta }`$; we have already discussed how this can be computed. We now have integrands containing factors of the form
$$\frac{\left(K^\pm (q)\right)^{\pm \frac{1}{2}}}{\left(K^+(0)\right)^{\frac{1}{2}}}=\mathrm{exp}\frac{1}{4\pi i}_{\mathrm{}}^+\mathrm{}\frac{q\mathrm{ln}K(\xi )}{\xi (\xi q)}𝑑\xi .$$
(53)
The $`\pm `$ refers on the right hand side to the sign of the imaginary part of $`q`$. As we will see later on, we need to calculate this function only for positive $`q`$. Then, the integral in the exponent can be written as
$$\frac{1}{2\pi i}_0^+\mathrm{}\frac{Re\mathrm{ln}K(q\xi )}{\xi ^2(1\pm i0)^2}𝑑\xi +\frac{1}{2\pi }_0^+\mathrm{}\frac{Im\mathrm{ln}K(q\xi )}{\xi (\xi ^2(1\pm i0)^2)}𝑑\xi .$$
(54)
We break up the region of integrations into three parts,$`(0,1/2)`$, $`(1/2,3/2)`$ and $`(3/2,+\mathrm{})`$. In the first interval, we numerically calculate the integral directly as written. For the third interval, we proceed as discussed in the last section for a similar semi-infinite integral; this leaves us with having to integrate numerically a function which behaves as $`1/\xi ^4`$ near infinity. Finally, for the integral near $`1`$ we add an integral with $`K(q\xi )`$ replaced by $`K(q)`$; this added integral is done analytically and the resultant subtracted integral with the integrand now containing $`\mathrm{ln}K(q\xi )/K(q)`$ is done numerically.
Applying the described procedure, we can evaluate the functions $`\left(K^\pm (q)\right)^{\pm \frac{1}{2}}/\left(K^+(0)\right)^{\frac{1}{2}}`$ for any real positive $`q`$. We must now perform the final integration over the variable $`q`$. First, we note that the requirement of having a real displacement necessitates $`^\pm (q)=^\pm (q)^{}`$. This allows us to change the region of integration to $`(0,+\mathrm{})`$ in (51); we will return shortly to 52. The trickiest part of this calculation concerns the behavior near $`q=0`$. For $`^+`$, there is a $`1/q^{1/2}`$ behavior. Here the leading order term can be integrated analytically over the interval $`(0,1)`$ and then the overall function with this term subtracted can be used for a numerical integration. This works in a straightforward manner.
We need to be more careful with the integral (52), because here the integrand behaves near zero as $`1/q^{3/2}`$. To proceed, we first divide the interval of integration into three parts: $`(\mathrm{},1)`$, $`(1,1)`$, and $`(1,+\mathrm{})`$. The integrals over $`(\mathrm{},1)`$ and $`(1,+\mathrm{})`$ can again be combined to yield an integral twice the real part of the integrand over the latter range. For the range spanning zero, we first subtract from the integrand the leading term
$$\frac{1}{\sqrt{A}S^+|_{q=\frac{i}{\eta _kv}}(iq+0)^{3/2}}.$$
After this subtraction, the integrand becomes of order $`1/q^{1/2}`$. Now, the one-sided integral converges, and thus we can again transform the range to $`(0,1)`$. This subtracted integral is evaluated numerically from a very small value of $`q`$ (typically $`q=10^3`$) to $`1`$. In the remaining part of the range, i.e. from $`0`$ to $`10^3`$, we evaluate the integral analytically after approximating it by its leading small $`q`$ behavior, $`\alpha /q^{1/2}`$; the value of the constant $`\alpha `$ is determined by fitting the the behavior of the function near the point $`10^3`$. Finally we need to perform analytically the integral of the subtracted piece $`1/(\sqrt{A}S^+|_{q=\frac{i}{\eta _kv}})/(iq+0)^{3/2}`$. To accomplish this, we deform the contour of integration from $`(1,1)`$ on the real axis to the curve from $`1`$ to $`1`$ going over the lower half of the unit circle. Note that the branch-cut for this function must be taken as before over negative real axis; this choice guarantees the fact that $`f(q)=f^{}(q)`$. Note that once all our transformations are complete, we only need values for the integrand for positive (real) values of $`q`$; this was used earlier in the technique for calculating $`K^\pm `$. A good check of this complex numerical technique of computing the Fourier transformation is applying it to the Fourier transform $`^+(q)`$ for $`\tau <0`$ and $`^{}(q)`$ for $`\tau >0`$, as in this case result must be equal to zero.
We first examine what happens in the standard case, $`k=1`$ and $`\eta _k=\stackrel{~}{\eta }=\eta `$. Figure 4 shows the behavior of $`(\tau )`$ for $`\eta =.2`$ and various values of the velocity. As discussed above, we see the two different kinds of inconsistencies exhibited by these solutions. For example, the solution for $`v=0.2`$ is inconsistent as $`(\tau )/(0)>1`$ for $`\tau >0`$. This type of inconsistency, seen at small velocities, corresponds to the region of backward behavior on $`v\mathrm{\Delta }`$ graph. As already claimed by Marder and Gross, this region is therefore unphysical.
More interesting is the inconsistency which sets in at large velocity, and has been argued to be related to the experimentally observed microbranching. What happens here is that the elongation of the horizontal bond, given by $`_{hor}(\tau )=((\tau 1)(\tau ))/2`$ increases as a function of increasing driving. At some critical speed it reaches the value $`(0)`$ and the bond breaks. Figure 5 shows the dependence of the critical speed on $`\eta `$. We note in passing that these graphs agree with what can be gotten by extrapolating a similar graph for the finite $`N_y`$ lattice, a graph that can be obtained using the methods of (Kessler and Levine, 1998). The actual location in $`\tau `$ at which the horizontal break occurs at the critical velocity is shown in Fig. 6.
Several details of our findings are worthy of note. First, there is always a critical speed for any dissipation, but this speed gets very close to the maximal crack speed for large $`\eta `$. The value of the critical speed for the dissipation-less limit is $`v_{cr}.725`$; this is above what has been seen in some experiments but recall that we are doing mode III on a square lattice, a far cry from mode I in an amorphous system. The is a curious break in the curve at $`\eta .665`$, which corresponds in Fig. 6 to the point where the break location varies most strongly with $`\eta `$.
It should be noted that this inconsistency does not appear related in any obvious way to the Yoffe (1951) criterion. First, the Yoffe criterion, concerning the direction of maximal stress, is calculated for the continuum elastic field, and is thus completely independent of the viscosity $`\eta `$ (Kessler and Levine, 1998). The onset of the inconsistency is strongly $`\eta `$ dependent however. Also, the Yoffe criterion is stated for Mode I cracks, and its applicability to Mode III is subject to question.
We turn now to the case of $`k<1`$, with weakened bonds along the crack path, again keeping $`\eta _k=\eta `$. Figures 5, 6 show our results. As is reasonable, the inconsistencies persist, but are pushed to higher speeds as $`k`$ decreases. It is interesting to speculate that since for finite width systems, velocities greater than the wave speed are possible, sufficiently small $`k`$ might enable stable steady-state fracture at supersonic velocities. Also, intersonic wave speeds are possible in the case of mode II fracture, and therefore weakened bonds on the path of the crack would in this case as well allow steady-state propagation at higher speeds (Rosakis, et al., 1999).
## VI Discussion
We have shown how to extend the Slepyan approach to cracks in ideally brittle infinite lattice systems to the case of having a Kelvin viscosity for each lattice spring. In addition, we have examined the onset of the additional, off-axis cracking at high velocity. We showed how dissipation delays but does not eliminate the microbranching instability, an effect impossible to see within the context of any continuum elastic treatment.
In the following work, we will present the results of similar calculations for mode I cracks on a triangular lattice. That system is much closer to those that have been studied experimentally and one can therefore hope for more direct lessons to emerge. Also, we should mention complementary studies on nonlinear lattice models which relax the assumption of ideally brittle springs and thereby allow one to view the onset of inconsistent behavior as the limiting case of more standard bifurcations (Kessler and Levine, 1999, 2000).
###### Acknowledgements.
DAK acknowledges the support of the Israel Science Foundation. The work of HL and LP is supported in part by the NSF, grant no. DMR94-15460. DAK and LP thank Prof. A. Chorin and the Lawrence Berkeley National Laboratory for their hospitality during the inital phase of this work.
## VII Appendix
We discuss here some of our conventions and notations regarding Fourier transformations. We use Fourier transformation in the form
$`u^F(q)={\displaystyle _{\mathrm{}}^+\mathrm{}}u(\tau )e^{iq\tau }𝑑\tau `$
$`u(\tau )={\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}u^F(q)e^{iq\tau }𝑑q`$ (55)
We define also $`u^+`$ and $`u^{}`$ as
$`u^+(q)`$ $`=`$ $`{\displaystyle _0^+\mathrm{}}u(\tau )e^{iq\tau }𝑑\tau `$ (56a)
$`u^{}(q)`$ $`=`$ $`{\displaystyle _{\mathrm{}}^0}u(\tau )e^{iq\tau }𝑑\tau `$ (56b)
This gives
$`u(\tau )\theta (\tau )`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}u^+(q)e^{iq\tau }𝑑q;`$ (57a)
$`u(\tau )\theta (\tau )`$ $`=`$ $`{\displaystyle \frac{1}{2\pi }}{\displaystyle _{\mathrm{}}^+\mathrm{}}u^{}(q)e^{iq\tau }𝑑q,`$ (57b)
which means that $`u^+`$ has poles only in the lower half plane and $`u^{}`$ only in the upper half plane.
Now let us define $`p=iq`$. Then
$$_0^+\mathrm{}u(\tau )e^{iq\tau }𝑑\tau \stackrel{_{p+\mathrm{}}}{}u(0)_0^+\mathrm{}e^{p\tau }𝑑\tau =\frac{1}{p}u(0).$$
(58)
Thus
$$\underset{iq+\mathrm{}}{lim}iqu^+=u(0).$$
(59)
Similarly we can get for $`p=iq`$
$$_{\mathrm{}}^0u(\tau )e^{iq\tau }𝑑\tau \stackrel{_{p+\mathrm{}}}{}u(0)_{\mathrm{}}^0e^{p\tau }𝑑\tau =\frac{1}{p}u(0).$$
(60)
So
$$\underset{iq+\mathrm{}}{lim}iqu^{}=u(0).$$
(61)
The Fourier transform of $`\theta (\tau )\frac{d}{d\tau }u(\tau )`$ is given by
$$_{\mathrm{}}^0\frac{du(\tau )}{d\tau }e^{iq\tau }𝑑\tau =u(0)iqu^{}$$
(62)
Finally, we consider separating an arbitrary function $`S(q)`$ into the product of two pieces $`S^+`$ and $`S^{}`$, with poles and zeroes in lower and upper half-planes respectively. These can be done using the identity,
$`{\displaystyle \frac{1}{2\pi i}}{\displaystyle _{\mathrm{}}^+\mathrm{}}{\displaystyle \frac{\mathrm{ln}S(\xi )}{\xi q}}𝑑\xi ={\displaystyle \frac{1}{2\pi i}}{\displaystyle _{\mathrm{}}^+\mathrm{}}\left({\displaystyle \frac{\mathrm{ln}S^+(\xi )}{\xi q}}+{\displaystyle \frac{\mathrm{ln}S^{}(\xi )}{\xi q}}\right)𝑑\xi `$
$`=\{\begin{array}{cc}forImq>0,\hfill & \mathrm{ln}S^+(q)\hfill \\ forImq<0,\hfill & \mathrm{ln}S^{}(q)\hfill \end{array}.`$ (65)
|
warning/0002/hep-th0002124.html
|
ar5iv
|
text
|
# Boundary States for a Free Boson Defined on Finite Geometries
## 1 Introduction
Renormalization transformations are often defined on spaces of fields or parameters of statistical physics models. It is assumed that the existence of a non-trivial fixed point of these transformations requires that the space it acts on be infinite-dimensional. Or at least that physical relevance of such fixed points stems from the infinite number of degrees of freedom. Langlands introduced a family of finite models inspired by percolation, each endowed with a renormalization transformation with a non-trivial fixed point. Numerical analysis shows that, already for the coarsest models in the family, the critical exponents bear some similarities with those accepted in the literature for percolation.
The calculation presented here is a step towards extending Langlands’ construction for percolation to other models with interaction. In Langlands calculated the partition function of the free boson on a cylinder with fixed boundary conditions. This partition function may be interpreted as the probability of measuring given restrictions of the boson at the cylinder extremities. Langlands argued that the space of such probability distributions might be a natural set upon which to construct a renormalization transformation. He showed moreover how to construct vectors $`|𝔵_B`$ in the Fock space that represent the restriction at a given extremity $`B`$ in such a way that the partition function is simply the expectation value of the evolution operator between the two boundary states:
$$Z(\phi |_{B_1},\phi |_{B_2})=𝔵_{B_1}|q^{L_0\overline{L}_0}|𝔵_{B_2}.$$
(1)
(The notation will be clarified in the following.) This is nothing but the Feynman-Kac formula. The present paper goes back to this formula to make Langland’s expression for the state $`|𝔵_B`$ more explicit. We should underline that it is a rather unusual use of the Feynman-Kac formula. While most of its applications involve the computation of a partition function from the evolution operator and boundary conditions, we use it as a tool to define boundary states in a chosen algebraic structure. The exercice is more than aesthetical; using this new expression we can show that the boundary states transform properly under conformal transformations that stabilize the boundary.
## 2 Notations
The description of free bosons is based on the Heisenberg algebra and its representations. The generators are the creation ($`𝔞_k,k>0`$), annihilation ($`𝔞_k,k>0`$) and central ($`𝔞_0`$) operators, obeying the commutation rule
$$[𝔞_n,𝔞_m]=n\delta _{n+m,0}.$$
(2)
The Fock space $`_\alpha `$ is a highest weight representation. The action of the generators on the highest weight vector $`|\alpha `$ is given by:
$`𝔞_k|\alpha `$ $`=0,k>0,`$ (3)
$`𝔞_0|\alpha `$ $`=\alpha |\alpha ,`$ (4)
and physical states are generated by polynomials in the $`𝔞_k,k>0`$. A basis for $`_\alpha `$ is given by the vectors
$$|\alpha ;n_1,n_2,\mathrm{}=𝔞_1^{n_1}𝔞_2^{n_2}\mathrm{}|\alpha ,$$
(5)
where the non-negative integers $`n_i`$’s are all zero but finitely many. The inner product on $`_\alpha `$ is defined by
$`\alpha ^{};n_1^{},n_2^{},\mathrm{}|\alpha ;n_1,n_2,\mathrm{}`$ $`=\alpha ^{}|\mathrm{}𝔞_3^{n_3^{}}𝔞_2^{n_2^{}}𝔞_1^{n_1^{}}𝔞_1^{n_1}𝔞_2^{n_2}𝔞_3^{n_3}\mathrm{}|\alpha `$
$`=\delta _{\alpha ,\alpha ^{}}\left({\displaystyle \underset{k=1}{\overset{\mathrm{}}{}}}k^{n_k}n_k!\delta _{n_k,n_k^{}}\right).`$ (6)
The elements (5) can thus be easily normalized to form an orthonormal basis.
The Hilbert space of the free boson is the direct sum of tensor products of the form $`_\alpha _{\overline{\alpha }}`$, $`_\alpha `$ and $`_{\overline{\alpha }}`$ characterizing modes in the holomorphic and anti-holomorphic sectors, respectively. States in these tensor products are generated by the action of polynomials in the $`𝔞_k`$ and $`\overline{𝔞}_k`$ on the highest weight vector $`|\alpha ;\overline{\alpha }=|\alpha |\overline{\alpha }`$. The generators $`𝔞_k`$ are understood to act as $`𝔞_k1`$ and the $`\overline{𝔞}_k`$ as $`1𝔞_k`$.
Fock spaces are given the structure of a Virasoro module by defining the conformal generators
$`L_n`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{m}{}}:𝔞_{nm}𝔞_m:n0`$
$`L_0`$ $`=`$ $`{\displaystyle \underset{n>0}{}}𝔞_n𝔞_n+{\displaystyle \frac{1}{2}}𝔞_0^2.`$ (7)
The expression for $`L_0`$ implies that $`_\alpha `$ is a highest weight module with highest weight $`\alpha ^2/2`$. As will be described in the next section, the boson field is to be compactified on a circle of radius $`R`$. The pairs ($`\alpha `$, $`\overline{\alpha }`$) are then restricted to take the values
$`\alpha =\alpha _{u,v}=\left({\displaystyle \frac{u}{2R}}+vR\right)`$ $`h_{u,v}={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{u}{2R}}+vR\right)^2`$ (8)
$`\overline{\alpha }=\overline{\alpha }_{u,v}=\alpha _{u,v}`$ $`\overline{h}_{u,v}={\displaystyle \frac{1}{2}}\left({\displaystyle \frac{u}{2R}}vR\right)^2,`$ (9)
with $`u`$ and $`v`$ integers and where $`h_{u,v}`$ and $`\overline{h}_{u,v}`$ are the values of $`L_0=L_01`$ and $`\overline{L}_0=1L_0`$ acting on $`_{\alpha _{u,v}}_{\overline{\alpha }_{u,v}}`$. We will denote $`_{\alpha _{u,v}}`$ by $`_{(u,v)}`$ and $`_{\overline{\alpha }_{u,v}}`$ by $`\overline{}_{(u,v)}`$.
In his calculation, Langlands chose the Virasoro algebra Vir as the fundamental structure. He was able to construct explicitly the map $`𝔵`$ for irreducible Verma modules over Vir. However Verma modules over Vir are reducible for some rational highest weights. (Rational compactification radii $`R`$ do lead to such highest weights.) It was his suggestion that we look for an alternative definition that would encompass reducible cases. Using the Heisenberg algebra as the basic structure avoids this difficulty and leads as well to an elegant form for $`𝔵`$.
## 3 Explicit calculation of the partition function
We identify the cylinder with the quotient of the infinite strip $`0<\text{Re}w<\mathrm{ln}q`$, $`0<q<1`$, by the translations $`ww+2\pi ik,k`$. It can be mapped on the annulus $`𝒜`$ of center 0, outer radius 1 and inner radius $`q`$ by the conformal map $`z=e^w`$. The angle $`\theta `$ of the annular geometry parametrizes both extremities of the cylinder.
The partition function is defined as
$$𝒟\phi e^{_𝒜(\phi )d^2z}$$
(10)
where $`_𝒜`$ denotes the integration over the annulus, and the Lagrangian density is given by
$$(\phi )=_z\phi _{\overline{z}}\phi .$$
(11)
The usual mode expansion of $`\phi (z,\overline{z})`$ is
$$\phi (z,\overline{z})=\phi _0+a\mathrm{ln}z+b\mathrm{ln}\overline{z}+\underset{n0}{}\left(\phi _nz^n+\overline{\phi }_n\overline{z}^n\right).$$
The restriction $`\phi _{B_1}`$ of this field to the inner circle where $`z=qe^{i\theta }`$ and $`\overline{z}=qe^{i\theta }`$ is of the form:
$$\phi _{B1}(\theta )=\phi _0+(a+b)\mathrm{ln}q+i\theta (ab)+\underset{k0}{}b_ke^{ik\theta },b_k=\overline{b}_k$$
and the restriction $`\phi _{B_2}`$ to the outer circle ($`z=e^{i\theta },\overline{z}=e^{i\theta }`$):
$$\phi _{B2}(\theta )=\phi _0+i\theta (ab)+\underset{k0}{}a_ke^{ik\theta },a_k=\overline{a}_k.$$
(The relationship between $`a_k,b_k`$ and $`\phi _n`$ will be given below.) Since it is the field $`e^{i\phi /R}`$ that really matters, $`\phi `$ need not be periodic but should only satisfy the milder requirement $`\phi (e^{2\pi i}z,e^{2\pi i}\overline{z})=\phi (z,\overline{z})+2\pi vR`$, $`v`$. This statement is equivalent to the compactification of the field $`\phi `$ on a circle of radius $`R`$ and implies that
$$ab=ivR,v.$$
The Lagrangian density does not depend on $`\phi _0`$ and this constant may be set to zero. Therefore only the difference of the constant terms in $`\phi _{B_1}`$ and $`\phi _{B_2}`$ remains. We choose to parametrize this difference by a real number $`x[0,2\pi R)`$ and an integer $`m`$:
$$(a+b)\mathrm{ln}q=x+2\pi mR.$$
The reason for this parametrization is again the compactification of $`\phi `$: even though the various pairs $`(\phi _{B_1}+2\pi mR,\phi _{B_2})`$, $`m`$, will give different contributions to the functional integral, they all represent the same restriction of $`e^{i\phi /R}`$ at the boundary.
We are interested in computing the partition function $`Z(\phi _{B_1},\phi _{B_2})=Z(x,\{b_k\},\{a_k\})`$ defined as
$$Z(x,\{b_k\},\{a_k\})=_B𝒟\phi e^{_𝒜(\phi )d^2z},$$
(12)
where $`_B`$ denotes the integration on the space of functions $`\phi `$ such that the restrictions of $`e^{i\phi /R}`$ at the inner and outer boundaries coincide with $`e^{i\phi _{B_1}/R}`$ and $`e^{i\phi _{B_2}/R}`$. (The dependence on the compactification radius $`R`$ is always implicit.) The decomposition of the field in a classical part verifying the boundary conditions and fluctuations vanishing at the extremities leads to
$$Z(x,\{b_k\},\{a_k\})=\mathrm{\Delta }^{1/2}Z_{\text{class}}(x,\{b_k\},\{a_k\}).$$
(13)
The factor $`\mathrm{\Delta }`$ is the $`\zeta `$-regularization of the determinant for the annulus and is known to be (see for example ):
$$\mathrm{\Delta }^{1/2}=(i\tau )^{1/2}\eta ^1(\tau ),\text{with }q=e^{i\pi \tau },$$
(14)
where $`\eta (\tau )=e^{i\pi \tau /12}_{m=1}^{\mathrm{}}(1e^{2im\pi \tau })`$ is the Dedekind $`\eta `$ function. The factor $`Z_{\text{class}}`$ is the integration (sum) over all classical solutions compatible with the boundary conditions in the above sense. To obtain $`Z_{\text{class}}`$ we solve the classical equations ($`_z_{\overline{z}}\phi =0`$) with boundary conditions given by $`(\phi _{B_1},\phi _{B_2})`$. The condition at the outer circle ($`z=e^{i\theta },\overline{z}=e^{i\theta }`$) is $`\phi _n+\overline{\phi }_n=a_n`$ and that at the inner one ($`z=qe^{i\theta },\overline{z}=qe^{i\theta }`$) is $`q^n\phi _n+q^n\overline{\phi }_n=b_n`$. The solution can be written as the sum
$$\phi =a\mathrm{ln}z+b\mathrm{ln}\overline{z}+\stackrel{~}{\phi }_1+\stackrel{~}{\phi }_2$$
(15)
where the two function $`\stackrel{~}{\phi }_1`$ and $`\stackrel{~}{\phi }_2`$ are harmonic inside the annulus and take respectively the values $`\phi _{B1}`$ and $`0`$ on the inner boundary and the values $`0`$ and $`\phi _{B2}`$ on the outer one. These functions are
$$\stackrel{~}{\phi }_1(z,\overline{z})=\underset{k0}{}\frac{b_k}{q^k\frac{1}{q^k}}(z^k\overline{z}^k),$$
$$\stackrel{~}{\phi }_2(z,\overline{z})=\underset{k0}{}\frac{a_k}{\frac{1}{q^k}q^k}\left(\left(\frac{z}{q}\right)^k\left(\frac{\overline{z}}{q}\right)^k\right).$$
Hence the classical solution $`\phi `$ is completely determined by the data $`(x,\{b_k\},\{a_k\})`$ up to the two integers $`m,v`$ that determine $`a`$ and $`b`$. The factor $`Z_{\text{class}}`$ is consequently the sum:
$$\underset{m,v}{}e^{(\phi _{(m,v)})}$$
where $`\phi _{(m,v)}`$ is the solution (15) with $`(a+b)\mathrm{ln}q=x+2\pi mR`$ and $`ab=ivR`$.
Using this expression and the Poisson summation formula on the index $`m`$, Langlands computed the desired partition function as the product
$$Z(x,\{b_k\},\{a_k\})=\mathrm{\Delta }^{1/2}Z_1(x)Z_2(\{b_k\},\{a_k\})$$
(16)
where
$$Z_1(x)=\underset{u,v}{}e^{iux/R}q^{\frac{u^2}{4R^2}+v^2R^2}=\underset{u,v}{}e^{ix(\alpha _{u,v}+\overline{\alpha }_{u,v})}q^{h_{u,v}+\overline{h}_{u,v}}$$
(17)
and
$$Z_2(\{b_k\},\{a_k\})=\underset{k=1}{\overset{\mathrm{}}{}}\mathrm{exp}\left(2k\left(\frac{1+q^{2k}}{1q^{2k}}(a_ka_k+b_kb_k)\frac{2q^k}{1q^{2k}}(a_kb_k+b_ka_k)\right)\right).$$
(18)
## 4 Explicit form of the boundary states
In this section we rewrite (16) as a sum over $`u,v`$ of terms of the form
$$Z^{(u,v)}(x,\{b_k\},\{a_k\})=𝔵^{(u,v)}(\phi _{B1})|q^{L_0+\overline{L}_0}|𝔵^{(u,v)}(\phi _{B2}),$$
(19)
with $`|𝔵^{(u,v)}(\phi )_{(u,v)}\overline{}_{(u,v)}`$ and where
$$Z^{(u,v)}(x,\{b_k\},\{a_k\})=\mathrm{\Delta }^{\frac{1}{2}}e^{iux/R}q^{\frac{u^2}{4R^2}+v^2R^2}Z_2(\{b_k\},\{a_k\}).$$
The goal is therefore to find a map $`𝔵^{(u,v)}`$ such that (19) holds. To do so we will first set $`Z_2(\{b_k\},\{a_k\})`$ in the form
$$Z_2(\{b_k\},\{a_k\})=\underset{k=1}{\overset{\mathrm{}}{}}\underset{m,n}{}B_{m,n}^kq^{k(m+n)}A_{m,n}^k$$
(20)
where $`A_{m,n}^k=A_{m,n}^k(a_k,a_k)`$ and $`B_{m,n}^k=B_{m,n}^k(b_k,b_k)`$ are functions of only two variables. With $`Q=q^k`$, $`a_\pm =2i\sqrt{k}a_{\pm k}`$ and $`b_\pm =2i\sqrt{k}b_{\pm k}`$ for $`k1`$, the terms $`\mathrm{\Delta }^{1/2}Z_2`$ of (16) are a product over $`k`$ of
$$e^{a_+a_{}/2}e^{b_+b_{}/2}\frac{e^{(a_+a_{}Q^2+a_+b_{}Q+b_+a_{}Q+b_+b_{}Q^2)/(1Q^2)}}{1Q^2},$$
(21)
up to a constant depending only on $`\tau `$. The factors in front are clearly factorizable and can be absorbed in the definition of $`A_{m,n}^k`$ and $`B_{n,m}^k`$. The remaining mixed term can be developped as a power series in $`Q`$:
$$\underset{i,j,k,l=0}{\overset{\mathrm{}}{}}\frac{(a_+a_{}+b_+b_{})^i}{i!}\frac{(a_+b_{})^j(b_+a_{})^k}{j!k!}\frac{(i+j+k+l)!}{(i+j+k)!l!}Q^{2i+j+k+2l}.$$
(22)
To achieve the form (20), the coefficient of the term $`q^{k(m+n)}`$ (i.e. of $`Q^{2i+j+k+2l}`$ with $`2i+j+k+2l=m+n`$) in the above expression has to be the product of two functions, one of $`(a_k,a_k)`$, the other of $`(b_k,b_k)`$. We concentrate on the terms with $`jk`$ and denote by $`S_{m,n}`$ the factor of $`(a_+b_{})^{mn}Q^{m+n}`$ with $`jk=mn`$. The terms with $`j<k`$ are treated similarly. With the use of $`x=a_+a_{}`$ and $`y=b_+b_{}`$, $`S_{m,n}`$ can be written as
$$S_{m,n}(x,y)=\underset{i+k+l=n}{}\frac{(x+y)^i(xy)^k}{i!(mn+k)!k!}\frac{(m+k)!}{(mn+i+2k)!l!}.$$
This function is clearly symmetric in $`x`$ and $`y`$. Define
$$R_{m,n}(x)S_{m,n}(x,0)$$
and
$$T_{m,n}R_{m,n}(0)=S_{m,n}(0,0).$$
Casting (22) in the form (20) will only be possible if
$$S_{m,n}(x,y)T_{m,n}=R_{m,n}(x)R_{m,n}(y).$$
(23)
That this condition is verified is highly non-trivial. It was found in that it is, although in disguised form, the Saalschütz identity . We thus have the factorization if we define
$`A_{m,n}^k(a_k,a_k)={\displaystyle \frac{R_{m,n}(x)}{\sqrt{T_{m,n}}}}a_+^{mn}e^{a_+a_{}/2},\text{if }mn`$ (24)
where
$$R_{m,n}=\underset{i=0}{\overset{n}{}}\frac{m!x^i}{i!(mn)!(mn+i)!(ni)!},mn,$$
and
$$T_{m,n}=\frac{m!}{n!((mn)!)^2},mn.$$
This expression for $`R_{m,n}`$ shows that it is related to the $`n`$-th Laguerre polynomial of the $`(mn)`$-th kind by
$$R_{m,n}(x)=\frac{L_n^{(mn)}(x)}{(mn)!},mn.$$
(25)
A similar calculation leads to
$$R_{m,n}(x)=\frac{L_m^{(nm)}(x)}{(nm)!},nm.$$
(26)
Going back to the initial notation, we finally get the desired form with:
$$A_{m,n}^k(a_k,a_k)=\{\begin{array}{cc}(2i\sqrt{k}a_k)^{mn}\sqrt{\frac{n!}{m!}}e^{2k|a_k|^2}L_n^{(mn)}(4k|a_k|^2)\hfill & mn\hfill \\ (2i\sqrt{k}a_k)^{nm}\sqrt{\frac{m!}{n!}}e^{2k|a_k|^2}L_m^{(nm)}(4k|a_k|^2)\hfill & nm,\hfill \end{array}$$
(27)
and $`B_{m,n}^k(b_k,b_k)=\overline{A_{m,n}^k(b_k,b_k)}`$. (An orientation on the boundary must be chosen to define the map $`𝔵`$. For example moving in the positive direction of the parameter $`\theta `$ should put the cylinder at one’s left. This explains the interchange $`b_kb_k`$ in the functions $`B`$.)
It is now straightforward to define the map from the boundary conditions to the Hilbert space. We have just shown that the contribution of the $`(u,v)`$ sector to the partition function can be written as
$$Z^{(u,v)}(x,\{b_k\},\{a_k\})=q^{h_{u,v}+\overline{h}_{u,v}}e^{ix(\alpha _{u,v}+\overline{\alpha }_{u,v})}\underset{k}{}\underset{m,n=0}{\overset{\mathrm{}}{}}B_{m,n}^kq^{k(m+n)}A_{m,n}^k.$$
(Note that both sectors $`(u,v)`$ and $`(u,v)`$ contribute the same quantity to $`Z`$. There seems therefore to be a freedom to attach $`(u,v)`$ to either $`_{(u,v)}\overline{}_{(u,v)}`$ or $`_{(u,v)}\overline{}_{(u,v)}`$. This choice is resolved in the next section.) Using the fact that
$$u,v|\frac{(𝔞_k^m𝔞_k^n)(𝔞_k^{}^m^{}𝔞_k^{}^n^{})}{\sqrt{k^{m+n}k_{}^{}{}_{}{}^{m^{}+n^{}}m!n!m^{}!n^{}!}}|u,v=\delta _{k,k^{}}\delta _{m,m^{}}\delta _{n,n^{}},$$
we get
$`Z^{(u,v)}`$ $`(x,\{b_k\},\{a_k\})=`$ (28)
$`=q^{h_{u,v}+\overline{h}_{u,v}}e^{ix(\alpha _{u,v}+\overline{\alpha }_{u,v})}{\displaystyle \underset{k,k^{}}{}}{\displaystyle \underset{m,m^{},n,n^{}}{}}B_{m,n}^kq^{k(m+n)}A_{m^{},n^{}}^k^{}\delta _{k,k^{}}\delta _{m,m^{}}\delta _{n,n^{}}`$ (29)
$`=𝔵^{(u,v)}(x_1,\{b_k\})|q^{L_0\overline{L}_0}|𝔵^{(u,v)}(x_2,\{a_k\})`$ (30)
where
$$|𝔵^{(u,v)}(x_2,\{a_k\})=e^{ix_2(\alpha _{u,v}+\overline{\alpha }_{u,v})}\underset{k=1}{\overset{\mathrm{}}{}}\underset{m,n=0}{\overset{\mathrm{}}{}}A_{m,n}^k(a_k,a_k)\frac{𝔞_k^m𝔞_k^n}{\sqrt{k^{m+n}m!n!}}|u,v.$$
(31)
We have reintroduced, somewhat arbitrarily, the constant term $`x_2`$ in $`\phi _{B2}`$. Again, only the difference $`x=x_2x_1`$ between the constant term $`x_2`$ in $`\phi _{B2}`$ and $`x_1`$ in $`\phi _{B1}`$ has a physical meaning. We now have an explicit form for the map $`𝔵`$.
The vector $`|𝔵^{(u,v)}(x_2,\{a_k\})`$ can be cast into a simpler form. With the help of the following recursion identities
$$(n+1)L_{n+1}^{(m(n+1))}(x)[x_xx+(mn)]L_n^{(mn)}(x)=0,m1n0$$
and
$$L_n^{((m+1)n)}(x)+[_x1]L_n^{(mn)}(x)=0,mn0,$$
we can prove by induction on both indices that
$$A_{m,n}^k=\frac{(_{}+\frac{1}{2}a_+)^m(_++\frac{1}{2}a_{})^n}{\sqrt{m!n!}}e^{a_+a_{}/2},$$
(32)
where, we recall, $`a_\pm =2i\sqrt{k}a_{\pm k}`$ and $`_\pm =\frac{}{a_\pm }`$. Defining $`\alpha _k=\frac{i}{2}(_k+2ka_k),\overline{\alpha }_k=\frac{i}{2}(_k+2ka_k)`$ and $`\mathrm{\Omega }_k=A_{0,0}^k=e^{a_+a_{}/2}=e^{2k|a_k|^2}`$, we get
$$A_{m,n}^k=\frac{\alpha _k^m\overline{\alpha }_k^n}{\sqrt{k^{m+n}m!n!}}\mathrm{\Omega }_k.$$
(33)
The correspondence $`𝔞_ki\alpha _k`$ and $`\overline{𝔞}_ki\overline{\alpha }_k`$ induces an isomorphism with a subalgebra of the Heisenberg algebra since the $`\alpha _k`$’s and $`\overline{\alpha }_k`$’s satisfy the commutation rules:
$$[\alpha _n,\alpha _m]=n\delta _{n+m,0}[\overline{\alpha }_n,\overline{\alpha }_m]=n\delta _{n+m,0}$$
$$[\alpha _n,\overline{\alpha }_m]=0.$$
If $`|\alpha _{uv}|\overline{\alpha }_{u,v}`$ is identified with $`\mathrm{\Omega }=_k\mathrm{\Omega }_k`$ and $`\alpha _0`$ (resp. $`\overline{\alpha }_0`$) is defined as acting by multiplication by $`\alpha _{u,v}`$ (resp. $`\overline{\alpha }_{u,v}`$), this correspondence can then be extended to an isomorphism of Heisenberg modules. Since the $`𝔞_k`$’s and $`\alpha _k`$’s, $`k>0`$, all commute with one another, we are able to write down an exponential form for the boundary state:
$`|𝔵^{(u,v)}(x_2,\{a_k\})`$ $`=`$ $`e^{ix_2(\alpha _{u,v}+\overline{\alpha }_{u,v})}{\displaystyle \underset{k}{}}\left\{{\displaystyle \underset{m,n}{}}{\displaystyle \frac{(\alpha _k𝔞_k)^m(\overline{\alpha }_k\overline{𝔞}_k)^n}{m!n!k^mk^n}}\mathrm{\Omega }_k\right\}|u,v`$ (34)
$`=`$ $`e^{ix_2(\alpha _{u,v}+\overline{\alpha }_{u,v})}{\displaystyle \underset{k}{}}\left\{e^{\alpha _k𝔞_k/k}e^{\overline{\alpha }_k\overline{𝔞}_k/k}\right\}\mathrm{\Omega }|u,v`$
$`=`$ $`e^{ix_2(\alpha _{u,v}+\overline{\alpha }_{u,v})}{\displaystyle \underset{k}{}}\left\{e^{(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}\right\}\mathrm{\Omega }|u,v`$
$`=`$ $`e^{ix_2(\alpha _{u,v}+\overline{\alpha }_{u,v})}e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}\mathrm{\Omega }|u,v.`$
Up to the factor $`(i\tau )^{1/2}e^{i\pi \tau /12}`$ the partition function takes the following form:
$$Z(x,\{b_k\},\{a_k\})=𝔵(x_1,\{b_k\})|q^{L_0+\overline{L}_0}|𝔵(x_2,\{a_k\}),$$
(35)
in which we have defined
$$|𝔵(x_2,\{a_k\})=e^{ix_2(𝔞_0+\overline{𝔞}_0)}e^{_{k=1}^{\mathrm{}}(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}\mathrm{\Omega }|\mathrm{\Lambda }$$
(36)
$$|\mathrm{\Lambda }=\underset{u,v}{}|u,v,$$
(37)
where the operators $`𝔞_0=𝔞_01`$ and $`\overline{𝔞}_0=1𝔞_0`$ act as the identity times $`\alpha _{u,v}`$ and $`\overline{\alpha }_{u,v}`$ on $`|u,v=|\alpha _{u,v}|\overline{\alpha }_{u,v}`$. The boundary states $`|𝔵(x_2,\{a_k\})`$ belong to the direct sum of Fock spaces $`_{u,v}_{(u,v)}\overline{}_{(u,v)}`$ or, more precisely, to the sum $`_{u,v}(_{(u,v)}\overline{}_{(u,v)})^c`$ of some completions that contains formal series like (36). In the next section, it will turn out to be useful to include the $`x`$-dependence in $`\mathrm{\Omega }`$, which will then be noted $`\mathrm{\Omega }_{u,v}`$, to highlight its sector:
$$\mathrm{\Omega }_{u,v}=e^{ix(\alpha _{u,v}+\overline{\alpha }_{u,v})}\mathrm{\Omega }.$$
(38)
The boundary state $`|𝔵(x_2,\{a_k\})`$ has an analogue in string theory. The propagation of a closed string can be wrtitten as the expectation value of an evolution operator between in and out states. In light-cone coordinates, these states are labeled by Fourier coefficients of $`(D2)`$ periodic coordinate functions, $`D`$ being the spacetime dimension, and their expression has been given in . Restricting the expressions (12, 13) in to a single coordinate function, the dependency on the Fourier coefficients $`a_k,k0`$, can be shown (after a somewhat lengthy calculation) to be identical to ours. Both expressions, theirs and ours, have an overall phase that depends on the zero mode $`a_0`$. Their phase does not depend on winding numbers but ours does. The next section will show that, for the problem at hand, the phase in $`|𝔵(x_2,\{a_k\})`$ is crucial to assure the proper behavior of $`|𝔵`$ under conformal transformations in each Fock sector $`_{(u,v)}\overline{}_{(u,v)}`$.
## 5 Conformal Transformations and Boundary States
Having found an explicit and concise form for the boundary states, we can now study their properties under conformal transformations. Let $`g`$ be an infinitesimal conformal transformation that leaves the boundary unchanged and $`G`$ the corresponding element in the Virasoro algebra. The purpose of this section is to show that the action of $`g`$ on the boundary condition $`\phi `$ and that of $`G`$ on $`|𝔵(\phi )`$ commute:
$$|𝔵(g\phi )=G|𝔵(\phi ).$$
(39)
We first discuss the actions $`g`$ and $`G`$ and the correspondence between them.
One can easily convince oneself that the only infinitesimal conformal transformations that preserve the center and radius of a circle in the complex plane are linear combinations of
$$(l_p\overline{l}_p),p,$$
(40)
where the conformal generators $`l_p`$ and $`\overline{l}_p`$ are defined as $`l_p=z^{p+1}_z`$ and $`\overline{l}_p=\overline{z}^{p+1}_{\overline{z}}`$. Note that the subalgebra $`_p(L_p\overline{L}_p)\text{Vir}\overline{\text{Vir}}`$ is centerless and the mapping defined by $`(l_p\overline{l}_p)(L_p\overline{L}_p)`$ of the boundary preserving conformal transformation into $`\text{Vir}\overline{\text{Vir}}`$ is an isomorphism. However the tranformations $`(l_p\overline{l}_p),p0`$, do not preserve the reality condition imposed on the boundary functions. The generators $`(l_p+\overline{l}_p)`$ and $`i(l_p\overline{l}_p)`$ do. Both reality and geometry preserving conditions are therefore satisfied by the infinitesimal transformations
$`g_0^{(1)}`$ $`=`$ $`1+iϵ\left\{l_0\overline{l}_0\right\}`$ (41)
$`g_p^{(1)}`$ $`=`$ $`1+iϵ\left\{(l_p+l_p)(\overline{l}_p+\overline{l}_p)\right\},p>0`$ (42)
and
$$g_p^{(2)}=1+ϵ\left\{(l_pl_p)+(\overline{l}_p\overline{l}_p)\right\},p>0.$$
(43)
We shall show that (39) holds if the $`g_p^{(i)}`$’s are defined as above and the corresponding $`G_p^{(i)}`$’s are taken to be
$`G_0^{(1)}`$ $`=`$ $`1+iϵ\{L_0\overline{L}_0\},`$ (44)
$`G_p^{(1)}`$ $`=`$ $`1+iϵ\left\{(L_p+L_p)(\overline{L}_p+\overline{L}_p)\right\}`$ (45)
and
$$G_p^{(2)}=1+ϵ\left\{(L_pL_p)+(\overline{L}_p\overline{L}_p)\right\}.$$
(46)
Since, for $`p0`$, we have
$$[g_p^{(1)}1,g_0^{(1)}1]=ϵp(g_p^{(2)}1),$$
the property for the second family of transformations follows directly if it is proven to be true for the first one. The action of $`G`$ in the rhs of (39) is simply left-multiplication. On the lhs, the action is defined as usual by $`(g\phi )(z,\overline{z})=\phi g^1(z,\overline{z})`$. We first study the case $`p>0`$. For $`p=0`$, the particularity of the Sugawara construction will modify the analysis. We will end this section by examining this case.
Let us first compute $`|𝔵(g_p\phi )`$ with $`g_p=g_p^{(1)}`$. Note that, due to the use of the Poisson summation formula to obtain (17), the constant $`m`$ in $`(a+b)\mathrm{ln}q=x+2\pi mR`$ is not anymore well-defined in the sector $`(u,v)`$. However the difference $`(ab)`$ still is. Only $`ab`$ will appear in the variation $`g_p\phi `$. As observed in the previous paragraph, the two contributions $`Z^{(u,v)}`$ and $`Z^{(u,v)}`$ are equal. It turns out that equation (39) holds when the functions $`\phi `$ with a given $`v`$ are mapped into the sectors $`_{(u,v)}\overline{}_{(u,v)}`$, $`u`$. (For the other choice $`_{(u,v)}\overline{}_{(u,v)}`$, the actions $`g`$ and $`G`$ fail to commute.) The function on the boundary must have the form
$$\phi (\theta )=x+vR\theta +\underset{k>0}{}(a_ke^{ik\theta }+a_ke^{ik\theta })$$
or, equivalently
$$\phi (z,\overline{z})=x+(a\mathrm{ln}z+b\mathrm{ln}\overline{z})+\underset{k>0}{}(a_kz^k+\overline{a}_k\overline{z}^k)$$
with $`z=e^{i\theta }`$ and $`\overline{z}=e^{i\theta }`$ and the reality condition $`a_k=\overline{a}_k`$. A direct calculation gives
$$g_p\phi =\stackrel{~}{x}+vR\theta +\underset{k>0}{}c_ke^{ik\theta }+\overline{c}_ke^{ik\theta }$$
where
$`c_k`$ $`=a_k+iϵ\left((k+p)a_{k+p}+(kp)a_{kp}\right)+ϵvR\delta _{k,p}`$ (47)
$`\overline{c}_k`$ $`=\overline{a}_kiϵ\left((k+p)\overline{a}_{k+p}+(kp)\overline{a}_{kp}\right)+ϵvR\delta _{k,p}`$ (48)
$`\stackrel{~}{x}`$ $`=x+iϵp(a_p\overline{a}_p).`$ (49)
One can see that the reality condition imposed on $`\phi `$ is indeed preserved. Since
$`|𝔵^{(u,v)}(\stackrel{~}{x},\{c_k\})`$ $`=\left(e^{_{k>0}(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}\mathrm{\Omega }_{u,v}\right)|_{g_p\phi }|u,v`$ (50)
$`=e^{_{k>0}(\stackrel{~}{\alpha }_k𝔞_k+\stackrel{~}{\overline{\alpha }}_k\overline{𝔞}_k)/k}\stackrel{~}{\mathrm{\Omega }}_{u,v}|u,v`$ (51)
where we have defined
$`\stackrel{~}{\alpha }_k`$ $`={\displaystyle \frac{i}{2}}({\displaystyle \frac{}{\overline{c}_k}}+2kc_k),`$ (52)
$`\stackrel{~}{\overline{\alpha }}_k`$ $`={\displaystyle \frac{i}{2}}({\displaystyle \frac{}{c_k}}+2k\overline{c}_k),`$ (53)
$`\stackrel{~}{\mathrm{\Omega }}_{u,v}`$ $`=e^{i\stackrel{~}{x}(\alpha _{u,v}+\overline{\alpha }_{u,v})}e^{2_{k>0}kc_k\overline{c}_k},`$ (54)
a first step is to express $`\stackrel{~}{\alpha }_k`$, $`\stackrel{~}{\overline{\alpha }}_k`$, and $`\stackrel{~}{\mathrm{\Omega }}_{u,v}`$ in terms of $`x`$ and the $`a_k`$’s. This can be easily achieved. The expressions for $`c_k`$ and $`\overline{c}_k`$ given above can be inverted in order to obtain closed form expressions for $`a_k`$ and $`\overline{a}_k`$. It is then a simple exercise to show that
$`\stackrel{~}{\alpha }_k`$ $`={\displaystyle \frac{i}{2}}({\displaystyle \frac{}{\overline{c}_k}}+2kc_k)=\{\begin{array}{cc}\alpha _k+iϵk(\alpha _{k+p}+\alpha _{kp}),\hfill & kp,\hfill \\ \alpha _p+iϵp(\alpha _{2p}+\alpha _{uv}),\hfill & k=p,\hfill \end{array}`$
$`\stackrel{~}{\overline{\alpha }}_k`$ $`={\displaystyle \frac{i}{2}}({\displaystyle \frac{}{c_k}}+2k\overline{c}_k)=\{\begin{array}{cc}\overline{\alpha }_kiϵk(\overline{\alpha }_{k+p}+\overline{\alpha }_{kp}),\hfill & kp,\hfill \\ \overline{\alpha }_kiϵp(\overline{\alpha }_{2p}+\overline{\alpha }_{uv}),\hfill & k=p.\hfill \end{array}`$
Using these expressions, the functional $`\stackrel{~}{\mathrm{\Omega }}_{u,v}=\mathrm{\Omega }_{u,v}|_{g_p\phi }`$ can be expressed in terms of the original variables. A careful treatment of the infinite sums leads to
$$\stackrel{~}{\mathrm{\Omega }}_{u,v}=\left(1+iϵ\left(\frac{1}{2}\underset{0<k<p}{}(\alpha _{pk}\alpha _k\overline{\alpha }_{pk}\overline{\alpha }_k)+\alpha _p\alpha _{u,v}\overline{\alpha }_p\overline{\alpha }_{u,v}\right)\right)\mathrm{\Omega }_{u,v}.$$
(55)
Finally we can rewrite $`|𝔵^{(u,v)}(g_p\phi )`$ to first order in $`ϵ`$ as
$`|𝔵^{(u,v)}(g_p\phi )`$ $`=e^{_k(\stackrel{~}{\alpha }_k𝔞_k+\stackrel{~}{\overline{\alpha }}_k\overline{𝔞}_k)/k}\stackrel{~}{\mathrm{\Omega }}_{u,v}|u,v`$
$`=e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}`$
$`\times (1+iϵ{\displaystyle \underset{k>0}{}}((\alpha _{k+p}+\alpha _{kp})𝔞_k(\overline{\alpha }_{k+p}+\overline{\alpha }_{kp})\overline{𝔞}_k)`$
$`+iϵ(\alpha _p𝔞_0\overline{\alpha }_p\overline{𝔞}_0+{\displaystyle \frac{1}{2}}{\displaystyle \underset{0<k<p}{}}(\alpha _{pk}\alpha _k\overline{\alpha }_{pk}\overline{\alpha }_k)))\mathrm{\Omega }_{u,v}|u,v.`$ (56)
We now turn our attention to the rhs of (39), namely $`G_p|𝔵(\phi )`$. It is convenient to introduce the operator $`u_{m,n}^k`$ defined as
$$u_{m,n}^k=\frac{𝔞_k^m𝔞_k^n}{\sqrt{k^{m+n}m!n!}}.$$
(57)
It is such that
$$u,v|u_{m,n}^k|𝔵(\phi )=A_{m,n}^k(\phi )\frac{\mathrm{\Omega }_{u,v}}{\mathrm{\Omega }_k},$$
(58)
where, we recall, $`\mathrm{\Omega }_k=e^{2k|a_k|^2}`$. Moreover
$$\left(u_{m^{},n^{}}^k^{}|u,v\right)^{}\left(u_{m,n}^k|u,v\right)=\delta _{k,k^{}}\delta _{m,m^{}}\delta _{n,n^{}}$$
(59)
and
$$\underset{k=1}{\overset{\mathrm{}}{}}\underset{m,n=0}{\overset{\mathrm{}}{}}u_{m,n}^k|u,vu,v|u_{m,n}^k$$
(60)
acts as the identify on $`_{(u,v)}\overline{}_{(u,v)}`$.
From the Sugawara construction
$$L_p=\frac{1}{2}\underset{k}{}:𝔞_{pk}𝔞_k:,p0,$$
(61)
we see that
$$[L_p,𝔞_k^m]=mk𝔞_k^{m1}𝔞_{k+p}$$
(62)
with similar equations for the anti-holomorphic sector. Also, since $`u,v|𝔞_k=0`$ if $`k<0`$,
$`u,v|L_p`$ $`=`$ $`u,v|\left({\displaystyle \frac{1}{2}}{\displaystyle \underset{0<k<p}{}}𝔞_k𝔞_{pk}+𝔞_p𝔞_0\right).`$ (63)
For the same reason, the operator $`L_p`$ annihilates $`u,v|`$ when acting from the right. We thus have
$`u,v|u_{m,n}^kL_p`$ $`=`$ $`u,v|\left([u_{m,n}^k,L_p]+L_pu_{m,n}^k\right)`$ (64)
$`=`$ $`u,v|\left(mk{\displaystyle \frac{(𝔞_k^{m1}𝔞_{k+p})𝔞_k^n}{\sqrt{k^{m+n}m!n!}}}+L_pu_{m,n}^k\right)`$
and
$`u,v|u_{m,n}^kL_p`$ $`=`$ $`u,v|\left([u_{m,n}^k,L_p]\right)`$ (65)
$`=`$ $`u,v|\left(mk{\displaystyle \frac{(𝔞_k^{m1}𝔞_{kp})𝔞_k^n}{\sqrt{k^{m+n}m!n!}}}\right).`$
With these two relations and their equivalent in the anti-holomorphic sector, we can compute $`u,v|u_{m,n}^kG_p`$:
$`u,v|u_{m,n}^kG_p`$ $`=`$ $`u,v|\left(u_{m,n}^k+iϵk{\displaystyle \frac{𝔞_k^{m1}𝔞_k^{n1}}{\sqrt{k^{m+n}m!n!}}}(m(𝔞_{k+p}+𝔞_{kp})\overline{𝔞}_kn𝔞_k(\overline{𝔞}_{k+p}+\overline{𝔞}_{kp}))\right)`$ (66)
$`+iϵu,v|\left(𝔞_p𝔞_0\overline{𝔞}_p\overline{𝔞}_0+{\displaystyle \frac{1}{2}}{\displaystyle \underset{0<l<p}{}}𝔞_l𝔞_{pl}\overline{𝔞}_l\overline{𝔞}_{pl}\right)u_{m,n}^k.`$
We thus get
$`u,v|u_{m,n}^kG_p|𝔵(\phi )`$ $`=\left(1+iϵ\left(\alpha _p\alpha _{u,v}\overline{\alpha }_p\overline{\alpha }_{u,v}+{\displaystyle \frac{1}{2}}{\displaystyle \underset{0<l<p}{}}\alpha _l\alpha _{pl}\overline{\alpha }_l\overline{\alpha }_{pl}\right)\right)A_{m,n}^k{\displaystyle \frac{\mathrm{\Omega }_{u,v}}{\mathrm{\Omega }_k}}`$
$`+iϵk{\displaystyle \frac{\alpha _k^{m1}\overline{\alpha }_k^{n1}}{\sqrt{k^{m+n}m!n!}}}\left(m(\alpha _{k+p}+\alpha _{kp})\overline{\alpha }_kn\alpha _k(\overline{\alpha }_{k+p}+\overline{\alpha }_{kp})\right)\mathrm{\Omega }_{u,v}.`$ (67)
Once again $`\overline{\alpha }_0`$ acts on $`\mathrm{\Omega }_{u,v}`$ by multiplication by $`\overline{\alpha }_{u,v}`$. Using the completeness relation (60), we can reconstruct $`G_p|𝔵^{(u,v)}(\phi )`$. First, summing over $`m`$ and $`n`$ gives
$`{\displaystyle \underset{m,n}{}}{\displaystyle \frac{u,v|u_{m,n}^kG_p|𝔵^{(u,v)}(\phi )}{_{lk}\mathrm{\Omega }_l}}u_{m,n}^k|u,v`$
$`={\displaystyle \frac{1}{_{lk}\mathrm{\Omega }_l}}\left(1+iϵ\left({\displaystyle \frac{1}{2}}{\displaystyle \underset{0<l<p}{}}\alpha _l\alpha _{pl}\overline{\alpha }_l\overline{\alpha }_{pl}+\alpha _p𝔞_0\overline{\alpha }_p\overline{𝔞}_0\right)\right)\left({\displaystyle \underset{m,n}{}}A_{m,n}^k{\displaystyle \underset{lk}{}}\mathrm{\Omega }_l{\displaystyle \frac{𝔞_k^m\overline{𝔞}_k^n}{\sqrt{k^{m+n}m!n!}}}\right)|u,v`$
$`+{\displaystyle \frac{iϵ}{_{lk}\mathrm{\Omega }_l}}\left(k(\alpha _{k+p}+\alpha _{kp}){\displaystyle \underset{m1,n0}{}}m\alpha _k^{m1}\overline{\alpha }_k^n{\displaystyle \frac{𝔞_k^m\overline{𝔞}_k^n}{k^{m+n}m!n!}}\right)\mathrm{\Omega }_{u,v}|u,v`$
$`{\displaystyle \frac{iϵ}{_{lk}\mathrm{\Omega }_l}}\left(k(\overline{\alpha }_{k+p}+\overline{\alpha }_{kp}){\displaystyle \underset{m0,n1}{}}n\alpha _k^m\overline{\alpha }_k^{n1}{\displaystyle \frac{𝔞_k^m\overline{𝔞}_k^n}{k^{m+n}m!n!}}\right)\mathrm{\Omega }_{u,v}|u,v.`$
The last two terms can be rewritten as
$`{\displaystyle \frac{iϵ}{_{lk}\mathrm{\Omega }_l}}\left((\alpha _{k+p}+\alpha _{kp})𝔞_k{\displaystyle \underset{m0,n0}{}}\alpha _k^m\overline{\alpha }_k^n{\displaystyle \frac{𝔞_k^m\overline{𝔞}_k^n}{k^{m+n}m!n!}}\right)\mathrm{\Omega }_{u,v}|u,v`$
$`{\displaystyle \frac{iϵ}{_{lk}\mathrm{\Omega }_l}}\left((\overline{\alpha }_{k+p}+\overline{\alpha }_{kp})\overline{𝔞}_k{\displaystyle \underset{m0,n0}{}}\alpha _k^m\overline{\alpha }_k^n{\displaystyle \frac{𝔞_k^m\overline{𝔞}_k^n}{k^{m+n}m!n!}}\right)\mathrm{\Omega }_{u,v}|u,v.`$ (68)
Putting all this together, we finally have
$`G_p|𝔵^{(u,v)}(\phi )`$ $`=e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}`$
$`\times [1+iϵ{\displaystyle \underset{k>0}{}}((\alpha _{k+p}+\alpha _{kp})𝔞_k(\overline{\alpha }_{k+p}+\overline{\alpha }_{kp})\overline{𝔞}_k)`$
$`+iϵ(\alpha _p𝔞_0\overline{\alpha }_p\overline{𝔞}_0+{\displaystyle \frac{1}{2}}{\displaystyle \underset{0<k<p}{}}(\alpha _{pk}\alpha _k\overline{\alpha }_{pk}\overline{\alpha }_k))]\mathrm{\Omega }_{u,v}|u,v`$
$`=|𝔵^{(u,v)}(g_p\phi ).`$ (69)
We have thus established the desired property for $`p>0`$. It is the phase $`e^{ix(\alpha _{uv}+\overline{\alpha }_{uv})}`$ in $`|𝔵`$ and the winding term $`vR\theta `$ in $`\phi `$ that are responsible for the terms $`𝔞_0`$ and $`\overline{𝔞}_0`$ in eq. (5). Their role is therefore crucial to prove the conformal property (39).
The transformation $`g_0^{(1)}=1+iϵ(l_0\overline{l}_0)`$ is nothing but an infinitesimal rotation. The gaussian terms are invariant under these transformations, because Fourier coefficients only pick up a phase. Hence $`\mathrm{\Omega }_{u,v}|_{g_0^{(1)}\phi }=(1+iϵvR(\alpha _{u,v}+\overline{\alpha }_{u,v}))\mathrm{\Omega }_{u,v}|_\phi `$. The computation of $`|𝔵^{(u,v)}(g_0^{(1)}\phi )`$ is straightforward and one gets
$`|𝔵^{(u,v)}(g_0^{(1)}\phi )`$ $`=`$ $`e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}`$ (70)
$`\times \left(1+iϵ\left({\displaystyle \underset{k>0}{}}\alpha _k𝔞_k\overline{\alpha }_k\overline{𝔞}_k\right)+iϵvR(\alpha _{u,v}+\overline{\alpha }_{u,v})\right)\mathrm{\Omega }_{u,v}|u,v`$
$`=`$ $`e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}`$
$`\times \left(1+iϵ\left({\displaystyle \underset{k>0}{}}\alpha _k𝔞_k\overline{\alpha }_k\overline{𝔞}_k\right)+iϵ(h_{u,v}\overline{h}_{u,v})\right)\mathrm{\Omega }_{u,v}|u,v.`$
The action of $`G_0`$ is somehow different. In this case, the $`\overline{L}_0`$ term does not annihilate $`u,v|`$, but rather acts on it by multiplying by $`\overline{h}_{u,v}`$. We thus get
$`G_0|𝔵^{(u,v)}(\phi )`$ $`=`$ $`e^{_k(\alpha _k𝔞_k+\overline{\alpha }_k\overline{𝔞}_k)/k}`$ (71)
$`\times \left(1+iϵ\left({\displaystyle \underset{k>0}{}}\alpha _k𝔞_k\overline{\alpha }_k\overline{𝔞}_k\right)+iϵ(h_{u,v}\overline{h}_{u,v})\right)\mathrm{\Omega }_{u,v}|u,v`$
$`=`$ $`|𝔵^{(u,v)}(g_0^{(1)}\phi ).`$
This completes the proof.
## 6 Concluding remarks
This simple yet quite instructive calculation gives an example of a conformal theory with non-conformally invariant boundary conditions. Can the map $`\phi |𝔵(\phi )`$ for the free boson be used to investigate minimal models with general boundary conditions? It is well known that minimal models can be constructed from the $`c=1`$ CFT, using the Coulomb gas technique. This was succesfully done on the plane by Dotsenko and Fateev and on the torus by Felder . This might be one path to construct the map for these models.
Langlands and the two authors have recently studied numerically the statistical distribution of the Fourier coefficients of a field defined for the Ising model. This distribution is more intricate than the boson’s as the Fourier coefficients of the field at one boundary do not appear now to be mutually independent. The map $`\phi |𝔵(\phi )`$, if it exists for the Ising model, might be a rich object.
## Acknowledments
The authors would like to thank Robert Langlands for explaining his original construction and stressing its limitations, Philippe Zaugg for helpful discussions and Rafael Nepomechie for bringing to our attention the papers in string theory relevant to the present work.
M.-A. L. gratefully acknowledges fellowships from the NSERC Canada Scholarships Program and the Celanese Foundation, and Y. S.-A. support from NSERC (Canada) and FCAR (Québec).
|
warning/0002/math0002066.html
|
ar5iv
|
text
|
# Untitled Document
Cubic surfaces and Borcherds products
February 7, 2000
2000 MSC: 11F55, 14J10
Daniel Allcock Department of Mathematics Harvard University Cambridge, MA 02138 ALLCOCK@ MATH.HARVARD.EDU Eberhard Freitag Mathematisches Institut Im Neuenheimer Feld 288 D69120 Heidelberg FREITAG@ MATHI.UNI-HEIDELBERG.DE
1. Introduction
The moduli space $``$ of marked cubic surfaces can be identified with the Baily-Borel compactification of $`_4/\mathrm{\Gamma }`$, where $`_4`$ denotes the complex $`4`$-ball and $`\mathrm{\Gamma }`$ is a certain arithmetic reflection group. (See \[ACT2\] and also \[ACT1\].) In this paper we use the methods of R. Borcherds to construct automorphic forms on $`_4`$. We will obtain an embedding of $``$ into the $`9`$-dimensional projective space $`P^9(C)`$, whose image is the intersection of 270 explicitly known cubic $`8`$-folds. This map is compatible with the actions of the Weyl group $`W(E_6)`$ on $``$ and $`P^9`$. The former action arises because $`W(E_6)`$ permutes the markings of cubic surfaces, and the latter action arises from the unique irreducible $`10`$-dimensional representation of $`W(E_6)`$. Furthermore, the cubic $`8`$-folds are all equivalent under $`W(E_6)`$. The $`10`$-dimensional linear system associated to this map into $`P^9(C)`$ contains $`270`$ automorphic forms with known zeros, which play a central role in our investigation. In particular, there is a direct connection between them and the classical invariants of cubic surfaces introduced by Cayley. He considered the 27 lines on a smooth cubic surface and a certain configuration of 45 planes that they determine. By considering 4-tuples of these planes that meet along one of the 27 lines, Cayley constructed 270 cross-ratios, and showed that these allow one to recover the original surface. We show that Cayley’s cross-ratios coincide not with our Borcherds products but rather with the quotients of certain pairs of them. This relies on work of Naruki \[Na\] and is the main part of our proof that our map of $``$ into $`P^9`$ is an embedding. We are grateful to R. Borcherds, B. van Geemen, and R. Vakil for helpful discussions.
2. The complex reflection group
Let
$$=Z[\omega ],\omega =[\sqrt[3]{1}]=\frac{1}{2}+\frac{\sqrt{3}}{2},$$
be the ring of Eisenstein integers. We consider the lattice
$$\mathrm{\Lambda }=^{1,4},$$
which is the $``$-module $`^5`$ equipped with the hermitian form of signature $`(1,4)`$ given by
$$a,b=\overline{a}_0b_0\overline{a}_1b_1\mathrm{}\overline{a}_4b_4.$$
$`(2.1)`$
Let $`Aut(\mathrm{\Lambda })`$ be the unitary group of this lattice, i.e. the group of $``$-module automorphisms which preserve the hermitian form. Complex conjugation acts as the identity on the residue field
$$F_3=/\sqrt{3},$$
which has order 3, so the hermitian form induces a $`F_3`$-valued quadratic form on the 5-dimensional $`F_3`$-vector space $`V=\mathrm{\Lambda }/\sqrt{3}\mathrm{\Lambda }`$. We denote the orthogonal group of $`V`$ by $`\mathrm{O}(5,3)`$ and define $`\mathrm{\Gamma }`$ to be the kernel of the action of $`Aut(\mathrm{\Lambda })`$ on $`V`$. We have the exact sequence
$$1\mathrm{\Gamma }Aut(\mathrm{\Lambda })\mathrm{O}(5,3)1.$$
For future reference we mention that $`V`$ contains 242 nonzero elements, of which 80 have norm $`0`$, 90 have norm $`1`$ and 72 have norm $`1`$. Nonzero vectors in $`V`$ are equivalent under $`\mathrm{O}(5,3)`$ if and only if they have the same norm. The subgroup of $`\mathrm{O}(5,3)`$ generated by the reflections in the norm $`1`$ vectors is isomorphic to the Weyl group $`W(E_6)`$, and $`\mathrm{O}(5,3)W(E_6)\times \{\pm 1\}`$. Furthermore, $`W(E_6)`$ contains a simple subgroup of index 2 and order $`\mathrm{25\hspace{0.17em}920}`$. A lattice vector $`a\mathrm{\Lambda }`$ is called primitive if it cannot be divided in $`\mathrm{\Lambda }`$ by a non-unit of $``$. Also, $`a`$ is called
isotropic if $`a,a=0`$, a short root if $`a,a=1`$, or a long root if $`a,a=2`$.
The roots are important because $`Aut(\mathrm{\Lambda })`$ contains reflections in them. If $`a`$ is a short root and $`\zeta `$ is a unit of $``$ (a sixth root of unity) then the map
$$vv(1\zeta )\frac{a,v}{a,a}a$$
is an automorphism of $`\mathrm{\Lambda }`$. (In the special case $`\zeta =\pm 1`$ this is also true if $`a`$ is a long root.) This automorphism fixes the orthogonal complement of $`a`$ and maps $`a`$ to $`\zeta a`$. We call this automorphism a reflection if $`\zeta 1`$. The order of a reflection is two, three or six, and we sometimes call reflections of these orders biflections, triflections and hexflections. The third roots of unity are congruent to 1 mod $`\sqrt{3}`$, and therefore the triflections belong to the congruence group $`\mathrm{\Gamma }`$. We remark that these triflections actually generate $`\mathrm{\Gamma }`$ \[ACT2\], although we will not need this fact. We need some information about the orbit structure of $`\mathrm{\Lambda }`$ with respect to $`\mathrm{\Gamma }`$. If $`a,b\mathrm{\Lambda }`$ are in the same $`\mathrm{\Gamma }`$-orbit, then their images in $`V`$ coincide. In some special cases the converse is true:
2.1 Proposition. Let $`a`$ and $`b`$ be two primitive isotropic vectors, or two short roots, or two long roots. Then $`a`$ and $`b`$ are equivalent under $`\mathrm{\Gamma }`$ if and only if their images in $`V`$ coincide. The number of $`\mathrm{\Gamma }`$-orbits of lines $`Ca`$, where $`a`$ is a primitive isotropic vector, a short root or a long root, is 40, 36 or 45, respectively.
Proof. The “only if” part is trivial. To prove the converse, we use the fact that $`Aut(\mathrm{\Lambda })`$ acts transitively on primitive isotropic vectors, on short roots, and on long roots (see Theorems 7.22 and 9.15 of \[ACT2\]). It is a general fact that if a group $`G`$ acts transitively on a set $`X`$, $`N`$ is a normal subgroup, and $`xX`$ has stabilizer $`G_x`$ in $`G`$, then the orbits of $`N`$ on $`X`$ are in 1-1 correspondence with the cosets in $`G/N`$ of the image of $`G_x`$. We apply this with $`G=Aut(\mathrm{\Lambda })`$, $`N=\mathrm{\Gamma }`$, and $`x`$ a primitive isotropic vector, short root or long root of $`\mathrm{\Lambda }`$. Then the number of $`\mathrm{\Gamma }`$-orbits into which the $`Aut(\mathrm{\Lambda })`$-orbit of $`x`$ splits is equal to the index in $`\mathrm{O}(5,3)`$ of the reduction modulo $`\sqrt{3}`$ of $`Aut(\mathrm{\Lambda })_x`$. We will now compute these reductions. We first take $`x`$ to be a primitive null vector. According to paragraph 7.8 of \[ACT2\], its stabilizer in $`Aut(\mathrm{\Lambda })`$ contains as a normal subgroup a Heisenberg group with center $`Im()`$ and central quotient $`^3`$, and the stabilizer modulo this Heisenberg group is the isometry group $`(Z/6)^3:S_3`$ of the lattice $`^3`$. By considering the matrices for these transformations, it is easy to see that the center of the Heisenberg group acts trivially on $`V`$, that $`^3`$ acts as $`^3/(\sqrt{3}^3)(Z/3)^3`$, that $`(Z/6)^3`$ acts as $`(Z/2)^3`$, and that $`S_3`$ acts faithfully. The image of the stabilizer in $`\mathrm{O}(5,3)`$ is a group $`3^3:2^3:S_3`$, which has index 80 in $`\mathrm{O}(5,3)`$. Next we take $`x`$ to be a short root of $`\mathrm{\Lambda }`$, say $`(0,0,0,0,1)`$, and $`\overline{x}`$ to be its image in $`V`$. Then the stabilizer of $`\overline{x}`$ is the orthogonal group of $`\overline{x}^{}`$, which is generated by the reflections in the nonisotropic elements of $`\overline{x}^{}`$. One can enumerate these vectors and check that each is the image of a root of $`x^{}`$. The biflections in these roots reduce to reflections of $`V`$, proving that the stabilizer of $`x`$ in $`Aut(\mathrm{\Lambda })`$ surjects to the stabilizer of $`\overline{x}`$, which has index 72 in $`\mathrm{O}(5,3)`$. Exactly the same argument applies if $`x`$ is a long root, say $`(0,0,0,1,1)`$, yielding an index of 90. We have shown that there are 80 (resp. 72, 90) orbits of primitive isotropic vectors (resp. short roots, long roots) in $`\mathrm{\Lambda }`$, which is the same as the number of nonzero elements of $`V`$ of norm 0 (resp. $`1`$, $`1`$). Since the map from $`\mathrm{\Gamma }`$-orbits of such lattice vectors to the corresponding set of vectors in $`V`$ is onto, it is bijective. This proves the first claim of the theorem, and the second follows immediately. $``$$``$
3. The ball quotient
The group $`\mathrm{\Gamma }`$ acts on a complex 4-ball in the projective space of $`C^{1,4}=\mathrm{\Lambda }_{}C`$. We will describe this in some generality, for convenience in later sections. Let o be an order in an imaginary quadratic number field. An o-lattice $`L`$ is a finitely generated projective o-module equipped with a Hermitian pairing $`,`$ on $`L`$ that takes value in the field of fractions of o. We take such pairings to be antilinear in the first and linear in the second variable. We say that $`L`$ is Lorentzian when its signature is $`(1,n)`$ with $`n1`$. A point of the projective space $`P(L_\text{o}C)`$ is called positive if it is represented by a vector of positive norm. When $`L`$ is Lorentzian, the positive points form an open $`n`$-ball $`(L)`$ in projective space, which is also called the complex hyperbolic space of $`L`$. $`Aut(L)`$ acts properly discontinuously on $`(L)`$, and there is a natural compactification of the quotient, due to Baily and Borel \[BB\]. A cusp is an element of $`P(L_\text{o}C)`$ that can be represented by an isotropic lattice vector. The cusps are the rational boundary points of $`(L)`$, and there are only finitely many orbits under $`Aut(L)`$. We denote by $`^{}(L)`$ the union of $`(L)`$ with the set of all cusps. The group $`Aut(L)`$ acts on this extension, and the quotient of $`^{}(L)`$ by any finite-index subgroup of $`Aut(L)`$ carries the structure of a projective algebraic variety. In our setting we have $`\text{o}=`$ and $`L=\mathrm{\Lambda }`$. The hermitian form on $`C^{1,4}=\mathrm{\Lambda }_{}C`$ is given by Eq. (2.1), and the identification of $`(\mathrm{\Lambda })`$ with the complex 4-ball is easy. Namely, any element of $`(\mathrm{\Lambda })`$ has a unique representative $`zC^{1,4}`$ whose $`z_0`$-component is $`1`$. Considering the remaining coordinates identifies $`(\mathrm{\Lambda })`$ with the set of all $`(z_1,\mathrm{},z_4)C^4`$ satisfying
$$|z_1|^2+\mathrm{}+|z_4|^2<1.$$
$`(3.1)`$
We will write $`_4`$ for $`(\mathrm{\Lambda })`$. We are interested in the quotient $`X`$ of $`_4^{}`$ by $`\mathrm{\Gamma }Aut(\mathrm{\Lambda })`$. By Prop. 2.1, there are forty $`\mathrm{\Gamma }`$-orbits of cusps in $`_4^{}`$, so the boundary of the Baily-Borel compactification of $`_4/\mathrm{\Gamma }`$ consists of 40 points. Let $`a\mathrm{\Lambda }`$ be a vector of negative norm. The orthogonal complement $`a^{}`$ of $`a`$ in $`P(C^{1,4})`$ meets $`_4`$ nontrivially because $`a`$ has negative norm. We can consider its intersection with $`_4^{}`$ and the image of this in $`X`$. It is known that this is an algebraic subvariety of codimension one, and we are interested in this construction for $`a`$ a root of $`\mathrm{\Lambda }`$. In this case we call $`a^{}_4^{}`$ a mirror of $`_4^{}`$. The terminology derives from the fact that the mirror is the fixed-point set of the reflection(s) in $`a`$, and we call the mirror short or long according to whether $`a`$ is short or long. The image in $`X`$ of a short (long) mirror is called a short (long) mirror of $`X`$. For convenience we sometimes call a vector in $`V`$ short (resp. long) if it has norm $`1`$ (resp. 1). The short (long) vectors in $`V`$ are exactly the images of the short (long) roots of $`\mathrm{\Lambda }`$. The short (long) mirrors in $`X`$ correspond to the 36 (45) pairs $`\{\pm a\}`$ of short (long) vectors of $`V`$. We will need some results about the intersection behavior of mirrors. Orthogonality of mirrors in $`_4^{}`$ is defined in the obvious way, and we call two mirrors in $`X`$ orthogonal if the corresponding elements of $`V`$ are orthogonal. If two mirrors in $`_4^{}`$ are orthogonal then so are their images in $`X`$.
3.1 Lemma. Two short mirrors in $`_4`$ are either orthogonal or disjoint.
Proof. We take $`x`$ and $`y`$ to be short roots whose mirrors are the given mirrors. If the mirrors meet in $`^4`$ then a point of the intersection represents a positive-definite one-dimensional subspace of $`C^{1,4}`$. Its orthogonal complement is negative definite and contains $`x`$ and $`y`$. Hence the Gram matrix of $`x,y`$ must be positive, so $`x,xy,y|x,y|^2>0`$, so $`|x,y|^2<1`$. Since $`x,y`$ we must have $`x,y=0`$. $``$$``$
We now introduce the notion of a cross. This is fundamental for the paper because the automorphic forms we will construct vanish exactly along the points of a cross in $`_4^{}`$. The word “cross” is meant to suggest several mutually orthogonal objects.
3.2 Definition. A cross in $`V`$ is a set of 5 pairwise orthogonal pairs $`\pm a_i`$, one pair consisting of long vectors and the others consisting of short vectors. The associated cross in $`X`$ is the union of the mirrors of the $`\pm a_i`$; it follows that a cross in $`X`$ is a set of 5 pairwise orthogonal mirrors, one long and 4 short. The associated cross in $`_4^{}`$ is the preimage of the cross in $`X`$. A point of $`_4^{}`$ lies in this cross just if it is orthogonal to a root whose projection to $`V`$ is one of the $`\pm a_i`$.
Since the three types of cross are in natural bijection, we will pass between them without comment.
3.3 Lemma. There are 135 crosses, three containing each of the 45 long mirrors of $`X`$, and all 135 crosses are all equivalent under $`\mathrm{O}(5,3)`$. More precisely, if $`\mathrm{}`$ is a long mirror in $`X`$ then the 12 short mirrors orthogonal to $`\mathrm{}`$ decompose in a unique way into three sets of 4 mirrors which are pairwise orthogonal, and the stabilizer of $`\mathrm{}`$ in $`\mathrm{O}(5,3)`$ permutes these sets transitively.
Proof. The transitivity of $`\mathrm{O}(5,3)`$ on crosses in $`V`$ is obvious, and the rest is just a calculation. Namely, the orthogonal complement of a long vector $`a`$ contains 12 pairs $`\{\pm v\}`$ of short vectors, orthogonality is (surprisingly) a transitive relation, generating an equivalence relation with three classes of size 4. By symmetry it suffices to check this for a single long vector $`a`$, say $`(1,0,0,0,0)`$. Then the three classes are
| $`\{\pm (0,1,0,0,0),`$ | $`\pm (0,0,1,0,0),`$ | $`\pm (0,0,0,1,0),`$ | $`\pm (0,0,0,0,1)\},`$ |
| --- | --- | --- | --- |
| $`\{\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1)\},`$ |
| $`\{\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1),`$ | $`\pm (0,1,1,1,1)\}.`$ |
$``$$``$
The purpose of the following theorem is to allow us to prove in section 4 that the automorphic forms we construct there have no common zeros.
3.4 Theorem. No point of $`X`$ lies on all 135 crosses. Furthermore, if $`p`$ is the point of $`_4`$ represented by $`(1,0,0,0,0)\mathrm{\Lambda }`$, then the image of $`p`$ in $`X`$ is the only point of $`X`$ that lies on all the crosses containing it. Finally, for each boundary point $`b`$ of $`X`$, $`b`$ is the only point of $`X`$ that lies on all the crosses containing it.
In order to prove the theorem we will need to understand the $`\mathrm{\Gamma }`$-orbits of points of $`_4`$ that, like $`p`$, lie on four short mirrors. If $`q`$ is such a point, then $`q^{}`$ is a copy of the unimodular lattice $`^{0,4}`$, and it follows that $`q`$ is represented by a lattice vector of norm $`1`$, and indeed by six such vectors. The images in $`V`$ of these vectors and of the short roots of $`q^{}`$ form a cross, which we call the cross associated to $`q`$.
3.5 Lemma. The map just defined, which associates a cross to each point of $`_4`$ that lies on $`4`$ short mirrors, defines a bijection between the set of $`\mathrm{\Gamma }`$-orbits of such points and the set of crosses. If each of $`p,q_4`$ lies on four short mirrors, and the images in $`V`$ of the short roots of $`p^{}`$ coincide with the images of the short roots of $`q^{}`$, then $`p`$ and $`q`$ are $`\mathrm{\Gamma }`$-equivalent.
Proof. For the first claim one uses the argument of 2.1. The essential facts are that $`Aut(\mathrm{\Lambda })`$ acts transitively on such points of $`_4`$ and that the stabilizer in $`Aut(\mathrm{\Lambda })`$ of such a point of $`_4`$ is $`(Z/6)\times (Z/6)^4:S_4`$, which reduces modulo $`\sqrt{3}`$ to $`(Z/2)^5:S_4`$, of index 135 in $`\mathrm{O}(5,3)`$. (The transitivity statement follows from the fact that such points in $`_4`$ correspond bijectively to the decompositions of $`\mathrm{\Lambda }`$ as a direct sum $`^{1,0}^{0,4}`$.) The second claim is a consequence of the first: the short vectors of a cross determine the cross uniquely, so the crosses associated to $`p`$ and $`q`$ coincide. $``$$``$
Proof of Theorem 3.4. Most of the proof consists of computer calculations concerning combinatorics in $`V`$; we will describe the ideas in sufficient detail for them to be reproduced easily. One can enumerate the roots orthogonal to $`p`$, and their images in $`V`$. A cross contains $`p`$ just if it contains one of these images. One can compute the set $`𝒞`$ of crosses satisfying this condition, and one finds $`|𝒞|=69<135`$. In particular, $`p`$ does not lie on all 135 crosses. Now we will show that $`p`$ is the only point of $`_4`$ (up to $`\mathrm{\Gamma }`$-equivalence) that lies on all the crosses containing $`p`$. Suppose $`q_4`$ lies on every cross of $`𝒞`$; we will show that $`q`$ is $`\mathrm{\Gamma }`$-equivalent to $`p`$. First we will show that $`q`$ lies on 4 short mirrors. For otherwise the short roots orthogonal to $`q`$ project into some triple $`T`$ of mutually orthogonal antipodal pairs of short vectors of $`V`$. If $`q`$ lies on every cross in $`𝒞`$ then there is a way to choose a root in $`q^{}`$ for each $`C𝒞`$, such that the image in $`V`$ of the root is one of the vectors of $`C`$. In particular, there is a way to choose an element $`vC`$ for each $`C𝒞`$, such that (1) if $`v`$ is short then $`vT`$, and (2) the span of all the $`v`$’s has dimension at most $`4`$. For each of the 540 possibilities for $`T`$ one can count the number of ways to choose vectors $`v`$ satisfying (1) and (2). It turns out that there are no ways to make such a choice, and it follows that $`q`$ cannot lie on only 3 (or fewer) short mirrors. We have shown that the short roots of $`q^{}`$ project onto some quadruple of mutually orthogonal antipodal pairs of short vectors of $`V`$, which we will denote by $`T`$. As in the previous paragraph, there is a way to choose an element $`vC`$ for each $`C𝒞`$, such that (1) and (2) are satisfied. For each of the 135 possibilities for $`T`$, one can count the number of ways to make such a set of choices. It turns out that for only one quadruple is there a way to do this, and this quadruple consists of $`\pm (0,1,0,0,0),\mathrm{},\pm (0,0,0,0,1)`$. Therefore the images in $`V`$ of the short roots of $`q^{}`$ are these 8 vectors. Since these are also the images of the short roots of $`p^{}`$, the $`\mathrm{\Gamma }`$-equivalence of $`p`$ and $`q`$ follows from Lemma 3.5. Now we turn to the boundary points of $`X`$. If $`b`$ is a boundary point of $`_4^{}`$ then we may represent it by a primitive isotropic lattice vector $`w`$, and a cross contains $`b`$ just if it contains the image in $`V`$ of a root orthogonal to $`w`$. One can check that every nonisotropic element of $`V`$ that is orthogonal to the image $`\overline{w}`$ of $`w`$ is the image of a root in $`w^{}`$. (This is easy to check for any given $`w`$, and the result follows for general $`w`$ because of the transitivity of $`Aut\mathrm{\Lambda }`$.) It follows that the set of crosses $`𝒞_b`$ containing $`b`$ consists of the crosses which contain a vector of $`V`$ orthogonal to $`\overline{w}`$. It is easy to compute the sets $`𝒞_b`$ for each of the 40 orbits of boundary points, and to check that no $`𝒞_b`$ is a subset of $`𝒞`$. This proves the second part of the theorem. The first part then follows, because no point of $`X`$ except for the image of $`p`$ lies on every cross in $`𝒞`$, and this point lies on only 69 of the 135 crosses. Now we show that no point of $`/\mathrm{\Gamma }`$ lies on all the crosses in $`𝒞_b`$, for any boundary point $`b`$. The proof is almost identical to the one used above. By symmetry it suffices to treat just one $`𝒞_b`$. If $`q_4`$, then the short roots of $`q^{}`$ project into some quadrouple $`T`$ of mutually orthogonal short vectors of $`V`$. If $`q`$ lies on every cross in $`𝒞_b`$ then there is a way to choose an element $`vC`$ for each $`C𝒞_b`$, such that (1) and (2) are satisfied. An enumeration shows that there is no way to make such a choice, and the claim follows. Finally, it is easy to compare the $`𝒞_b`$’s with each other as $`b`$ varies over the boundary points, and check that none of the $`𝒞_b`$’s contains any other. It follows that for each boundary point $`b`$ of $`X`$, $`b`$ is the only point of $`X`$ that lies on all the crosses containing $`b`$. This completes the proof. We verified the enumerations with a computer program written in C++, which ran to completion in less than a minute. Repeatedly checking condition (2) required more than $`4\times 10^8`$ row-reduction operations, and we did this efficiently by enumerating the $`3^5`$ elements of $`V`$ and preparing a lookup table of all $`3^{52}`$ possible row-reductions. $``$$``$
4. Automorphic forms on the ball
Borcherds has given two constructions for automorphic forms on $`\mathrm{O}(2,n)`$, which we will use to build automorphic forms on the 4-ball. Here we will use his additive lift \[Bo1,§14\], which generalizes correspondences of Shimura, Doi-Naganuma, Maass, Gritsenko, and others. In the next section we will discuss his other construction, which uses infinite products. We begin in the setting of section 3, with o an order in an imaginary quadratic number field, $`L`$ an o-lattice of signature $`(1,n)`$, $`(L)`$ the associated ball in projective space, and $`^{}(L)`$ the union of the ball with the cusps. We assume that $`L`$ is integral (all inner products lie in o) and that $`n>1`$, so that $`L`$ has dimension at least 3. We define $`\stackrel{~}{}(L)`$ and $`\stackrel{~}{}^{}(L)`$ to be the preimages of $`(L)`$ and $`^{}(L)`$ in $`L_\text{o}C`$. If $`G`$ is a subgroup of $`Aut(L)`$ and $`v:GS^1C^{}=C\{0\}`$ is a character of $`G`$ then an automorphic form of weight $`kZ`$ with respect to $`G`$ and $`v`$ is a holomorphic function $`f:\stackrel{~}{}(L)C`$ satisfying
a) $`f(tz)=t^kf(z)`$ for $`tC^{}`$, and
b) $`f(\gamma z)=v(\gamma )f(z)`$ for $`\gamma G`$. (If $`n`$ were 1, so that $`(L)`$ were one-dimensional, then we would impose an additional condition of regularity at the cusps.) We denote the space of all such forms by $`[G,k,v]`$, or by $`[G,k]`$ if $`v`$ is trivial. One can extend an automorphic form $`f:\stackrel{~}{}(L)C`$ to $`\stackrel{~}{}^{}(L)`$ in a natural way, providing boundary values for $`f`$. If $`a`$ is an isotropic element of $`\stackrel{~}{}^{}(L)`$, so that it represents a cusp, then by the non-degeneracy of $`,`$ we may choose $`bL_\text{o}C`$ satisfying $`a,b0`$. For all $`\tau C`$ with sufficiently large imaginary part, $`\tau a+2\text{i}a,bb`$ has positive norm. The limit
$$f(a):=\underset{Im\tau \mathrm{}}{lim}f(\tau a+2\text{i}b,ab)$$
exists and is independent of the choice of $`b`$. This follows from the Fourier Jacobi expansion of $`f`$ at a cusp; we refer to \[Sh\] for more details. An automorphic form $`f[G,k,v]`$ is of course not a function on $`(L)`$ unless $`k=0`$. But it is clear that the zero-locus of $`f`$ is preserved by $`G`$ and scalar multiplication, so the set of zeros of $`f`$ in $`^{}(L)/G`$ is well-defined. It is a closed algebraic subvariety of pure codimension one.
Borcherds’ additive lift
We consider the $`Z`$-lattice $`M`$ underlying $`L`$, which is the underlying $`Z`$-module equipped with the even integral bilinear form
$$(a,b):=a,b+b,a,$$
which has signature $`(2,2n)`$. The dual lattice with respect to $`(,)`$ is denoted $`M^{}`$, and $`M^{}/M`$ is a finite group. We remark that if $`\alpha ,\beta M^{}/M`$ then $`(\alpha ,\alpha )`$ and $`(\beta ,\beta )`$ are well-defined modulo 2, while $`(\alpha ,\beta )`$ is well-defined modulo 1. The group $`SL(2,Z)`$ acts on the group ring $`C[M^{}/M]`$ by means of the Weil representation $`\varrho _M`$, which is defined in terms of the standard generators
$$T=\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right),S=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)$$
by
$$\begin{array}{cc}\hfill \varrho _M(T)e_\alpha & =\mathrm{exp}(\pi \mathrm{i}(\alpha ,\alpha ))e_\alpha ,\hfill \\ \hfill \varrho _M(S)e_\alpha & =\frac{\text{i}^{n1}}{\sqrt{|M^{}/M|}}\underset{\beta M^{}/M}{}\mathrm{exp}(2\pi \mathrm{i}(\alpha ,\beta ))e_\beta .\hfill \end{array}$$
(We denote the standard generators of the group ring $`C[M^{}/M]`$ by $`e_\alpha `$, with $`\alpha `$ varying over $`M^{}/M`$.) The Weil representation factors through $`SL(2,Z/NZ)`$, where $`N`$ is the smallest natural number such that $`\frac{N}{2}(a,a)`$ is integral for all $`aM^{}`$. The inputs of Borcherds’ additive lift are vector valued modular forms $`f:HC[M^{}/M]`$ on the usual upper half plane $`H`$ with respect to the Weil representation. More precisely, we require that $`f=(f_\alpha )_{\alpha M^{}/M}`$ satisfy
$$\begin{array}{ccc}& f_\alpha (\tau +1)=e^{\pi \mathrm{i}(\alpha ,\alpha )}f_\alpha (\tau ),\hfill & 1.\hfill \\ & f_\alpha \left(\frac{1}{\tau }\right)=\tau ^{k+1n}\frac{\text{i}^{n1}}{\sqrt{|M^{}/M|}}\underset{\beta M^{}/M}{}e^{2\pi \mathrm{i}(\alpha ,\beta )}f_\beta (\tau )\text{, and}\hfill & 2.\hfill \\ & f\text{ is holomorphic at the cusp infinity. }\hfill & 3.\hfill \end{array}$$
Borcherds’ additive lift allows also inputs which have poles at the cusps, but we do not need this extension. But even in the case of modular forms which are regular at the cusps, Borcherds extended previous constructions because he imposes no restriction on the weight of $`f`$, and does not require that $`f`$ be a cusp form. The additive lift is a linear map $`\mathrm{\Psi }`$ from the space of such $`f`$ into a certain space of automorphic forms on $`(L)`$. We give its important properties in the following theorem, which is a specialization of Theorem 14.3 in \[Bo1\] to $`\mathrm{U}(1,n)\mathrm{O}(2,2n)`$.
4.1 Theorem. Let $`G`$ be the subgroup of $`Aut(L)`$ that acts trivially on $`M^{}/M`$. There exists a linear map $`\mathrm{\Psi }`$ (the additive lift) from the space of elliptic modular forms with the properties 1–3 above into the space $`[G,k]`$ of automorphic forms of weight $`k`$ with respect to $`G`$ and the trivial character. This lifting is equivariant with respect to the action of $`Aut(L)`$. ($`Aut(L)`$ acts on $`[G,k]`$ because $`G`$ is normal in $`Aut(L)`$, and on the space of elliptic modular forms via its action on $`M^{}/M`$.)
Furthermore, Borcherds shows how to compute the values of $`\mathrm{\Psi }(f)`$ at the cusps of $`\stackrel{~}{}^{}(L)`$ from the Fourier coefficients of $`f`$. We now turn to the case of interest, with $`\text{o}=`$ and $`L=\mathrm{\Lambda }`$. The $`Z`$-lattice underlying the 1-dimensional lattice $``$ is the $`A_2`$ root lattice (the hexagonal lattice in the plane with minimal norm 2), which has index 3 in its dual. From the definition of $`\mathrm{\Lambda }`$ as a direct sum, we see that $`M^{}/M`$ has order $`3^5`$. Indeed more is true: $`M^{}`$ coincides with $`(\sqrt{3})^1\mathrm{\Lambda }`$, so that $`M^{}/M`$ is canonically isomorphic to the $`F_3`$-vector space $`V=\mathrm{\Lambda }/\sqrt{3}\mathrm{\Lambda }`$ introduced in section 2. In particular, $`G`$ is the congruence subgroup $`\mathrm{\Gamma }`$. One can check that if $`\alpha ,\beta M^{}/M`$ then $`(\alpha ,\beta )`$ is $`0`$, $`2/3`$ or $`2/3`$ (modulo 1) according to whether the corresponding elements of $`V`$ have inner product $`0`$, $`1`$ or $`1`$ (in $`F_3`$). Similarly, if $`\alpha M^{}/M`$ then $`(\alpha ,\alpha )`$ is $`0`$, $`2/3`$ or $`2/3`$ (modulo 2) according to whether the corresponding element of $`V`$ has norm $`0`$, $`1`$ or $`1`$. It follows from this that the level of the Weil representation is $`N=3`$, so that the representation factors through $`SL(2,F_3)`$. We will usually write $`V`$ in place of $`M^{}/M`$ to lighten the notation. We apply 4.1 in the simplest case, where $`f`$ is a modular form of weight $`0`$, hence a constant, which is to say an element of $`C[V]^{SL(2,F_3)}`$. The weight being $`0`$ means that the exponent $`1k+n`$ of $`\tau `$ in the second transformation rule is $`0`$, so that $`k=n1=3`$. Therefore Borcherds’ additive lift gives a linear map
$$C[V]^{SL(2,F_3)}[\mathrm{\Gamma },3].$$
We remark that since $`\mathrm{\Gamma }`$ contains the cube roots of unity acting as scalars, every automorphic form on $`\stackrel{~}{}_4`$ for $`\mathrm{\Gamma }`$, with trivial character has weight divisible by 3. Our first task is to find some elements of $`C[V]^{SL(2,F_3)}`$.
4.2 Lemma. Let $`a_0,\mathrm{},a_4`$ be an orthogonal basis for $`V`$ consisting of one long vector and four short vectors, and let $`C=(C_\alpha )_{\alpha V}C[V]`$ be defined by the condition that $`C_\alpha `$ is the complex number $`1`$, $`0`$ or $`1`$ according to whether $`_i(\alpha ,a_i)`$ is the element $`1`$, $`0`$ or $`1`$ of $`F_3`$. Then $`C`$ is invariant under the Weil representation. Furthermore, $`C`$ changes sign under reflection in any of the $`a_i`$, and is characterized up to a scalar by this property.
To avoid the impression that the $`C`$’s were discovered by clever guesswork, we should mention that we found this construction quite late, following extensive computer work. Proof. The behavior of $`C`$ under the reflections is obvious, and the invariance under $`SL(2,F_3)`$ may be checked by a computer calculation. To see the last claim, suppose $`D=(D_\alpha )C[V]`$ has the stated property. If $`\alpha `$ is orthogonal to one of the $`a_i`$ then we have $`D_\alpha =D_\alpha `$ by the transformation rule, so that $`D_\alpha =0`$. The remaining $`\alpha `$ fall into a single orbit of size 32 under the group $`(Z/2)^5`$ generated by the reflections in the $`a_i`$, so all the remaining $`D_\alpha `$ are determined by any one of them. $``$$``$
It is easy to work this out explicitly in an example: if $`a_0,\mathrm{},a_4`$ are $`(1,0,\mathrm{},0),\mathrm{},(0,\mathrm{},0,1)`$ then $`C`$ is supported on those $`\alpha `$ of the form $`(\pm 1,\mathrm{},\pm 1)`$, with $`C_\alpha =+1`$ or $`1`$ according to whether there are an even or odd number of minus signs. Note that $`C`$ is supported on the isotropic vectors in $`V`$, which is not immediately obvious from the construction. It follows from the lemma and Theorem 4.1 that to each cross there is associated an automorphic form on $`_4`$, well-defined up to sign. We will see below that the zero-locus of this form is exactly the associated cross in $`_4`$. To resolve the sign ambiguity it is convenient to introduce the notion of a signed cross. This is just a basis $`\{a_0,\mathrm{},a_4\}`$ as in the lemma, modulo the equivalence relation that $`\{a_0,\mathrm{},a_4\}\{a_0^{},\mathrm{},a_4^{}\}`$ if the $`a_i^{}`$ differ from the $`a_i`$ by a permutation and evenly many sign changes. It is clear that there are two signed crosses for every cross, and that the lemma assigns an element of $`C[V]`$ to each of the 270 signed crosses.
4.3 Lemma. The space
$$C[M^{}/M]^{SL(2,Z)}=C[V]^{SL(2,F_3)}$$
has dimension $`10`$ and is spanned by the elements of $`C[V]`$ associated to the signed crosses. The group $`\mathrm{O}(5,3)`$ acts irreducibly on this space, with $`W(E_6)`$ acting by its unique 10-dimensional irreducible representation and the central involution acting by $`1`$. The multiplicity of this representation in $`C[V]`$ is one.
Proof. It is easy to make a computer construct the elements $`C`$ of $`C[V]`$ associated to the signed crosses and check that their complex span $`Z`$ is 10-dimensional. Consulting the character table shows that any 10-dimensional representation of $`W(E_6)`$ is either trivial, or the irreducible representation in 10 dimensions, or else the sum of the (unique) irreducible 6-dimensional representation and a 4-dimensional trivial one. These may be distinguished by the trace of almost any group element, say a short reflection $`R`$, which has ATLAS \[C\] conjugacy class 2C. The fixed space of $`R`$ in $`Z`$ is spanned by the vectors $`C+R(C)`$ where $`C`$ is as above. It is easy to check that this space has dimension 5, so that $`R`$ has trace 0, so that $`W(E_6)`$ acts irreducibly on $`Z`$. It is obvious that each $`C`$ changes sign under the central involution of $`\mathrm{O}(5,3)`$. If the multiplicity of this $`\mathrm{O}(5,3)`$-representation in $`C[V]`$ were more than one, then the subspace of $`C[V]`$ that changed sign under the reflections of each vector in a cross would have dimension $`>1`$, contrary to Lemma 4.2. To see that $`Z`$ is all of $`C[V]^{SL(2,F_3)}`$, suppose $`C=(C_a)_{aV}`$ is an element of $`C[V]^{SL(2,F_3)}`$. Invariance under $`T`$ means that $`C`$ is supported on the $`81`$ isotropic elements. Invariance under $`S^2=E`$ means $`C_a=C_a`$. Invariance under $`S`$ can be read as a linear equation in $`40`$ indeterminates, and it is easy to make a computer check that the space of solutions has only 10 dimensions. If one is prepared to do more work with group characters, one can of course find the complete decomposition of $`C[V]`$ under $`SL(2,F_3)\times \mathrm{O}(5,3)`$; this is done in \[Fr\]. $``$$``$
We will write $`W`$ for the image of Borcherds’ additive lift $`C[V]^{SL(2,F_3)}[\mathrm{\Gamma },3]`$. Our next theorem asserts that the automorphic forms we have constructed are nontrivial:
4.4 Proposition. Borcherds’ additive lift
$$C[V]^{SL(2,F_3)}W[\mathrm{\Gamma },3]$$
is an $`\mathrm{O}(5,3)`$-equivariant embedding.
Proof. The $`\mathrm{O}(5,3)`$-equivariance is part of Theorem 4.1. To prove injectivity, we construct an inverse by using the boundary values of the automorphic forms. Namely, if $`f[\mathrm{\Gamma },3]`$ then we define $`C=(C_\alpha )_{\alpha V}`$ by taking $`C_\alpha =0`$ if $`\alpha `$ is zero or nonisotropic, and $`C_\alpha =f(\stackrel{~}{\alpha })`$ otherwise, where $`\stackrel{~}{\alpha }`$ is any primitive isotropic vector in $`\mathrm{\Lambda }`$ representing $`\alpha `$. This definition is independent of the choice of $`\stackrel{~}{\alpha }`$ because $`f`$ is $`\mathrm{\Gamma }`$-invariant and all the primitive isotropic preimages of $`\alpha `$ are $`\mathrm{\Gamma }`$-equivalent (Lemma 2.1). The irreducibility of $`C[V]^{SL(2,F_3)}`$ as an $`\mathrm{O}(5,3)`$-module and the fact that its multiplicity in $`C[V]`$ is one imply that the composition
$$C[V]^{SL(2,F_3)}\stackrel{\text{additive lift}}{}W[\mathrm{\Gamma },3]\stackrel{\text{boundary values}}{}C[V]$$
is a scalar. The problem is to show that this scalar is nonzero. This is a straightforward but tedious calculation using Borcherds’ formulae for the Fourier expansions of additive lifts (\[Bo1\], 14.3) and the explicit embedding of $`_4`$ into the hermitian symmetric space of $`\mathrm{O}(2,8)`$. The latter space consists of two-dimensional positive definite real subspaces of $`M_ZR`$. Every positive definite complex line in $`\mathrm{\Lambda }_{}C`$ (i.e. a point in $`_4`$) defines such a subspace. One has to express this embedding in the coordinates which Borcherds uses in his theorem 14.3. Details of this calculation can be found in section 6 of \[Fr\]. $``$$``$
Lemma 4.3 shows that our 270 automorphic forms satisfy many linear equations. Some of these are easy to see, and those treated in the following lemma will receive an elegant geometric interpretation in section 7. To formulate the lemma we note that there is an $`\mathrm{O}(5,3)`$-invariant inner product on $`W`$, which is unique up to scale, by the irreducibility of the representation.
4.5 Lemma. Let $`v`$ be a long vector of $`V`$. Then the automorphic forms associated to the six signed crosses involving $`v`$ lie in a 2-dimensional subspace of $`W`$, and form a scaled copy of the $`A_2`$ root system, i.e., the vertices of a regular hexagon centered at $`0`$.
Proof. One can check this by computing the inner products of the 6 elements of $`C[V]`$, using the restriction of the inner product
$$((C_\alpha ),(D_\alpha ))=\underset{\alpha V}{}C_\alpha D_\alpha ,$$
which is obviously $`\mathrm{O}(5,3)`$-invariant and therefore the natural inner product. But here is a better argument. The reflection $`R`$ in $`v`$ is not in the simple subgroup of $`W(E_6)`$, but $`R`$ is, and has conjugacy class $`2A`$ in ATLAS notation. Consulting the character table shows that $`R`$ has trace $`6`$, so that the subspace $`Z`$ of $`W`$ that $`R`$ negates has dimension 2. Lemma 4.2 associates to each of the three crosses a one-dimensional subspace of $`Z`$, with two generators coming from the associated signed crosses. The reflection of $`V`$ in any short root of one of these crosses preserves that cross, acts as $`1`$ on the the associated subspace, and exchanges the other two crosses and hence the corresponding subspaces. Therefore the three subspaces meet each other at angles of $`\pi /3`$, the subgroup of $`\mathrm{O}(5,3)`$ generated by the reflections in the short vectors orthogonal to $`v`$ acts on $`Z`$ by the $`A_2`$ Weyl group, and the 6 elements of $`Z`$ coming from the crosses form a copy of the $`A_2`$ root system. $``$$``$
We recall the notion of the divisor of an automorphic form. Let $`YX`$ be an irreducible subvariety of codimension one. We denote by $`e_Y`$ the ramification degree with respect to the natural projection $`\pi :_4X`$ (counted as $`1`$ if $`\pi `$ is unramified along $`Y`$). If $`Y`$ is a short mirror this ramification degree is three, because the triflections are contained in $`\mathrm{\Gamma }`$. For any other $`Y`$, such as a long mirror, it is one. If $`F`$ is a nonzero automorphic form for $`\mathrm{\Gamma }`$ and the trivial character, then the vanishing order of $`F`$ along $`\pi ^1(Y)`$ is divisible by $`e_Y`$. We call the quotient of this vanishing order by $`e_Y`$ the order of $`F`$ along $`Y`$ and denote it by $`n_Y(F)`$. The divisor of $`F`$ in $`X`$ is the finite sum
$$(F):=\underset{YX}{}n_Y(F)Y.$$
We consider a cross in $`X`$ as a divisor with multiplicity one at all 5 of its mirrors. A fundamental result for this paper is
4.6 Theorem. Let $`F0`$ lie in the one dimensional space of automorphic forms associated to a cross. Then the divisor of $`F`$ in $`X`$ is exactly this cross. The 270 automorphic forms associated to the signed crosses have no common zeros in $`_4^{}`$.
To prove this we will need a result whose proof we postpone to the next section. We remark that the form $`\chi _4`$ given here was first discovered by Borcherds \[Bo3\].
4.7 Theorem. There are automorphic forms $`\chi _4[Aut(\mathrm{\Lambda }),4,v]`$ and $`\chi _{75}[Aut(\mathrm{\Lambda }),75,v^{}]`$, for some characters $`v`$ and $`v^{}`$ of $`\mathrm{\Lambda }`$, such that the divisors of $`\chi _4`$ and $`\chi _{75}`$ in $`_4`$ are the sum of the short mirrors and the sum of the long mirrors, respectively, with multiplicity one.
Proof of Theorem 4.6. Suppose the cross is $`\{\pm a_0,\mathrm{},\pm a_4\}V`$. If $`\stackrel{~}{a}`$ is a root of $`\mathrm{\Lambda }`$ representing any of the $`\pm a_i`$, and $`R`$ is the biflection in $`\stackrel{~}{a}`$, then the relation $`FR=F`$ (which follows from the construction of $`F`$) implies that $`F`$ vanishes along the mirrors of $`\stackrel{~}{a}`$. Furthermore, if $`\stackrel{~}{a}`$ is a short root then $`F`$ is invariant under the triflection in $`\stackrel{~}{a}`$, so that the multiplicity in $`_4`$ is at divisible by 3. It follows that the divisor of $`F`$ in $`_4`$ contains the short mirrors of the cross with multiplicity 3, plus the long mirrors of the cross. To prove the theorem it suffices to show that this is the full divisor of $`F`$. To see this we construct the product $`P`$ of all 270 automorphic forms, and divide $`P`$ by $`\chi _4^{90}\chi _{75}^6`$, where $`\chi _4`$ and $`\chi _{74}`$ are as in Theorem 4.7. The quotient is holomorphic because $`P`$ vanishes to order at least $`6`$ along each long mirror in $`_4`$ and least $`27034/36=90`$ along each short mirror. The quotient has weight $`2703904756=0`$, so is constant. It is nonzero because each $`F`$ is nonzero. Therefore the divisor of $`P`$ is the same as that of $`\chi _4^{90}\chi _{75}^6`$; since this is also the sum of the “known” divisors of the various $`F`$, the first statement of the theorem follows. The second follows immediately from this and Theorem 3.4. $``$$``$
5. Borcherds products (and proof of theorem 4.7)
We recall some facts about automorphic forms on $`\mathrm{O}(2,n)`$, where $`\mathrm{O}(2,n)`$ is the orthogonal group of a real vector space $`V`$ with a symmetric bilinear form $`(,)`$ of signature $`(2,n)`$. Let $`_n`$ denote the hermitian symmetric space associated to $`\mathrm{O}(2,n)`$. It can be realized as an open subset of the quadric defined by $`(z,z)=0`$ in the projective space $`P(V(C))`$, where we extend $`(,)`$ to a $`C`$-bilinear form on $`V(C)`$. Namely, it is one of the two connected components of the open subset defined by $`(z,\overline{z})>0`$. A subgroup $`\mathrm{O}^{}(V)`$ of index two of the orthogonal group $`\mathrm{O}(V)`$ acts biholomorphically on $`_n`$. Let $`\stackrel{~}{}_n`$ denote the inverse image of $`_n`$ in $`V(C)`$. We restrict henceforth to the case $`n>2`$ for convenience. If $`M`$ is an even integral $`Z`$-lattice in $`V`$, then a meromorphic automorphic form of weight $`kZ`$ with respect to a subgroup $`G`$ of finite index in
$$\mathrm{O}^{}(M)=\mathrm{O}(M)\mathrm{O}^{}(V)$$
and a character $`v`$ of $`G`$ is a meromorphic function $`f`$ on $`\stackrel{~}{}_n`$ with the properties a) $`f(\gamma z)=v(\gamma )f(z)`$ for all $`\gamma G`$. b) $`f(tz)=t^kf(z)`$ for all $`tC^{}`$. We next recall the notion of a Heegner divisor: let $`m`$ be a negative rational number and let $`\alpha `$ be an element of $`M^{}/M`$, where $`M^{}`$ denotes the dual lattice. The Heegner divisor $`H(\alpha ,m)_n`$ is the union of the orthogonal complements $`v^{}_n`$ where $`v`$ runs through all elements of $`M^{}`$ satisfying
$$v\alpha modM\text{and}(v,v)=2m.$$
We consider $`H(\alpha ,m)`$ as a divisor on $`_n`$ by attaching multiplicity $`1`$ to all components. It is obvious that $`H(\alpha ,m)=H(\alpha ,m)`$, so that the divisor depends only on $`m`$ and the image of $`\alpha `$ in $`(M^{}/M)/\pm 1`$. Borcherds introduced in \[Bo1\] a method for constructing automorphic forms on $`_n`$ whose divisors are sums of Heegner divisors. Then, in \[Bo2\], he constructed a ‘space of obstructions’ to the use of this technique for constructing automorphic forms with divisor equal to some given sum of Heegner divisors. This space consists of all elliptic modular forms of weight
$$k:=(2+n)/2$$
with respect to the dual $`\varrho ^{}:=\varrho _M^{}`$ of the Weil representation. We restrict to the case of even $`n`$ since the Weil representation simplifies and this is the only case we need. Such a form $`(f_\alpha )_{\alpha M^{}/M}`$ is required to be holomorphic at the cusp at infinity and satisfy the transformation laws
$$\begin{array}{ccc}& f_\alpha (\tau +1)=e^{\pi \mathrm{i}(\alpha ,\alpha )}f_\alpha (\tau )\hfill & 1.\hfill \\ & f_\alpha \left(\frac{1}{\tau }\right)=\sqrt{\frac{\tau }{\text{i}}}^{2+n}\frac{1}{\sqrt{|M^{}/M|}}\underset{\beta M^{}/M}{}e^{2\pi \mathrm{i}(\alpha ,\beta )}f_\beta (\tau ).\hfill & 2.\hfill \end{array}$$
As in section 4, we note that $`(\alpha ,\alpha )`$ is well-defined modulo 2 and $`(\alpha ,\beta )`$ is well-defined modulo 1, so that these formulas make sense. Elements of the space of obstructions can be constructed by means of Eisenstein series, as follows. We write $`R`$ for the group ring $`C[M^{}/M]`$ and $`R_0`$ for the subspace on which $`(1)^k\varrho ^{}(E)`$ acts trivially. Since $`(1)^k\varrho ^{}(E)`$ acts by exchanging $`e_\alpha `$ and $`e_\alpha `$, where the $`e_\alpha `$ form the standard basis of $`R`$ as $`\alpha `$ varies over $`M^{}/M`$, a basis for $`R_0`$ is given by the elements
$$e_\alpha +e_\alpha ,\alpha (M^{}/M)/\pm 1.$$
If $`\xi R_0`$ satisfies
$$\varrho ^{}\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)\xi =\xi $$
then $`(c\tau +d)^k\varrho ^{}(Q)^1\xi `$ remains unchanged if one replaces $`Q`$ by $`PQ`$, where $`P=\pm \left(\genfrac{}{}{0pt}{}{1}{0}\genfrac{}{}{0pt}{}{n}{1}\right)`$ lies in the stabilizer $`SL(2,Z)_{\mathrm{}}`$ of $`\mathrm{}`$ and $`Q=\left(\genfrac{}{}{0pt}{}{a}{c}\genfrac{}{}{0pt}{}{b}{d}\right)SL(2,Z)`$. This lets us define the Eisenstein series
$$E_\xi (\tau )=\underset{Q=\left(\genfrac{}{}{0pt}{}{a}{c}\genfrac{}{}{0pt}{}{b}{d}\right)SL(2,Z)_{\mathrm{}}\backslash SL(2,Z)}{}(c\tau +d)^k\varrho ^{}(Q)^1\xi ,$$
which is a modular form of weight $`k`$ with respect to $`\varrho ^{}`$, so it lies in the space of obstructions. Furthermore,
$$\underset{Im\tau \mathrm{}}{lim}E_\xi (\tau )=\xi .$$
In particular, if $`\xi 0`$ then $`E_\xi `$ is not a cusp form.
5.1 Remark. Under our assumption $`n>2`$ we have $`k>2`$, and in this case the space of Eisenstein series of weight $`k`$ and with respect to $`\varrho ^{}`$ is isomorphic the space of all $`\xi V_0`$ with
$$\varrho ^{}\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)\xi =\xi \left(\text{and }\varrho ^{}(E)\xi =(1)^k\xi \right).$$
Using Remark 5.1 one can reformulate a fundamental result of Borcherds \[Bo1\], \[Bo2\] as follows.
5.2 Theorem. Suppose $`D`$ is a finite $`Z`$-linear combination
$$\underset{\alpha (M^{}/M)/\pm 1,m<0}{}C(\alpha ,m)H(\alpha ,m)$$
of Heegner divisors and $`G`$ is the subgroup of $`\mathrm{O}^{}(M)`$ that acts trivially on $`M^{}/M`$. Then $`D`$ is the divisor of a meromorphic automorphic form on $`_n`$ with respect to some character of $`G`$ if for every cusp form $`f`$ in the space of obstructions, say
$$f_\alpha (\tau )=\underset{mQ}{}a_\alpha (m)\mathrm{exp}(2\pi \mathrm{i}m\tau ),$$
$`(5.1)`$
the relation
$$\underset{m<0,\alpha M^{}/M}{}a_\alpha (m)C(\alpha ,m)=0$$
$`(5.2)`$
holds. The weight of such an automorphic form is
$$\underset{mQ,\alpha M^{}/M}{}b_\alpha (m)C(\alpha ,m),$$
where $`b_\alpha (m)`$ denotes the Fourier coefficients of the (unique) Eisenstein series with constant term
$$b_\alpha (0)=\{\begin{array}{cc}1/2\hfill & \text{if }\alpha =0\text{,}\hfill \\ 0\hfill & \text{otherwise.}\hfill \end{array}$$
We want to apply this theorem to our lattice $`\mathrm{\Lambda }=^{1,4}`$, or rather to its underlying $`Z`$-lattice $`M`$. The obstructions have weight $`k=(2+8)/2=5`$, and if the space of obstructions vanished then the existence of the forms of Theorem 4.7 would follow easily from Theorem 5.2 by restriction from $`_8`$ to $`_4`$. There are obstructions, and even cuspidal obstructions, but we will show that there are no $`\mathrm{O}^{}(M)`$-invariant cusp forms in the space of obstructions, and this turns out to be enough to establish Theorem 4.7. Here, $`\mathrm{O}^{}(M)`$ acts via its action on $`M^{}/M`$.
5.3 Theorem. For $`M`$ equal to the $`Z`$-lattice underlying $`\mathrm{\Lambda }=^{1,4}`$, the space of $`\mathrm{O}^{}(M)`$-invariant obstructions has dimension two and is spanned by Eisenstein series. The space of invariant cuspidal obstructions vanishes.
Proof. The $`\mathrm{O}^{}(M)`$-invariant part of $`C[M^{}/M]`$ has dimension 4, because $`\mathrm{O}^{}(M)`$ acts with 4 orbits (or ‘types’) on $`M^{}/M`$. The type of an element $`\alpha M^{}/M`$ is defined as $`00`$ if $`\alpha `$ is the zero element and as $`t\{0,1,2\}`$ if $`\alpha `$ is different from zero and $`(\alpha ,\alpha )2t/3`$ mod $`2`$. There are 1, 80, 90 and 72 elements of $`M^{}/M`$ of types 00, 0, 1 and 2, respectively. We will express an invariant obstruction $`h`$ as $`(h_{00},h_0,h_1,h_2)`$, where each $`h_t`$ is the sum of the $`h_\alpha `$ as $`\alpha `$ varies over the elements of $`M^{}/M`$ of type $`t`$. A calculation allows one to determine the action of $`SL(2,Z)`$ with respect to this basis. It turns out that the standard generators $`T=\left(\genfrac{}{}{0pt}{}{1}{0}\genfrac{}{}{0pt}{}{1}{1}\right)`$ and $`S=\left(\genfrac{}{}{0pt}{}{0}{1}\genfrac{}{}{0pt}{}{1}{0}\right)`$ act by
$$\varrho ^{}(T)=\left(\begin{array}{cccc}1& & & \\ & 1& & \\ & & \omega ^2& \\ & & & \omega \end{array}\right)\text{and}\varrho ^{}(S)=\frac{\text{i}}{3^{5/2}}\left(\begin{array}{cccc}1& 1& 1& 1\\ 80& 1& 8& 10\\ 90& 9& 9& 0\\ 72& 9& 0& 9\end{array}\right).$$
Borcherds \[Bo2\] gives a formula for the dimension of the space of elliptic modular forms of given weight that tranform according to some given representation of $`SL(2,Z)`$, in terms of the eigenvalues of certain elements of $`SL(2,Z)`$. Applying this formula to the 4-dimensional representation above shows that the space of obstructions has dimension 2. On the other hand, the space of Eisenstein series is also 2-dimensional, because another calculation shows that the subspace of $`C[M^{}/M]^{\mathrm{O}^{}(M)}`$ whose elements satisfy the conditions of Remark 5.1 is 2-dimensional. Since a cuspidal Eisenstein series vanishes identically, the theorem follows. $``$$``$
5.4 Corollary. With $`M`$ as in Theorem 5.3, every divisor $`D`$ as in Theorem 5.2 that is $`\mathrm{O}^{}(M)`$-invariant is the divisor of a form on $`_8`$ that is automorphic with respect to some character of $`\mathrm{O}^{}(M)`$.
Proof. Since $`D`$ is $`\mathrm{O}^{}(M)`$-invariant it satisfies condition (5.2) for all $`f`$ as in (5.1) if and only if it satisfies the condition for all such $`f`$ that are also $`\mathrm{O}^{}(M)`$-invariant. Therefore the theorem assures us of the existence of an automorphic form for $`G\mathrm{O}^{}(M)`$ with divisor $`D`$, and since $`D`$ is $`\mathrm{O}^{}(M)`$-invariant the form must be automorphic with respect to some character of $`\mathrm{O}^{}(M)`$ itself. $``$$``$
In order to find the weights of the forms constructed in this way, we need the Eisenstein series with constant term $`b_\alpha (0)=1/2`$ for $`\alpha =0`$ and $`b_\alpha (0)=0`$ for other $`\alpha `$. To compute this series we construct a basis for the space of Eisenstein series and then take a suitable linear combination. The Weil representation factors through $`SL(2,Z/3Z)`$, so our Eisenstein series are linear combinations of the classical elliptic Eisenstein series of level $`3`$, namely
$$G_k(\tau ;c,d,N):=\stackrel{}{}_{\genfrac{}{}{0pt}{}{mc}{nd}modN}\frac{1}{(m\tau +n)^k},$$
where the level $`N`$ is 3 and the weight $`k`$ is 5. We write $`E_1`$, $`E_2`$, $`E_3`$ and $`E_4`$ for the four classical Eisenstein series corresponding to the values $`(c,d)=(0,1)`$, $`(1,0)`$, $`(1,1)`$ and $`(1,2)`$. We will continue to use the notation introduced in the proof of Theorem 5.3.
5.5 Proposition. With $`M`$ as in Theorem 5.3, a basis for the space of $`\mathrm{O}^{}(M)`$-invariant obstructions consists of the Eisenstein series $`f`$ and $`g`$ given by
$$\begin{array}{cc}\hfill f_{00}& =\frac{\text{i}\sqrt{3}}{18}E_1\frac{1}{18}(E_2+E_3+E_4)\hfill \\ \hfill f_0& =\frac{5\text{i}\sqrt{3}}{9}E_1+\frac{5}{9}(E_2+E_3+E_4)\hfill \\ \hfill f_1& =0\hfill \\ \hfill f_2& =E_2+\omega ^2E_3+\omega E_4\hfill \end{array}\begin{array}{cc}\hfill g_{00}& =\frac{\text{i}\sqrt{3}}{18}E_1+\frac{1}{18}(E_2+E_3+E_4)\hfill \\ \hfill g_0& =\frac{4\text{i}\sqrt{3}}{9}E_1+\frac{4}{9}(E_2+E_3+E_4)\hfill \\ \hfill g_1& =E_2+\omega E_3+\omega ^2E_4\hfill \\ \hfill g_2& =0.\hfill \end{array}$$
Proof. If $`h=(h_{00},h_0,h_1,h_2)`$ is an $`\mathrm{O}^{}(M)`$-invariant Eisenstein series then each component of $`h`$ is a $`C`$-linear combination of $`E_1,\mathrm{},E_4`$. The manner in which the $`E_i`$’s transform into each other under $`SL(2,Z)`$ is known, and the transformation laws of $`h`$ with respect to $`\varrho ^{}`$ reduce to a set of linear conditions on the coefficients of the $`E_i`$’s. One simply solves the system of linear equations. (Of course, once one has the answer one can simply check it.) $``$$``$
The Fourier coefficients of the Eisenstein series can be found in many text books, for example \[He, no. 24, section 1\] or \[Fr\]. This lets one find the Fourier expansions for $`f`$ and $`g`$; once these are known then one can find the unique obstruction $`h`$ whose Fourier coefficients $`b_\alpha (m)`$ have constant term as in Theorem 5.2. The answer turns out to be given by
$$\begin{array}{cc}\hfill h_{00}& =1/2+12q+225q^2+1092q^3+2892q^4+\mathrm{}\hfill \\ \hfill h_0& =1080q+16200q^2+87480q^3+260280q^4+673920q^5+\mathrm{}\hfill \\ \hfill h_1& =225q^{2/3}+9360q^{5/3}+57825q^{8/3}+219600q^{11/3}+540450q^{14/3}+\mathrm{}\hfill \\ \hfill h_2& =12q^{1/3}+2892q^{4/3}+28824q^{7/3}+112320q^{10/3}+342744q^{13/3}+\mathrm{}\hfill \end{array}$$
5.6 Proposition. With $`M`$ as in Theorem 5.3, there exists an automorphic form on $`_8`$ for $`\mathrm{O}^{}(M)`$, of weight $`12`$ (resp. $`225`$), whose zeros are the orthogonal complements of the vectors $`vM^{}`$ satisfying $`(v,v)=2/3`$ (resp. $`(v,v)=1/3`$). The vanishing order is one.
Proof. This follows from Theorems 5.2and 5.3. For the form of weight 12 (resp. 225) we take $`D`$ to be the sum of the $`H(\alpha ,m)`$ where $`m=1/3`$ (resp. $`m=2/3`$) and $`\alpha `$ varies over the type 2 (resp. type 1) elements of $`(M^{}/M)/\pm 1`$. $``$$``$
We note that the form of weight 12 was found by Borcherds in \[Bo3\]. Proof of Theorem 4.7. We use the natural embedding $`\mathrm{U}(1,4)\mathrm{O}^{}(2,8)`$ and a compatible holomorphic embedding $`_4_8`$. The short (resp. long) mirrors in $`_4`$ are the intersections of the divisors described in Prop. 5.6. The orthogonal complements in Prop. 5.6 occur in triples having the same intersection with $`_4`$, since if $`r`$ is a root of $`\mathrm{\Lambda }`$ then $`r`$, $`\omega r`$ and $`\omega ^2r`$ are roots with the same orthogonal complement in $`_4`$ but different orthogonal complements in $`_8`$. Therefore the vanishing order of the restriction to $`_4`$ along each short (resp. long) mirror is three. Taking a cube root yields a form of weight $`12/3=4`$ (resp. $`225/3=45`$). $``$$``$
6. A model for the moduli space of marked cubic surfaces
Recall the ten dimensional space $`W`$ of automorphic forms for $`\mathrm{\Gamma }`$, the congruence subgroup of level $`\sqrt{3}`$ in $`Aut(\mathrm{\Lambda })`$. We know from Theorem 4.6 that these forms have no common zero. Therefore, by choosing a basis for $`W`$ we obtain an everywhere holomorphic map
$$\beta :X=_4^{}/\mathrm{\Gamma }P^9(C).$$
This map is algebraic by Chow’s theorem. By a result of Hilbert it is a finite map. Hence the image is a projective algebraic variety $`𝒱P^9`$ of dimension 4. In fact more is true:
6.1 Theorem. The map $`\beta :X𝒱`$ is birational.
After proving this theorem we will introduce a family of cubic 8-folds, each of which contains $`𝒱`$. Then we will sketch a proof that these cubic equations actually define $`𝒱`$. Our proof of 6.1 uses only our automorphic forms. In section 7 we will prove that $`\beta `$ is actually an embedding, but this relies heavily on the very extensive calculations and involved arguments of \[Na\]. Theorem 6.1 follows immediately from the lemma:
6.2 Lemma. Let $`p`$ be the point of $`_4`$ represented by $`(1,0,0,0,0)\mathrm{\Lambda }`$, and let $`\overline{p}`$ denote its image in $`X`$. Then $`\overline{p}`$ is the only point of $`X`$ mapping to $`\beta (\overline{p})`$, and $`\beta :X𝒱`$ is a local diffeomorphism at $`\overline{p}`$.
Proof. The first claim is a consequence of the second part of Theorem 3.4. In order to prove the second claim we will find four elements of $`W`$, the sum of whose divisors in $`X`$ is a normal crossing divisor at $`\overline{p}`$. For this we will need coordinates around $`\overline{p}`$. Coordinates around $`p_4`$ may be taken to be $`z_1,\mathrm{},z_4C^4`$, with $`_i|z_i|^2<1`$ as in formula (3.1). The stabilizer $`\mathrm{\Gamma }_p`$ of $`p`$, which is generated (modulo scalars) by the triflections in the short roots
$`(0,1,0,0,0)`$, $`(0,0,1,0,0)`$, $`(0,0,0,1,0)`$ and $`(0,0,0,0,1)`$,
acts by multiplying the $`z_i`$ by cube roots of unity. It follows that local coordinates for $`X`$ near $`\overline{p}`$ are given by the functions $`w_i=z_i^3`$. The four short mirrors of $`X`$ passing through $`\overline{p}`$ are given by the equations $`w_i=0`$. The long mirrors in $`_4`$ that pass through $`p`$ are the mirrors of the 216 roots $`(0,a_1,\mathrm{},a_4)`$, where two of the $`a_i`$ are zero and the others are sixth roots of unity. To work out their images in $`X`$ it suffices to treat the case where the nonzero $`a_i`$ lie in $`\{\pm 1\}`$, since the orbit of these under $`\mathrm{\Gamma }_p`$ is the entire set of 216. It is easy to check that the mirror $`z_i=\pm z_j`$ in $`_4`$ projects to the mirror $`w_i=\pm w_j`$ in $`X`$. It follows that in our local coordinates in $`X`$, the 12 long mirrors through $`\overline{p}`$ are given by the equations $`w_i=\pm w_j`$ for the various pairs $`ij`$. We claim that for each long mirror $`m`$ of $`X`$ passing through $`\overline{p}`$, there is a cross $`C`$ containing it whose short mirrors do not pass through $`\overline{p}`$. To see this, consider the three crosses containing $`m`$. Because only two of the short mirrors passing through $`\overline{p}`$ are orthogonal to $`m`$, one of the three crosses contains neither of these mirrors. Since it cannot contain either of the other short mirrors, it has the desired property and we take it to be $`C`$. Now, it is easy to find four long mirrors $`m_1,\mathrm{},m_4`$ whose sum is a normal crossing divisor at $`\overline{p}`$, for example those given by $`w_1=\pm w_2`$ and $`w_3=\pm w_4`$. We let $`C_i`$ be crosses associated to the $`m_i`$ as above, and $`f_i`$ be automorphic forms associated to the $`C_i`$. Then the $`f_i`$ are necessarily linearly independent, and we extend them to a basis $`f_1,\mathrm{},f_{10}`$ of $`W`$. Of course, one of the $`f_i`$, say $`f_{10}`$, does not vanish at $`\overline{p}`$, and then $`f_1/f_{10},\mathrm{},f_9/f_{10}`$ are affine coordinates near $`\beta (\overline{p})P^9`$. It follows from the implicit function theorem and the fact that $`f_i`$ ($`i=1,\mathrm{},4`$) has only a simple zero along $`m_i`$ that $`\beta `$ is a local diffeomorphism as $`\overline{p}`$. $``$$``$
Next we will find some cubic relations satisfied by our automorphic forms; these define cubic 8-folds in $`P^9(C)`$ which contain $`𝒱`$. It is easy to explain the origin of these relations: it can happen that there are three crosses $`C_1`$, $`C_2`$ and $`C_3`$, and another three crosses $`C_1^{}`$, $`C_2^{}`$ and $`C_3^{}`$, such that as divisors in $`X`$ they satisfy
$$C_1+C_2+C_3=C_1^{}+C_2^{}+C_3^{}.$$
$`(6.1)`$
If $`F_i`$ and $`F_i^{}`$ are nonzero automorphic forms in the one-dimensional subspaces of $`W`$ associated to the $`C_i`$ and $`C_i^{}`$, then the divisors of $`F_1F_2F_3`$ and $`F_1^{}F_2^{}F_3^{}`$ are equal and therefore the two products coincide up to a scalar. This relation would be trivial if the $`C_i^{}`$ were obtained by permuting the $`C_i`$, but nontrivial relations do arise and can be found by studying the geometry of $`V`$. Here are some nontrivial cubic relations, which turn out to be the only ones. (Only trivial relations can be found if one plays the same game with pairs rather than triples of crosses.)
6.3 Lemma. Let $`(a_1,a_2,a_3,b_1,b_2)`$ be an ordered orthonormal basis of $`V`$, $`S_i`$ be the signed cross given by the basis $`\{a_i,a_{i+1}\pm b_1,a_{i1}\pm b_2\}`$, and $`S_i^{}`$ be the signed cross given by $`\{a_i,a_{i+1}\pm b_2,a_{i1}\pm b_1\}`$, where the subscript of $`a_{i\pm 1}`$ should be read modulo 3. Writing $`F_i`$ and $`F_i^{}`$ for the automorphic forms associated to $`S_i`$ and $`S_i^{}`$, we have $`F_1F_2F_3=F_1^{}F_2^{}F_3^{}`$.
Proof. We write $`C_i`$ and $`C_i^{}`$ for the crosses underlying $`S_i`$ and $`S_i^{}`$. It is easy to check that (6.1) holds, and it follows that $`F_1F_2F_3`$ is a constant multiple of $`F_1^{}F_2^{}F_3^{}`$. To determine the constant, let $`\alpha `$ be the isotropic vector $`a_1+a_2+a_3V`$ and let $`\stackrel{~}{\alpha }`$ be any primitive isotropic element of $`\mathrm{\Lambda }`$ representing $`\alpha `$. Using the product formula of Lemma 4.2, it is easy to see that the element of $`C[V]^{SL(2,F_3)}`$ associated to each $`S_i`$ and $`S_i^{}`$ has component $`1`$ at $`\alpha `$. By the relationship between the values of elements of $`W`$ at cusps of $`\stackrel{~}{}_4^{}`$ and at the corresponding elements of $`V`$ (see the proof of Prop. 4.4), all the $`F_i`$ and $`F_i^{}`$ take the same value at $`\stackrel{~}{\alpha }`$. Therefore $`F_1F_2F_3(\stackrel{~}{\alpha })=F_1^{}F_2^{}F_3^{}(\stackrel{~}{\alpha })`$, and so $`F_1F_2F_3=F_1^{}F_2^{}F_3^{}`$. $``$$``$
Remarks: We will discuss coincidences among these relations, and the fact that they account for all the relations arising from crosses $`C_i,C_i^{}`$ satisfying (6.1). If $`(\widehat{a}_1,\widehat{a}_2,\widehat{a}_3,\widehat{b}_1,\widehat{b}_2)`$ is another ordered orthonormal basis for $`V`$, then the relations given by the two bases are essentially the same if
$$\{\pm a_1,\pm a_2,\pm a_3\}=\{\pm \widehat{a}_1,\pm \widehat{a}_2,\pm \widehat{a}_3\}\text{and}\{\pm b_1,\pm b_2\}=\{\pm \widehat{b}_1,\pm \widehat{b}_2\}.$$
$`(6.2)`$
By “essentially the same” we mean that each relation implies the other. There are $`|\mathrm{O}(5,3)|/2^5\mathrm{\hspace{0.17em}3}!\mathrm{\hspace{0.17em}2}!=270`$ equivalence classes of ordered orthonormal bases under the relation (6.2), yielding 270 cubic relations. It is easy to make a computer enumerate all nontrivial pairs of triples of crosses $`C_i`$ and $`C_i^{}`$ satisfying (6.1) and check that every one is a case of our construction. Therefore we have found all the relations arising from equalities of sums of triples of crosses. For convenience in enumerating the 270 relations, we remark that they are in 1-1 correspondence with the unordered triples of mutually orthogonal long mirrors in $`X`$. To find the relation associated to such a triple of mirrors, let $`a_1`$, $`a_2`$ and $`a_3`$ be long vectors of $`V`$ associated to the mirrors, extend them to an orthonormal basis of $`V`$, and apply the lemma.
6.4 Theorem. The variety $`𝒱`$ is the intersection of the cubic eightfolds defined by the relations of Lemma 6.3.
Proof sketch. Using one of the computer algebra systems MACAULEY or SINGULAR, it is easy to see that the dimension of the intersection $`𝒱^{}`$ of the 270 cubics has dimension 4. With either system it is possible to compute a projective resolution of $`R/J`$, where $`R=Q[Y_0,\mathrm{},Y_9]`$, $`Y_0,\mathrm{},Y_9`$ are a basis for $`W`$, and $`J`$ the ideal generated by the 270 cubic relations. The projective dimension of $`R/J`$ turns out to be 5, by a calculation that takes a few minutes in SINGULAR but several hours in MACAULAY. As a consequence, $`𝒱^{}`$ contains no component of dimension $`<4`$. It is more involved to prove that $`𝒱^{}`$ is irreducible. In principle one can simply ask the machine, but this seems to be too much for the computer. Instead, we consider the intersection of $`𝒱^{}`$ with a hyperplane corresponding to a cross. If $`𝒱^{}`$ is irreducible then the intersection should consist of 5 irreducible components. It is not hard to prove that in our situation the converse is also true. The hyperplane section is defined by a certain ideal $`\text{a}C[Y_0,\mathrm{},Y_9]`$. In principle one can ask the computer for the components of the ideal (e.g. by using “decompose” in MACAULEY), but again this does not work. Instead, one finds directly five ideals $`\text{a}_0,\mathrm{},\text{a}_4`$ containing a that come from the five mirrors of the cross and are constructed in an obvious way. After the ideals $`\text{a}_i`$ have been constructed, one can verify $`\text{a}=\text{a}_0\mathrm{}\text{a}_4`$ by means of MACAULEY or SINGULAR. The problem now is to prove that the varieties of the $`\text{a}_i`$ are irreducible. This means that we face a similar problem in a lower dimension, which can be treated in a similar manner. During this procedure several very interesting ball quotients of smaller dimension occur. This will be treated in a separate paper, where more details about the ideal $`JC[Y_0,\mathrm{},Y_9]`$ and the hyperplane sections will be given. We also intend to include proofs of the facts that $`J`$ is prime and that $`R/J`$ is normal. This has the important consequence that $`W`$ generates the ring of all automorphic forms on $`\mathrm{\Gamma }`$ with trivial multipliers. The normality can be used to give an alternate proof of Theorem 7.3 (that $`\beta `$ is an embedding). We will also give the Hilbert function of $`R/J`$. $``$$``$
7. Cross Ratios
In this section we will relate our automorphic forms to the original invariants of a cubic surface, the cross-ratios of Cayley. These are rational functions on the moduli space of marked cubic surfaces that encode the manner in which the 27 lines on a cubic surface lie in $`P^3`$. We will show below that Cayley’s cross-ratios are ratios of certain pairs of our 270 automorphic forms. Then we will use this to prove that the map $`\beta :P^9`$ of section 6 is an embedding. Suppose that $`A`$ and $`B`$ are two crosses with the same long mirror $`m`$. By Lemma 4.5, the subspace of $`W`$ that changes sign under reflection in $`m`$ is 2-dimensional, and the automorphic forms coming from the six signed crosses of $`m`$ form a regular hexagon in this plane, centered at $`0`$. Now, $`A`$ and $`B`$ define two diameters of this hexagon, and we choose an endpoint $`F`$ (resp. $`G`$) of the diameter associated to $`A`$ (resp. $`B`$), such that $`F`$ and $`G`$ are adjacent vertices of the hexagon. There is a unique way to do this, up to simultaneously negating $`F`$ and $`G`$, so the rational function $`F/G`$ does not depend on the choice made. We call this the cross ratio $`A/B`$. The reason for the name is Theorem 7.2 below, which identifies these rational functions with Cayley’s cross-ratios. It is a genuine accident of terminology that our cross-ratios may also be regarded as ratios of crosses. There are 270 cross-ratios, six for each of the 45 long mirrors. To identify our cross-ratios with Cayley’s we will need to describe the divisor of $`A/B`$:
7.1 Lemma. If $`m`$ is a long mirror in $`X`$ and $`A`$, $`B`$ and $`C`$ are its three crosses, then the divisor of the cross-ratio $`A/B`$ consists of the four short mirrors of $`A`$ with multiplicity $`1`$ (simple zeros) and the four short mirrors of $`B`$ with multiplicity $`1`$ (simple poles). Furthermore, $`A/B`$ takes the constant value $`1`$ at generic points of the short mirrors of $`C`$.
Proof. If $`F`$ and $`G`$ are automorphic forms chosen as in the discussion above, then their divisors in $`X`$ are the crosses $`A`$ and $`B`$, respectively. Since the long mirrors of $`A`$ and $`B`$ coincide and the short mirrors are distinct, the identification of the divisor of $`A/B`$ is complete. Finally, $`H=FG`$ is an endpoint of the third diameter of the hexagon, so that it lies in the 1-dimensional subspace of $`W`$ associated to $`C`$, and in particular it vanishes on the short mirrors of $`C`$. That is, $`F=G`$ on the mirrors of $`C`$ and so $`A/B=1`$ at generic points of the short mirrors of $`C`$. $``$$``$
Now we discuss Cayley’s cross-ratios; our basic reference is Naruki’s extensive study of them and a compactification $`C`$ of the moduli space $`M`$ of marked smooth cubic surfaces that they define \[Na\]. The biregular action of $`W(E_6)`$ on $`M`$ extends to a biregular action on $`C`$, and the complement of $`M`$ in $`C`$ has 76 components, which fall into orbits of size 40 and 36 under $`W(E_6)`$. The components in the orbit of size 40 are all disjoint and can be blown down to points. The variety $`\stackrel{ˇ}{C}`$ obtained by this blowing-down is the standard Geometric Invariant Theory (GIT) compactification of $`M`$, with its natural $`W(E_6)`$-action. Now, $`M`$ is also $`W(E_6)`$-equivariantly isomorphic to the complement in $`X`$ of the short mirrors, and the inclusion of this space into $`X`$ is also the standard GIT compactification. It follows that $`X`$ is $`W(E_6)`$-equivariantly isomorphic to $`\stackrel{ˇ}{C}`$, with the 36 short mirrors corresponding to the images in $`\stackrel{ˇ}{C}`$ of the remaining 36 components of $`CM`$. Naruki describes $`M`$ in terms of a maximal torus $`T`$ of the simple Lie group $`D_4`$ of adjoint type. He writes $`\mathrm{\Delta }`$ for the union of the subtori which are the fixed-point sets of the 12 reflections of $`W(D_4)`$, and realizes $`M`$ as the blowup of $`T`$ at the identity element, minus the proper transforms of the 12 components of $`\mathrm{\Delta }`$. He introduces multiplicative characters $`\lambda `$, $`\mu `$, $`\nu `$ and $`\rho `$ of $`T`$, which provide a coordinate system for $`T`$, and describes the action of $`W(E_6)`$ on $`M`$ by giving explicit rational self-maps of $`T`$ in terms of these coordinates. This group $`W(E_6)`$ contains the obvious group $`W(D_4)`$ and also the larger group $`W(F_4)`$ obtained by adjoining the automorphisms of $`T`$ arising from the automorphisms of the Dynkin diagram $`D_4`$. Naruki introduces 45 divisors in $`M`$ which $`W(E_6)`$ permutes transitively. One of these, $`\delta _0`$, is the exceptional divisor lying over the identity of $`T`$, and the rest are given by explicit equations in $`\lambda `$, $`\mu `$, $`\nu `$ and $`\rho `$. We claim that these 45 divisors correspond to our long mirrors. To see this, observe that $`\delta _0`$ is the fixed-point set of the lift (say $`\eta `$) to $`M`$ of the negation map on $`T`$. (All of the 2-torsion points of $`T`$ lie in $`\mathrm{\Delta }`$.) The conjugacy class of $`\eta `$ has size at most 45, since $`\eta `$ centralizes $`W(F_4)`$, of index 45 in $`W(E_6)`$, and at least 45, since $`\delta _0`$ has 45 translates under $`W(E_6)`$. Since $`W(E_6)`$ has a unique conjugacy class of involutions of size 45, and the elements of this class are our long reflections, $`\eta `$ corresponds to a long reflection $`\widehat{\eta }`$. Furthermore, $`\delta _0`$ corresponds to the fixed-point set of $`\widehat{\eta }`$, which must therefore be irreducible and (since it contains a long mirrors) consist entirely of the long mirror. Another way to prove our claim is to use the fact that each of \[Na\] and \[ACT\] proves that its set of 45 divisors represent the marked cubic surfaces that have an Eckardt point. The passage from $`T`$ to $`\stackrel{ˇ}{C}`$ involves compactifying $`T`$ and then performing a sequence of blowings-up and blowings-down. All that matters to us is that the identity of $`T`$ is blown up, and that the 12 components of $`\mathrm{\Delta }`$ (or rather their transforms in $`\stackrel{ˇ}{C}`$) are among the the 36 components of $`\stackrel{ˇ}{C}M`$. Naruki calls these 12 divisors the $`A_1`$-hypersurfaces. Finally, Naruki’s table 2 gives 45 of Cayley’s cross-ratios explicitly as rational functions of $`\lambda `$, $`\mu `$, $`\nu `$ and $`\rho `$. The full set of Cayley’s 270 cross-ratios is obtained by following these functions by the 6 projective linear transformations of $`P^1=C\{\mathrm{}\}`$ that preserve $`\{0,1,\mathrm{}\}`$. Of course, Cayley had much more explicit geometric concepts in mind when defining his cross-ratios; for details see Naruki’s paper.
7.2 Theorem. Cayley’s cross-ratios coincide with ours.
Proof: The idea is to check that the divisors coincide and that Cayley’s cross-ratios satisfy the normalization condition of lemma 7.1. By Cayley’s geometric considerations (\[Na\], §3), his cross-ratios do not take any of the values $`0`$, $`1`$ and $`\mathrm{}`$ in $`M`$, so their divisors consist of short mirrors with some multiplicities. For the short mirror $`S`$ given by $`\rho =1`$ in Naruki’s coordinates, simple substitution reveals the behavior along $`S`$ of the 45 cross-ratios given in Naruki’s table. Namely, exactly 7 vanish along it, exactly 7 have poles along it, and just one takes the constant value 1. Since the full set of Cayley’s cross-ratios is obtained by following these by the 6 linear fractional transformations preserving $`\{0,1,\mathrm{}\}`$, we see that exactly $`2(7+7+1)=30`$ of Cayley’s cross-ratios vanish along $`S`$, another 30 take the constant value 1, and a further 30 have poles along $`S`$. (Working with the full set of 270 restores the symmetry between $`0`$, $`1`$ and $`\mathrm{}`$ that Naruki’s choice of 45 conceals.) Now, by the transitivity of $`W(E_6)`$ on Cayley’s cross-ratios, each vanishes along the same number, say $`k`$, of short mirrors. By transitivity on the short mirrors, each short mirror lies in the zero-locus of exactly 30 of Cayley’s cross-ratios. These transitivities also show that $`270k=3630`$, so that $`k=4`$ and each of Cayley’s cross-ratios vanishes along exactly 4 short mirrors. The same argument also shows that each has poles along exactly 4 short mirrors. Now we consider Cayley’s cross-ratio $`r(\mathrm{w})`$, given in Naruki’s coordinates by
$$r(\mathrm{w})=\frac{(\lambda \rho 1)(\mu \rho 1)(\nu \rho 1)(\lambda \mu \nu \rho 1)}{(\mu \nu \rho 1)(\lambda \nu \rho 1)(\lambda \mu \rho 1)(\rho 1)}.$$
We will write simply $`r`$ for $`r(\mathrm{w})`$. The sets $`\chi =1`$, where $`\chi `$ is one of the characters $`\lambda \rho `$, $`\mu \rho `$, $`\nu \rho `$ and $`\lambda \mu \nu \rho `$ (resp. $`\mu \nu \rho `$, $`\lambda \nu \rho `$, $`\lambda \mu \rho `$ and $`\rho `$) appearing in the numerator (resp. denominator) are among Naruki’s $`A_1`$-hypersurfaces, so $`r`$ has a simple zero (resp. simple pole) along these four short mirrors. By the argument above, these constitute the entire divisor of $`r`$. Furthermore, the short mirrors along which $`r`$ vanishes (resp. has a pole) are orthogonal, in the sense that the reflections across them commute. To see this we do not even need to perform a calculation, because Naruki (p. 20) has already organized his twelve $`A_1`$-hypersurfaces into three sets each consisting of four mutually orthogonal divisors. Finally, all 8 of these short mirrors are orthogonal to the long mirror $`\delta _0`$, because their reflections obviously commute with the negation map of $`T`$. It follows that $`\delta _0`$ together with the four short mirrors coming from the numerator (resp. denominator) of $`r`$ form a cross $`C_n`$ (resp. $`C_d`$). Therefore $`r`$ has the same divisor as our cross-ratio $`C_n/C_d`$. To show that $`r=C_n/C_d`$ it now suffices to show that $`r=1`$ along the short mirrors of the third cross associated to $`\delta _0`$. Consulting again the table on Naruki’s p. 20, we see that these mirrors are given by $`\chi =1`$, where $`\chi `$ varies over the characters $`\lambda `$, $`\mu `$, $`\nu `$ and $`\lambda \mu \nu \rho ^2`$. Simple calculation verifies the condition, so $`r=C_n/C_d`$. Since one of Cayley’s cross-ratios coincides with one of ours, and $`W(E_6)`$ acts transitively on both sets of cross-ratios, the theorem follows. $``$$``$
Remark: B. van Geemen has also obtained this theorem, as a byproduct of a larger investigation. His idea is to construct and study the linear system on Naruki’s model of the moduli space that comes from our space $`W`$ of automorphic forms. After one understands this linear system (van Geemen identifies it with one introduced by Coble long ago), the result above follows immediately. His approach also has the advantage of allowing one to relate the moduli space $``$ to the variety $`𝒱`$ over fields other than $`C`$. (Note that $`𝒱`$ is defined over $`Z`$.)
7.3 Corollary. The map $`\beta :XP^9`$ of section 6 is an embedding.
Proof: We write $`X`$ for $`\overline{_4/\mathrm{\Gamma }}_4/\mathrm{\Gamma }`$, the set of 40 cusp points. One of Naruki’s main results is that the 270 cross-ratios, a priori defined as maps $`M(P^1\{0,1,\mathrm{}\})`$, extend to regular maps $`(XX)P^1`$ that embed $`XX`$ in $`(P^1)^{270}`$. Since the cross-ratios are quotients of the elements of $`W`$, $`\beta `$ must embed $`XX`$ in $`P^9`$. Unfortunately, this argument cannot be extended to show that $`\beta `$ embeds all of $`X`$ into $`P^9`$; the problem is that one must blow up the points of $`X`$ in order for the rational map from $`X`$ to $`(P^1)^{270}`$ to become regular. In order to prove the theorem we will first show that $`\beta `$ is injective as a map of sets, and then that $`\beta `$ is a local embedding at each point of $`X`$. The injectivity has essentially already been proven: Theorem 3.4 shows that for each $`xX`$, $`x`$ is the only point of $`X`$ that lies on all the crosses containing $`x`$. It follows that no point of $`X`$ is identified under $`\beta `$ with any other point of $`X`$. Since $`\beta `$ is already known to be injective on $`XX`$, $`\beta `$ is injective. Now we prove that $`\beta `$ is a local embedding at each point of $`X`$; we will use Naruki’s explicit description (see \[Na\], section 12) of these singularities. Namely, his $`T`$-equivariant compactification $`\stackrel{~}{T}`$ of $`T`$ adjoins 48 divisors, 24 of which he then blows down to obtain 24 of the points of $`X`$. Focusing on one of these divisors, which he denotes by $`\overline{\rho =0}`$ and we will denote by $`D`$, he gives 8 characters of $`T`$ which extend to regular functions $`z_1,\mathrm{},z_8`$ on a neighborhood $`𝒰`$ of $`D`$ in $`\stackrel{~}{T}`$, and which vanish along $`D`$. According to his theorem 12.1, the induced map $`𝒰C^8`$ gives the blowing-down of $`D`$ and thus embeds a neighborhood of the resulting singular point $`xX`$ into $`C^8`$. Furthermore, he explicitly describes the singularity as the cone on the Veronese embedding of $`P^1\times P^1\times P^1`$ in $`P^7`$. This makes it a simple matter to see that the divisor of each $`z_i`$ near $`x`$ has exactly three components, and these components meet each other away from $`x`$ as well as at $`x`$. Since each $`z_i`$ is the extension of a character of $`T`$, its divisor can consists only of the components of $`XM`$, which is to say, short mirrors. Since short mirrors that meet each other in $`XX`$ must be orthogonal, we have shown that the divisor of each $`z_i`$ near $`X`$ consists of three mutually orthogonal short mirrors. For each $`i=1,\mathrm{},8`$, we will find an automorphic form $`\psi _iW`$ whose divisor near $`x`$ coincides with that of $`z_i`$. We may also choose $`\psi ^{}W`$ whose divisor misses $`x`$ entirely. Then the evaluation of $`\psi _1/\psi ^{},\mathrm{},\psi _8/\psi ^{}`$ provides essentially the same map (of some neighborhood of $`x`$) into $`C^8`$ as Naruki’s. It follows that $`\beta `$ must embed a neighborhood of $`x`$ into $`P^9`$. All that remains is to show that if $`xX`$ and $`m_1`$, $`m_2`$ and $`m_3`$ are any three mutually orthogonal short mirrors that all meet $`x`$, then there exists $`\psi W`$ whose divisor near $`X`$ is just the sum of the $`m_i`$. We choose a primitive null vector $`v\mathrm{\Lambda }`$ representing $`x`$, and short roots $`r_iv^{}`$ whose mirrors represent the $`m_i`$. Denoting the images of these vectors in $`V`$ by $`\overline{x}`$ and $`\overline{r}_i`$, we may choose coordinates in $`V`$ so that the inner product is given by
$$(a,b)=a_0b_0a_1b_1\mathrm{}a_4b_4,$$
and $`\overline{r}_1=(0,0,1,0,0)`$, $`\overline{r}_2=(0,0,0,1,0)`$, $`\overline{r}_3=(0,0,0,0,1)`$ and $`\overline{v}=(1,1,0,0,0)`$. The standard cross, given by the pairs $`(\pm 1,0,0,0,0),\mathrm{},(0,0,0,0,\pm 1)`$, is the divisor of one of our Borcherds products, which we take to be $`\psi `$. It is obvious that the divisor of $`\psi `$ contains the $`m_i`$. To show that the other components of the divisor miss $`x`$, we observe that these components correspond to the orthogonal complements of the roots of $`\mathrm{\Lambda }`$ whose images in $`V`$ are $`(\pm 1,0,0,0,0)`$ or $`(0,\pm 1,0,0,0)`$. Any such root has inner product $`0`$ (mod 3) with $`v`$, so its mirror cannot contain $`v`$. $``$$``$
Literature
\[ACT1\] Allcock, D. Carlson, J. and Toledo, D.: A complex hyperbolic structure for the moduli of cubic surfaces, Comptes Rendus de l’Aacademie Scientifique Francaise 326, ser I, 49–54 (1988) (alg geom /970916)
\[ACT2\] Allcock, D. Carlson, J., and Toledo, D.: The complex hyperbolic geometry of the moduli space of cubic Surfaces, preprint
\[BB\] Baily, W.L., and Borel,A.: Compactification of arithmetic quotients of bounded symmetric domains, Annals of Math. 84, No 3, 442–528 (1966)
\[Bo1\] Borcherds, R.: Automorphic forms with singularities on Grassmannians, Invent. math. 132, 491–562 (1998) http://www.dpmms.cam.ac.uk/home/emu/reb/.my-home-page.html
\[Bo2\] Borcherds, R.: The Gross-Kohnen-Zagier theorem in higher dimensions, Duke Math. J. 97, No 219–233 (1999) http://www.dpmms.cam.ac.uk/home/emu/reb/.my-home-page.html
\[Bo3\] Borcherds, R.: An automorphic form related to cubic surfaces, 1997 (unpublished).
\[C\] Conway, J. H. et. al.: Atlas of Finite Groups, Oxford University Press, 1985.
\[EHV\] Eisenbud, D., Huneke, C., and Vasconcelos, W.: Direct methods for primary decomposition, Inv. Math 110, No 2, 207–235 (1992)
\[Fr\] Freitag, E.: Modular forms related to cubic surfaces, in preperation.
\[He\] Hecke, E.: Mathematische Werke, Göttingen Vandenhoeck & Ruprecht (1959)
\[Na\] Naruki, I.: Cross ratio variety as moduli space of cubic surfaces, Proc. London Math. Soc. (3) 45, 1–30 (1982)
\[Sh\] Shimura, G.: The arithmetic of automorphic forms with respect to a unitary group, Annals of Math. 107, 596–605 (1978)
|
warning/0002/quant-ph0002083.html
|
ar5iv
|
text
|
# Abstract
## Abstract
We show that the potential wells $`V(x)=x^{10}+ax^8+bx^6+cx^4+dx^2+f/x^2`$ with a central spike possess arbitrary finite multiplets of elementary exact bound states. Their strong asymptotic growth implies an ambiguity in their $`𝒫𝒯`$ symmetrically generalized quantization (via complex boundary conditions) but the three eligible recipes coincide at our exceptional solutions.
PACS 03.65.Ge
## 1 Introduction
Realistic calculations in quantum physics and field theory are often guided by a parallel study of a simplified quantum-mechanical model in one dimension. A typical example may be found, e.g., in ref. where a self-interacting scalar field theory is being modified in non-perturbative manner. It is underlined there that one has to pay due attention not only to the interaction potential $`V(x)`$ itself but also to the boundary conditions imposed upon solutions in infinity.
In general the latter conditions are not unique. One of the most transparent explicit examples of their ambiguity has been offered by Bender and Turbiner . In essence, they considered a partially solvable sextic potential $`V(x)=x^63x^2`$ and its zero-energy ground-state wave function $`\psi (x)=\mathrm{exp}(x^4/4)`$. In this model it is obvious that once you start working in the whole complex plane, $`x=\varrho e^\phi lC`$, the asymptotic normalizability of the wave function keeps satisfied not only on the real line but, in general, within the four different asymptotic wedges defined by the elementary equation $`\mathrm{Re}(x^4)<0`$. This gives the admissible angles $`\phi S_k`$, $`k=1,2,3,4`$ lying within the four separate intervals,
$$\begin{array}{c}S_1=(\pi /8,\pi /8),S_2=(3\pi /8,5\pi /8),\hfill \\ S_3=(7\pi /8,9\pi /8),S_4=(11\pi /8,13\pi /8).\hfill \end{array}$$
(1)
Schematically, the situation is depicted in Figure 1. One imagines that the eligible complex physical coordinates can be arbitrary curves with ends which do not enter the “forbidden” asymptotic domains. Thus, the current real line starts at $`\phi _LS_3`$ in the left infinity and ends at $`\phi _RS_1`$. Its slightly bizarre “Wick-rotated” alternative $`S_4+S_2`$ has also been discussed in the literature as a weird example of a system whose real spectrum is not bounded below . The elementary zero-energy bound state itself is shared by both these non-equivalent spectra.
From a retrograde point of view the choice of the sextic $`V(x)`$ proved unfortunate since its menu (1) is not yet rich enough. Further progress has only been achieved five years later when Bender and Boettcher came to the conclusion that a privileged role must be played by the pairs of sectors with a mirror left-right symmetry. This symmetry keeps the trace of its origin in field theory and is called $`𝒫𝒯`$ symmetry. In the quantum mechanical context the Hamiltonians $`H`$ have to commute with the product of $`𝒫`$ (parity) and $`𝒯`$ (complex conjugation or time reflection) . It is currently believed that this guarantees the reality of the spectrum for many non-Hermitian complex Hamiltonians .
In the very special class of these models $`V(x)=x^2(ix)^{2\delta }`$ the complex plane is to be cut upwards in order to keep the picture unique. Bender and Boettcher started from the smallest exponents $`\delta `$ and picked up the lower sectors
$$S_L=(3\mathrm{\Delta }\pi /2,\mathrm{\Delta }\pi /2),S_R=(\mathrm{\Delta }\pi /2,\mathrm{\hspace{0.17em}3}\mathrm{\Delta }\pi /2)$$
with the half-width $`\mathrm{\Delta }=\pi /(4+2\delta )`$. Of course, for any $`\delta >1/2`$ there emerges an alternative pair of sectors
$$S_{L2}=(5\mathrm{\Delta }\pi /2,3\mathrm{\Delta }\pi /2),S_{R2}=(3\mathrm{\Delta }\pi /2,\mathrm{\hspace{0.17em}5}\mathrm{\Delta }\pi /2)$$
which has been used by Buslaev and Grecchi in their study of asymptotically quartic potentials . In general the latter possibility remains compatible with the current real coordinates in the interval of $`\delta (1,3)`$.
For all the asymptotically power-law models the ambiguity of quantization is an interesting phenomenon. Numerically, this has been documented by several studies which made use of the limiting transition $`\delta \mathrm{}`$ or of the absence of the cut at $`\delta =1,2,\mathrm{}`$ . An even simpler form of the $`\delta `$ dependence takes place at the positive integers $`Z=1+\delta /2`$ in the potentials $`V(x)=x^{4Z2}+𝒪(x^{4Z3})`$ since, asymptotically, their wave functions $`\psi (x)\mathrm{exp}(x^{2Z}/2Z)`$ are symmetric on the real line.
There appear the two new $`𝒫𝒯`$ symmetric pairs of sectors $`(S_L,S_R)`$ at each odd value of the integer $`Z`$. Thus, the $`\delta =4`$ and $`Z=3`$ decadic oscillator is the first model with an ambiguity of this type. This is illustrated in Figure 2 where the single left-right pair of the asymptotic boundary conditions of Figure 1 is replaced by the triple choice. All the three angles $`\phi _LS_{Lj}`$ and $`\phi _RS_{Rj}`$ with $`j=1,2`$ or $`3`$ are equally compatible with the normalizability of the exponential $`\psi (x)\mathrm{exp}(x^6/6)`$. With this motivation we shall consider here all the decadic polynomial potentials complemented by a central spike,
$$V(r)=r^{10}+ar^8+br^6+cr^4+dr^2+f/r^2.$$
(2)
These forces contain as many as five independent coupling constants and their real forms have been studied by several authors in the literature . Here, we shall re-consider these “spiked decadic” interactions within the generalized quantum mechanics of Bender et al . It in effect weakens the current Hermiticity of the Hamiltonian to its mere $`𝒫𝒯`$ symmetry. For this reason one can construct more solutions in principle. We shall see below that such an expectation is well founded, indeed.
## 2 Decadic oscillators
The results of study of Hermitian Hamiltonians with interactions (2) were summarized by Ushveridze in sec. 2.4 of his monograph on the partially (so called quasi-exactly) solvable models. This summary implies that the Hermitian decadic force lies somewhere in between the numerous purely numerical models (for which “it seems absolutely unrealistic to find an exact solution” ) and quasi-exactly solvable models in the narrower sense (where one requires the algebraic solvability for a multiplet of K states at any pre-determined integer K).
The latter category (say, type-I QES) lies already quite close to the harmonic and other completely solvable models. Its properties are best exemplified by the standard Hermitian sextic model (cf. eq. (1.4.13) and related discussion in ref. ). The former extreme without any solvability is usually illustrated by the Hermitian quartic oscillator (cf. section 1.3 in ref. ). In this comparison, the “intermediate” decadic models (e.g., eq. (2.4.5) in the Ushveridze’s book) admit merely a few ($`K`$) exact bound states with a strongly limited multiplicity $`K5`$ . Such a property (let’s call it QES of type II) is entirely trivial at $`K=1`$ and still quite easily achieved at the first few $`K>1`$ .
### 2.1 $`𝒫𝒯`$ symmetric regularization
The best illustration of influence of the replacement of Hermiticity by the mere $`𝒫𝒯`$ symmetry of the Hamiltonian is provided by the popular quartic oscillators. In ref. these potentials were shown to belong to the type I of the QES category. Obviously, $`𝒫𝒯`$ symmetry plays a crucial role in such an improvement of solvability.
The picture will be completed in what follows. We are going to demonstrate that a $`𝒫𝒯`$ symmetrization of eq. (2) leads still to the maximal, type-I form of the QES property. We shall see that for the complexified decadic oscillators the integer K may be chosen arbitrarily large. It is worth noting that the same enhanced solvability admitting the arbitrarily large multiplets remains freely applicable in any spatial dimension $`D`$.
In a preparatory step of the explicit constructions let us remind the reader that in any ( = $`\mathrm{}`$-th) partial-wave projection the $`D`$dimensional differential Schrödinger equation with a virtually arbitrary regular central potential reads
$$\left[\frac{d^2}{dr^2}+\frac{L(L+1)}{r^2}+V(r)\right]\psi (r)=E\psi (r),L=\mathrm{}+(D3)/2.$$
(3)
In a more general perspective, we can add a centrifugal-like spike to the regular force. This is known to preserve the exact solvability of the spiked harmonic oscillator . In a close parallel, we do not need to change the notation too much, re-defining only the angular momenta $`L=L(f)`$ in such a way that
$$L(L+1)=f+\left(\mathrm{}+\frac{D3}{2}\right)\left(\mathrm{}+\frac{D1}{2}\right).$$
(4)
As a consequence, equation (3) may be assigned a pair of independent solutions with the well known behaviour near the central singularity $`1/r^2`$,
$$\psi _1(r)r^{L(f)},\psi _2(r)r^{L(f)+1}.$$
In such a setting the “forgotten possibility” of a suitable ansatz lies in the simultaneous use of both these solutions in the (complex) vicinity of the origin,
$$\psi (r)𝒞_1r^L[1+𝒪(r^2)]+𝒞_2r^{L+1}[1+𝒪(r^2)].$$
(5)
Similar idea is slightly counterintuitive but it has already been used in several papers on the quartic oscillators with $`\delta =1`$. In 1993, Buslaev and Grecchi paved the way by the mathematically rigorous example of introduction of the complex coordinates for a problem of the present type. They achieved a regularization of the centrifugal spike by a constant imaginary shift of the real axis,
$$r=r(x)=xi\epsilon ,x(\mathrm{},\mathrm{}),\epsilon >0.$$
(6)
This makes the centrifugal spikes smooth and fully regular at all $`x`$,
$$\frac{L(L+1)}{(xi\epsilon )^2}=\frac{L(L+1)(x+i\epsilon )^2}{(x^2+\epsilon ^2)^2}=\frac{L(L+1)}{\epsilon ^2}+𝒪(x^2).$$
A longer discussion of some consequences of this type of the $`𝒫𝒯`$ symmetric regularization has been provided by ref. . There, both the even and odd wave functions of the exactly solvable harmonic oscillator and other models were assigned their separate analytic continuations to the complex $`x`$. In accord with expectations, the spectrum of the energies remained real.
In the present paper we shall try to follow the same pattern and imagine that the two-term ansatz (5) may remain compatible with the single Taylor-series expansion of $`\psi (r)`$ (say, in the powers of $`r^2`$) whenever the ratio $`r^{L+1}/r^L`$ is itself equal to an integer power of $`r^2`$. In the other words, under a suitable convention $`L+1>l`$, our key assumption will read $`L+1/2=M`$ where $`M`$ can only be a positive integer, $`M=1,2,\mathrm{}`$.
We shall see below that this type of constraint will lead to significant simplifications of the solutions as well as to their easier interpretation. Indeed, the trivial choice of $`f=0`$ implies that $`M=\mathrm{}1+D/2=1`$. This mimics the four-dimensional $`s`$wave or two-dimensional $`p`$wave situation since, in both cases, $`L=1/2`$. Similarly we arrive at the alternative choice between the six-dimensional $`s`$wave, four-dimensional $`p`$wave or two-dimensional $`d`$wave Schrödinger equation at $`M=2`$, etc.
### 2.2 Correct asymptotics and recurrences
In 1998, Bender and Boettcher discovered the partial solvability of the $`𝒫𝒯`$ symmetrized quartic polynomial oscillators. In the present language this means that they just employed the ansatz of the type (5) at the particular $`L=0`$. This has been accompanied by the asymptotically bent choice of the complex integration path $`rr(x)|C`$. In a way inspired by ref. the latter construction has been extended to all $`L`$ in our recent remark .
It is amusing to summarize that except the pioneering sextic study and its harmonic-oscillator simplification , virtually all the available papers on the (partially) solvable $`𝒫𝒯`$ symmetric polynomial potentials pay an exclusive attention to the asymmetric models $`V(r)V(r)`$ of degree two and four . In this way all of them avoid the ambiguity problem but necessitate the complex couplings and acquire the counter-intuitive property $`\mathrm{Im}V(t)0`$ even on the real axis of coordinates $`t`$.
The unpleasant asymmetries disappear within the decadic model (2) which is such that $`V(r)=V(r)`$. Its asymptotic growth is also steeper than in the current solvable models . In what follows, we are going to show that this model represents in fact the “missing” last item in a list of all the quasi-exactly solvable polynomial models. The first step towards this not quite predictable result lies in the manifestly normalizable ansatz
$$\psi (r)=\mathrm{exp}\left(\frac{r^6}{6}\alpha \frac{r^4}{4}\beta \frac{r^2}{2}\right)\underset{n=0}{\overset{N1}{}}h_nr^{2nL}.$$
(7)
The insertion of this formula converts our differential Schrödinger equation (3) into the finite set of recurrences
$$A_nh_{n+1}+B_nh_n+C_nh_{n1}+D_nh_{n2}=0,n=0,1,\mathrm{},N.$$
(8)
The use of the asymptotically optimal WKB-inspired parameters
$$a=2\alpha ,b=\alpha ^2+2\beta ,c=c(N)=2\alpha \beta +2M4N2$$
enables us to simplify the coefficients significantly,
$$\begin{array}{c}A_n=(2n+2)(2n+22M),B_n=E\beta (4n+22M),\\ C_n=\beta ^2d\alpha (4n2M),D_n=4(N+1n).\end{array}$$
(9)
This is the concise, linear algebraic formulation of our present problem.
## 3 Terminating solutions
Centrifugal spike in eq. (3) binds the integers $`M`$ to its strength $`f0`$ in a way prescribed by formula (4). This means that certain non-vanishing spikes can emerge as the simple functions of the dimension and angular momentum,
$$f=M^2(\mathrm{}1+D/2)^2.$$
In particular, the most elementary choice of $`D=M=1`$ and $`\mathrm{}=0`$ implies that we just have to fix $`f=3/4`$. This value is, by the way, precisely equal to a boundary of a certain mathematical regularity domain (cf. our recent remark for more details).
### 3.1 Sturmian multiplets of couplings at $`M=1`$
We may notice that in accord with our definitions (9) two coefficients vanish completely, $`A_1=0`$ and $`D_{N+1}=0`$. This is vital for the consistency of our ansatz (7). The third consequence of our assumptions reads $`A_{M1}=0`$ and has no obvious interpretation. Seemingly, it is redundant. Let us now pay more attention to its crucial and beneficial role.
Starting from the first nontrivial choice of $`M=1`$ we get simply $`A_0=0`$. In such a case we can fix $`E=0`$ and discover that the whole over-determined set of our recurrences (8) degenerates to the mere square-matrix recipe. As long as $`C_n=C_n(d)=\beta ^2\alpha (4n2M)d`$ we can write
$$\left(\begin{array}{ccccc}C_1(0)& B_1& A_1& & \\ D_2& C_2(0)& \mathrm{}& \mathrm{}& \\ & \mathrm{}& \mathrm{}& B_{N2}& A_{N2}\\ & & D_{N1}& C_{N1}(0)& B_{N1}\\ & & & D_N& C_N(0)\end{array}\right)\left(\begin{array}{c}h_0\\ h_1\\ h_2\\ \mathrm{}\\ h_{N1}\end{array}\right)=d\left(\begin{array}{c}h_0\\ h_1\\ h_2\\ \mathrm{}\\ h_{N1}\end{array}\right).$$
(10)
We see that a more or less routine diagonalization of a four-diagonal matrix determines in principle the $`N`$ different eigen-couplings $`d=d_k`$ with $`k=1,2,\mathrm{},N`$. These values may be found numerically at an arbitrary $`N`$. In the other words, the whole problem becomes quasi-exactly solvable of type I. This is our first important result.
It makes sense to introduce a shifted coupling $`F=d\beta ^2+2N\alpha `$. For the first few smallest dimensions $`N0`$ this simplifies the formulae and leads to the transparent implicit definitions of the shifted couplings $`F(d,N)`$,
$$\begin{array}{c}F=0,N=1,\\ F^2+16\beta 4\alpha ^2=0,N=2,\\ F^3+\left(16\alpha ^264\beta \right)F+256=0,N=3,\\ F^4+\left(160\beta 40\alpha ^2\right)F^21536F+144\alpha ^41152\beta \alpha ^2+2304\beta ^2=0,N=4,\\ F^5+\left(80\alpha ^2320\beta \right)F^3+5376F^2+\left(8192\beta \alpha ^21024\alpha ^416384\beta ^2\right)F\\ +196608\beta 49152\alpha ^2=0,N=5,\\ \mathrm{}\end{array}$$
We may summarize that at $`M=1`$, our solutions remain purely non-numerical up to the degree $`N=4`$. In a sufficiently broad part of the $`(\alpha ,\beta )`$ plane (or, if you wish, of the $`(a,b)`$ plane of the octic and sextic coupling constants) the multiplets of exact and elementary bound states exist and are numbered by their (real) quadratic couplings $`d_n`$ at any value of the multiplicity $`N`$.
### 3.2 Multiplets of energies at $`M=2`$
We have seen that our $`M=1`$ multiplets have been formed by the so called Sturmian solutions. The role of their energy was marginal and fixed to the single value $`E=0`$. Returning now back to our main story we have to move to the next integer $`M=2`$. This replaces the above-mentioned choice of $`E=0`$ (i.e., in effect, of $`B_0=0`$) by the virtually equally efficient simplification
$$det\left(\begin{array}{cc}B_0& A_0\\ C_1& B_1\end{array}\right)=0$$
which leads immediately to the compact formula
$$d=d(E)=\frac{E^2}{4},M=2.$$
The insertion of this energy-dependent harmonic strength $`d`$ returns us back to the square-matrix secular equation (10). With its three innovated diagonals
$$A_n=(2n+2)(2n2),B_n=E\beta (4n2),C_n=\beta ^2E^2/4\alpha (4n4)$$
it defines the spectrum $`E_n`$ in the very similar manner as above, i.e., as roots of a certain polynomial. We can display its first nontrivial sample,
$$\begin{array}{c}E^52\beta E^4+\left(48\alpha +8\beta ^2\right)E^3+\left(96\beta \alpha +192+16\beta ^3\right)E^2\\ +\left(256\beta 512\alpha ^2+192\beta ^2\alpha 16\beta ^4\right)E\\ 1024\beta \alpha ^21280\beta ^2+4096\alpha 32\beta ^5+384\beta ^3\alpha =0,\\ M=2,N=3.\end{array}$$
Its roots are not non-numerical anymore but three of them remain manifestly real at $`\alpha =\beta =0`$ where their form remains closed, $`(E_1,E_2,E_3)=(0,0,\sqrt[3]{192})`$. The consequent and precise specification of the whole domain of the reality of these roots remains as numerical a task as, say, its parallel studied in ref. .
### 3.3 More complicated multiplets at $`M=3`$ etc.
Explicit formulae with the integers $`M3`$ become appreciably more complicated. For all the really large truncations $`NM`$ they may remain useful. Their derivation would proceed along the same lines as before. One starts from the general pre-conditioning requirement
$$det\left(\begin{array}{ccccc}B_0& A_0& & & \\ C_1& B_1& A_1& & \\ D_2& C_2& \mathrm{}& \mathrm{}& \\ & \mathrm{}& \mathrm{}& B_{M2}& A_{M2}\\ & & D_{M1}& C_{M1}& B_{M1}\end{array}\right)=0$$
(11)
and its solutions have to be inserted in the, presumably, much larger main secular determinant
$$det\left(\begin{array}{ccccc}C_1& B_1& A_1& & \\ D_2& C_2& \mathrm{}& \mathrm{}& \\ & \mathrm{}& \mathrm{}& B_{N2}& A_{N2}\\ & & D_{N1}& C_{N1}& B_{N1}\\ & & & D_N& C_N\end{array}\right)=0.$$
(12)
The procedure only becomes inefficient at the larger integers $`M`$. In such a setting it would be more appropriate to treat both the “small” and “large” secular equations (11) and (12) on an equal footing, as a mutually coupled algebraic system.
In the $`MN`$ setting one suddenly loses the main advantage of our present construction, viz., its reducibility to the single secular equation. In practice, one also has to replace the insertions of $`E`$ or $`d`$ by the use of the so called Gröbner bases. The algorithm determines both $`d`$ and $`E`$ at once and is routinely provided by the languages like MAPLE .
Although the generalized $`M>3`$ formulae and calculations need not necessarily become hopelessly complicated (and, in fact, give the nice results, e.g., in the $`M\mathrm{}`$ limit ) their use would definitely require another motivation. Once we get that far, we would have no reason for maintaining our main assumption (4). At any real quantity $`M=2L+1`$ (reflecting a free variability of the coupling $`f`$) we would only have to replace the “small” dimension $`M`$ in eq. (11) by the overall truncation $`N`$ itself.
## 4 Concluding remarks
Results of our explicit constructions share their transparent algebraic character: Mathematically, one eliminates the auxiliary or “redundant” root of “the simple” condition (11) and ends up with the single “effective-Hamiltonian-like” square-matrix eigenvalue problem (12). This is the main merit (and general feature) of the QES systems of type I. Indeed, once we return to the explicit Hermitian decadic $`N=2`$ construction of sec. 2.4 in ref. , we discover that, in the same language, the main shortcoming of the type-II QES constructions lies precisely in the necessity of solving several coupled eigenvalue problems at once.
Our present sample non-numerical constructions available at the first few smallest integers $`M`$ parallel in fact many other quasi-exactly solvable models. In particular, the Sturmian constant-energy form of the elementary multiplets characterized already the very first quasi-exactly solvable (viz., Coulomb plus harmonic) model as discovered by A. Hautot in 1972 . More complicated relation between the couplings and energies characterizes, e.g., the partially solvable anharmonicities $`(1+gr^2)^1`$ revealed by G. Flessas in 1981 and explained algebraically in 1982 .
In our present paper the most elementary multiplets using auxiliary integer $`M=1`$ proved purely Sturmianic. Their spike-shaped short-range part of the interaction $`f`$ is not arbitrary and can only vanish in even dimensions. Vice versa, the emergence of a spike for our $`s`$wave multiplets in three dimensions is reminiscent of the so called conditionally solvable models where the choice of $`f=3/4`$ is obligatory .
At any auxiliary $`M`$, in comparison with all the Hermitian QES-I systems related to certain tridiagonal matrix representations of Lie algebras , all the decadic examples are distinguished by the four-diagonal and $`N`$dimensional secular determinants (12). In the Hermitian setting this “four-diagonality” was in fact the main reason of the restricted type-II solvability. We have shown that only the appropriate complexification can move the decadic systems to a higher, type-I QES group. At present, this group which exhibits the appealing $`𝒫𝒯`$ symmetry already encompasses the asymptotically decreasing quartic forces of Bender et al and the Coulomb + harmonic model .
## Acknowledgement
Partially supported by the GA AS CR grant Nr. A 104 8004.
## Figure captions
### Figure 1. Permitted domain for sextic oscillators
### Figure 2. Permitted domain for decadic oscillators
|
warning/0002/hep-ex0002033.html
|
ar5iv
|
text
|
# 1 ATLAS Calorimetry
## 1 ATLAS Calorimetry
A view of the ATLAS calorimeters is presented in Figure 1. The calorimetry consists of an electromagnetic (EM) calorimeter covering the pseudorapidity region $`|\eta |<3.2`$, a hadronic barrel calorimeter covering $`|\eta |<1.7`$, hadronic end-cap calorimeters covering $`1.5<|\eta |<3.2`$, and forward calorimeters covering $`3.1<|\eta |<4.9`$.
The EM calorimeter is a lead/liquid-argon (LAr) detector with accordion geometry . Over the pseudorapidity range $`|\eta |<1.8`$, it is preceded by a presampler detector, installed immediately behind the cryostat cold wall, and used to correct for the energy lost in the material upstream of the calorimeter.
The hadronic calorimetry of ATLAS, presented in Figure 1, consists of three main devices. In the barrel region ($`|\eta |<1.7`$) there is the scintillating Tile Calorimeter . The Hadronic End-cap LAr Calorimeter (HEC) extends up to $`|\eta |=3.2`$. The range $`3.1<|\eta |<4.9`$ is covered by the high density Forward Calorimeter (FCAL). Up to $`|\eta |=2.5`$ the basic granularity of the hadron calorimeters is $`\mathrm{\Delta }\eta \times \mathrm{\Delta }\varphi =0.1\times 0.1`$. This region is used for precise measurements of the energy and angles of jets. In the region $`|\eta |>2.5`$, the basic granularity is approximately $`\mathrm{\Delta }\eta \times \mathrm{\Delta }\varphi =0.2\times 0.2`$.
A more detailed description of all ATLAS calorimeters is given in the Calorimeter TDRs (, and ).
The performance of the barrel and extended barrel sections of the ATLAS hadronic calorimeter for the measurement of charged pion energy is studied. The intrinsic energy resolution, the effects of dead material, electronic noise and limited cone size are discussed.
## 2 Energy Resolution
In the barrel region, the response of the calorimeter was studied at two pseudorapidity values: $`\eta =0.3`$ (central barrel) and $`\eta =1.3`$ (extended barrel). First, the energy sampled in the different calorimeter compartments is converted to the total deposited energy using the electromagnetic energy scale (EM scale). The intrinsic performance of the calorimeter is studied: the energy considered is not restricted to a cone and electronic noise is not added. These effects are discussed later in Section 4. The algorithm to reconstruct the pion energy is similar to the ”Benchmark Method” used to analyse the combined LAr-Tile test beam data
$$E_{rec}=\alpha E_{had}+\beta E_{em}+\gamma E_{em}^2+\delta \sqrt{E_{had1}E_{em3}}+\kappa E_{ITC}+\lambda E_{scint}.$$
(1)
The coefficients $`\alpha `$ and $`\beta `$ take into account the different response of the EM and Hadronic Calorimeters to the pion energy. The quadratic term $`\gamma E_{em}^2`$ provides an additional first order correction for non-compensation (the coefficient is negative, it suppresses the signal for events with a large fraction of electromagnetic energy). The term $`\delta \sqrt{E_{had1}E_{em3}}`$ estimates the energy loss in the cryostat wall separating the LAr and Tile Calorimeters. In the central barrel, the energy is taken from the geometric mean of the energies in the last compartment of the LAr EM barrel ($`E_{em3}`$) and the first compartment of the Tile barrel calorimeter ($`E_{had1})`$; whereas in the extended barrel the energy is taken from the geometric mean of the energies in the outer wheel of the EM end-cap and the first compartment of the Tile extended barrel calorimeter. The term $`\kappa E_{ITC}`$ corrects for the energy loss in the dead material in the vertical gap between the Tile central and extended barrels. It is sampled by the two Intermediate Tile Calorimeter (ITC) modules (see Figure 2). The last term $`\lambda E_{scint}`$ corrects for the energy loss in the barrel and end-cap vertical cryostat walls (see Figure 2).
The response and the energy resolution for pions in the energy range from $`E_0=20GeV`$ to $`1TeV`$ at $`\eta =0.3`$ and 1.3 are shown in Figures 3 and 4. The open crosses show the results when the coefficients of Equation (1) are independent of energy. With the simple ”Benchmark Method”, the effect of non-compensation is not fully corrected for and there is a residual non-linearity of the pion response of the order of $`45\%`$ between 20 GeV and 1 TeV. The test beam data show a 10% residual non-linearity between 20 and 300 GeV when using the same reconstruction method , reflecting the fact that G-CALOR predicts a lower degree of non-compensation and may not describe correctly the energy dependence of the fraction of electromagnetic energy produced in the pion interaction. The energy dependence of the resolution is fitted with the two-term formula
$$\sigma /E=a/\sqrt{E}b$$
(2)
where the sampling term $`a`$ is given in $`\%\sqrt{GeV}`$ and the constant term $`b`$ in $`\%`$. Although the resolutions obtained for low-energy pions are similar in both cases, at high energy there is some longitudinal leakage in the central barrel, yielding a resolution at 1 TeV of 3% instead of the 2% obtained in the extended barrel. When energy dependent parameters are applied (solid dots), the linearity of the response is restored <sup>1</sup><sup>1</sup>1The 1% residual non-linearity at 20 GeV results from the fact that the coefficients were obtained by minimizing the expression $`(E_{rec}E_0)^2`$ without the addition of a linear term $`(E_{rec}E_0)`$ with a Lagrange multiplier. and the resolution improved. The results are presented in Table 1:
## 3 Pseudorapidity Scan
A pseudorapidity scan with pions of constant transverse energy $`E_T=20`$ and 50 GeV was carried out to check that the linearity of the response can be maintained across the pseudorapidity range covered by the barrel and the extended barrel, and that no significant tail appears in the line shape. The algorithm, characterised by Equation (1), with energy dependent parameters was applied. The parameters were adjusted independently for the six sets of pion data, each one covering an interval of 0.4 in pseudorapidity.
The energy resolutions obtained for the two scans are shown in Figure 5. The solid lines show the energy resolution corresponding to Equation $`\sigma /E=39\%/\sqrt{E}1\%`$ for $`E_T=20`$ GeV and $`\sigma /E=49\%/\sqrt{E}2\%`$ for $`E_T=50`$ GeV. This performance allows to fulfil the goal for the jet energy resolution of the ATLAS hadronic calorimetry in the region $`|\eta |<3`$ of $`\sigma /E=50\%/\sqrt{E}3\%`$.
In the region of the cracks between the calorimeters, from about $`|\eta |=1.3`$ to $`|\eta |=1.5`$, where the amount of dead material is the largest, the resolution is somewhat worse.
Figure 6 shows the linearity of the response across $`\eta `$. The fitted mean is plotted for each interval of 0.05 in $`\eta `$. The RMS of the mean is 1.1% for $`E_T=50`$ GeV and 2.0% for $`E_T=20`$ GeV.
In addition, the tails of the distributions of the reconstructed energy were investigated. Figure 7 shows the events with a pion response more than three standard deviations away from the mean. No significant tails are present: the fraction of events in the tails does not exceed 1 – 2%. A few events out of a total of 5000 events per energy scan, mostly from the sample of pions of $`E_T=20`$ GeV, deposit relatively little energy. These correspond to pions decaying to muons before reaching the calorimeter.
## 4 Effects of Electronic Noise and Cone Size
The results presented so far were obtained without any restriction on the pion reconstruction volume. These results characterise the intrinsic performance of the calorimeters. The presence of electronic noise does not allow integration over a too wide region, therefore the measurement of the pion energy must be restricted to a cone
$$\mathrm{\Delta }R=\sqrt{\mathrm{\Delta }^2\eta +\mathrm{\Delta }^2\varphi }.$$
(3)
A compromise has to be found between the pion energy lost outside of this cone and the noise included inside. The optimum varies as a function of pseudorapidity, since the showers have a width which is characterised by the polar angle whereas the calorimeter cells subtend intervals of constant pseudorapidity. Hence, at higher values of pseudorapidity, the showers extend laterally over more cells.
For a cone of $`\mathrm{\Delta }R=0.6`$ ($`\mathrm{\Delta }R=0.3`$), the noise is above 3 GeV (1.5 GeV). Digital filtering allows noise suppression (approximately by a factor 1.6). But even this level of noise is large and is comparable to the intrinsic resolution of the calorimeters for pions with energy of a few tens of GeV. A smaller cone of $`\mathrm{\Delta }R=0.3`$ is preferable from this point of view; after digital filtering, noise can be kept around 1 GeV in the barrel region and below 3 GeV in the pseudorapidity region covered by the extended barrel.
The response and the energy resolution in the barrel region are presented in Figures 8 and 9 as a function of the cone size used for the pion energy reconstruction. Energy losses outside a cone noticeably increase with decreasing cone size, especially for 50 GeV pions. The energy resolution also becomes worse, but it is still acceptable for the cone of $`\mathrm{\Delta }R=0.3`$.
Selecting cells with energy deposition above a certain threshold decreases the noise contribution. A study to optimise the cone size and the noise cut was performed in the barrel region. A $`2\sigma `$-noise applied to the calorimeter cells within a cone of $`\mathrm{\Delta }R=0.3`$ leads to the best energy resolution. In Figure 10, the energy dependency of the resolution is plotted for two pseudorapidities: $`\eta =0.3`$ and $`\eta =1.3`$. The energy dependence of the resolution can be parametrised by the equation 2 with an additional noise term:
$$\sigma /E=a/\sqrt{E}bc/E$$
(4)
where $`c`$ is given in $`\%GeV`$. The results of the fit with the formula of Equation 4 are presented in Table 2.
## 5 Conclusions
The response of the barrel and extended barrel region of the ATLAS calorimeter system to single charge pions was investigated using full simulation. Pion energy scans from $`E=20`$ GeV to $`1000`$ GeV and pseudo rapidity scans with pions of constant transverse energy ($`E_T=20`$ and $`50`$ GeV) have been analysed. For the pion energy reconstruction, the ”Benchmark approach” was used: it provides a first order correction for non-compensation effects and accounts for the effect of the dead material by using the ITC’s and scintillators to sample the energy loss or interpolating between the energy deposited in adjacent calorimeter layers.
Energy and rapidity dependent and independent calibrations have been considered. The best results are obtained with energy and rapidity dependent parameters. The effect of electronic noise has been studied: cone size and cell energy cuts have been optimised. The energy dependence of the resolution can be parameterized as: $`(50\pm 4)\%/\sqrt{E}(3.4\pm 0.3)\%1.0/E`$ at $`\eta =0.3`$ and $`(68\pm 8)\%/\sqrt{E}(3.0\pm 0.7)\%1.5/E`$ at $`\eta =1.3`$. The larger constant term at $`\eta =0.3`$ can be explained by the longitudinal leakage from calorimeters in this region. The resolution, obtained for the pseudorapidity scans, is represented by: $`(39\pm 1)\%/\sqrt{E}(1\pm 5)\%`$ for $`E_T=20`$ GeV, $`(49\pm 9)\%/\sqrt{E}(2\pm 6)\%`$ for $`E_T=50`$ GeV, in the full range, except from about $`|\eta |=1.3`$ to $`|\eta |=1.5`$, where the resolution is deteriorated by the energy loss in the dead material although no significant tails in the energy spectrum appears.
## Acknowledgements
The authors are grateful to Andrei Kiryunin for fruitful discussions.
|
warning/0002/cond-mat0002122.html
|
ar5iv
|
text
|
# Spin-fermion model near the quantum critical point: one-loop renormalization group results
\[
## Abstract
We consider spin and electronic properties of itinerant electron systems, described by the spin-fermion model, near the antiferromagnetic critical point. We expand in the inverse number of hot spots in the Brillouin zone, $`N`$ and present the results beyond previously studied $`N=\mathrm{}`$ limit. We found two new effects: (i) Fermi surface becomes nested at hot spots, and (ii) vertex corrections give rise to anomalous spin dynamics and change the dynamical critical exponent from $`z=2`$ to $`z>2`$. To first order in $`1/N`$ we found $`z=2N/(N2)`$ which for a physical $`N=8`$ yields $`z2.67`$.
\]
The problem of fermions interacting with critical antiferromagnetic spin fluctuations attracts a lot of attention at the moment due to its relevance to both high temperature superconductors and heavy-fermion materials . The key interest of the current studies is to understand the system behavior near the quantum critical point (QCP) where the magnetic correlation length diverges at $`T=0`$ . Although in reality the QCP is almost always masked by either superconductivity or precursor effects to superconductivity, the vicinity of the QCP can be reached by varying external parameter such as pressure in heavy fermion compounds, or doping concentration in cuprates.
In this paper, we study the properties of the QCP without taking pairing fluctuations into account. We assume that the singularities associated with the closeness to the QCP extent up to energies which exceed typical energies associated with the pairing. This assumption is consistent with the recent calculations of the pairing instability temperature in cuprates . From this perspective, the understanding of the properties of the QCP without pairing correlations is a necessary preliminary step for subsequent studies of the pairing problem.
A detailed study of the antiferromagnetic QCP was performed by Hertz and later by Millis who chiefly focused on finite $`T`$ properties near the QCP. They both argued that if the Fermi surface contains hot spots (points separated by antiferromagnetic momentum $`Q`$, see Fig. 1), then spin excitations possess purely relaxational dynamics with $`z=2`$. They further argued that in $`d=2`$, $`d+z=4`$, i.e., the critical theory is at marginal dimension, in which case one should expect that spin-spin interaction yields at maximum logarithmical corrections to the relaxational dynamics. Millis argued that this is true provided that the effective Ginsburg-Landau functional for spins (obtained by integrating out the fermions) is an analytic function of the spin ordering field. This is a’priori unclear as the expansion coefficients in the Ginsburg-Landau functional are made out of particle-hole bubbles and generally are sensitive to the closeness to quantum criticality due to feedback effect from near critical spin fluctuations on the electronic subsystem. Millis however demonstrated that the quartic term in the Ginsburg-Landau functional is governed by high energy fermions and is free from singularities.
In this communication, we, however, argue that the regular Ginsburg-Landau expansion is not possible in 2D by the reasons different from those displayed in . Specifically, we argue that the damping term in the spin propagator (assumed to be linear in $`\omega `$ in ) is by itself made out of a particle hole bubble, and, contrary to $`\varphi ^4`$ coefficient, is governed by low-energy fermions. We demonstrate that due to singular vertex corrections, the frequency dependence of the spin damping term at the QCP is actually $`\omega ^{1\alpha }`$. In the one loop approximation, we find $`\alpha 0.25`$.
Another issue which we study is the form of the renormalized quasiparticle Fermi surface near the magnetic instability. In a mean-field SDW theory, the Fermi surface in a paramagnetic phase is not affected by the closeness to the QCP. Below the instability, the doubling of the unit cell induces a shadow Fermi surface at $`k_F+Q`$, with the residue proportional to the deviation from criticality. This gives rise to the opening of the SDW gap near hot spots and eventually (for a perfect antiferromagnetic long range order) yields a Fermi surface in the form of small pockets around $`(\pi /2,\pi /2)`$ and symmetry related points (see Fig. 1a).
Several groups argued that this mean-field scenario is modified by fluctuations, and the Fermi surface evolution towards hole pockets begins already within the paramagnetic phase. We show that the Fermi surface near hot spots does evolve as $`\xi \mathrm{}`$, but due to strong fermionic damping (not considered in ), this evolution is a minor effect which at $`\xi =\mathrm{}`$ only gives rise to a nesting at the hot spots (see Fig. 1b).
The point of departure for our analysis is the spin-fermion model which describes low-energy fermions interacting with their own collective spin degrees of freedom. The model is described by
$``$ $`=`$ $`{\displaystyle \underset{𝐤,\alpha }{}}𝐯_F(𝐤𝐤_F)c_{𝐤,\alpha }^{}c_{𝐤,\alpha }+{\displaystyle \underset{q}{}}\chi _0^1(𝐪)𝐒_𝐪𝐒_𝐪+`$ (2)
$`g{\displaystyle \underset{𝐪,𝐤,\alpha ,\beta }{}}c_{𝐤+𝐪,\alpha }^{}\sigma _{\alpha ,\beta }c_{𝐤,\beta }𝐒_𝐪.`$
Here $`c_{𝐤,\alpha }^{}`$ is the fermionic creation operator for an electron with momentum $`𝐤`$ and spin projection $`\alpha `$, $`\sigma _i`$ are the Pauli matrices, and $`g`$ measures the strength of the interaction between fermions and their collective bosonic spin degrees of freedom. The latter are described by $`𝐒_𝐪`$ and are characterized by a bare spin susceptibility which is obtained by integrating out high-energy fermions.
This spin-fermion model can be viewed as the appropriate low-energy theory for Hubbard-type lattice fermion models provided that spin fluctuations are the only low-energy degrees of freedom. This model explains a number of measured features of cuprates both in the normal and the superconducting states . Its application to heavy-fermion materials is more problematic as in these compounds conduction electrons and spins are independent degrees of freedom, and the dynamics of spin fluctuations may be dominated by local Kondo physics rather than the interaction with fermions .
The form of the bare susceptibility $`\chi _0(q)`$ is an input for the low-energy theory. We assume that $`\chi _0(q)`$ is non-singular and peaked at $`𝐐`$, i.e., $`\chi _0(𝐪)=\chi _0/(\xi ^2+(𝐪𝐐)^2)`$, where $`\xi `$ is the magnetic correlation length. In principle, $`\chi _0`$ can also contain a nonuniversal frequency dependent term in the form $`(\omega /W)^2`$ where $`W`$ is of order of fermionic bandwidth. We, however, will see that for a Fermi surface with hot spots which we consider here, this term will be overshadowed by a universal $`\omega ^{1\alpha }`$ term produced by low-energy fermions.
The earlier studies of the spin-fermion model have demonstrated that the perturbative expansion for both fermionic and bosonic self-energies holds in power of $`\lambda =3g^2\chi _0/(4\pi v_F\xi ^1)`$ where $`v_F`$ is the Fermi velocity at a hot spot. This perturbation theory obviously does not converge when $`\xi \mathrm{}`$. As an alternative to a conventional perturbation theory, we suggested the expansion in inverse number of hot spots in the Brillouin zone $`N`$ ($`=8`$ in actual case) . Physically, large $`N`$ implies that a spin fluctuation has many channels to decay into a particle-hole pair, which gives rise to a strong ($`N`$) spin damping rate. At the same time, a fermion near a hot spot can only scatter into a single hot spot separated by $`𝐐`$. Power counting arguments than show that a large damping rate appears in the denominators of the fermionic self-energy and vertex corrections and makes them small to the extent of $`1/N`$. The only exception from this rule is the fermionic self-energy due to a single spin fluctuation exchange, which contains a frequency dependent piece without $`1/N`$ prefactor due to an infrared singularity which has to be properly regularized .
The set of coupled equations for fermionic and bosonic self-energies at $`N=\mathrm{}`$ has been solved in , and we merely quote the result. Near hot spots, we have
$`G_k^1(\omega )`$ $`=`$ $`\omega ϵ_k+\mathrm{\Sigma }(\omega ),`$ (3)
$`\chi (q,\mathrm{\Omega }_m)`$ $`=`$ $`\chi _0\xi ^2/(1+(𝐪𝐐)^2\xi ^2i\mathrm{\Pi }_\mathrm{\Omega }).`$ (4)
Here $`ϵ_k=v_x\stackrel{~}{k}_x+v_y\stackrel{~}{k}_y`$, where $`\stackrel{~}{k}=kk_{hs}`$, and $`v_x`$, $`v_y`$, which we set to be positive, are the components of the Fermi velocity at a hot spot ($`v_F^2=v_x^2+v_y^2`$). The fermionic self-energy $`\mathrm{\Sigma }_k(\omega )`$ and the spin polarization operator $`\mathrm{\Pi }_\mathrm{\Omega }`$ are given by
$$\mathrm{\Sigma }(\omega )=2\lambda \frac{\omega }{1+\sqrt{1\frac{i|\omega |}{\omega _{sf}}}};\mathrm{\Pi }_\mathrm{\Omega }=\frac{|\mathrm{\Omega }|}{\omega _{sf}}$$
(5)
and $`\omega _{sf}=(4\pi /N)v_xv_y/(g^2\chi _0\xi ^2)`$.
We see from Eq.(5) that for $`\omega \omega _{sf}`$, $`G(k_{hs},\omega )=Z/(\omega +i\omega |\omega |/(4\omega _{sf}))`$, i.e., as long as $`\xi `$ is finite, the system preserves the Fermi-liquid behavior at the lowest frequencies. The quasiparticle residue $`Z`$ however depends on the interaction strength, $`Z=(1+\lambda )^1`$, and progressively goes down when the spin-fermion coupling increases. At larger frequencies $`\omega \omega _{sf}`$, the system crosses over to a region, which is in the basin of attraction of the quantum critical point, $`\xi =\mathrm{}`$. In this region, $`G^1(k_F,\omega )3g(v_xv_y\chi _0/\pi Nv_F^2)^{1/2}(i|\omega |)^{1/2}\mathrm{sgn}(\omega )`$ . At the same time, spin propagator has a simple $`z=2`$ relaxational dynamics unperturbed by strong frequency dependence of the fermionic self-energy .
Our present goal is to go beyond $`N=\mathrm{}`$ limit and analyze the role of $`1/N`$ corrections. The $`1/N`$ terms give rise to two new features: vertex corrections which renormalize both fermionic and bosonic self-energies, and static fermionic self-energy $`\mathrm{\Sigma }_k`$. The corresponding diagrams are presented in Fig 2.
The lowest-order $`1/N`$ corrections have been calculated before . Both vertex correction and the static self-energy are logarithmical in $`\xi `$:
$`{\displaystyle \frac{\mathrm{\Delta }g}{g}}`$ $`=`$ $`{\displaystyle \frac{Q(v)}{N}}\mathrm{log}\xi ,`$ (6)
$`\mathrm{\Delta }ϵ_k`$ $`=`$ $`ϵ_{k+Q}{\displaystyle \frac{12}{\pi N}}{\displaystyle \frac{v_xv_y}{v_F^2}}\mathrm{log}\xi `$ (7)
where $`ϵ_{k+Q}=v_x\stackrel{~}{k}_x+v_y\stackrel{~}{k}_y`$, and $`Q(v)=(4/\pi )\mathrm{arctan}(v_x/v_y)`$ interpolates between $`Q=1`$ for $`v_x=v_y`$, and $`Q=2`$ for $`v_y0`$.
Besides, the $`1/N`$ corrections also contribute $`(1/N)\omega \mathrm{log}\xi `$ to $`G_k^1(\omega )`$, but this term is negligible compared to $`\mathrm{\Sigma }(\omega )`$ and we neglect it.
We see from (6,7) that the $`1/N`$ corrections to the vertex and to the velocity of the excitations are almost decoupled from each other: the velocity renormalization does not depend on the coupling strength at all, while the renormalization of the vertex depends on the ratio of velocities only through a non-singular $`Q(v)`$. This is a direct consequence of the fact that the dynamical part of the spin propagator is obtained self-consistently within the model. Indeed, the overall factors in $`\mathrm{\Delta }ϵ_k`$ and $`\mathrm{\Delta }g/g`$ are $`g^2(\omega _{sf}\xi ^2)`$ where $`\omega _{sf}\xi ^2`$ comes from the dynamical part of the spin susceptibility. Since the fermionic damping is produced by the same spin-fermion interaction as the fermionic self-energy, $`\omega _{sf}`$ scales as $`1/g^2`$, and the coupling constant disappears from the r.h.s. of (6,7).
The logarithmical dependence on $`\xi `$ implies that $`1/N`$ expansion breaks down near the QCP, and one has to sum up the series of the logarithmical corrections. We will do this in a standard one-loop approximation by summing up the series in $`(1/N)\mathrm{log}\xi `$ but neglecting regular $`1/N`$ corrections to each term in the series. We verified that in this approximation, the cancellation of the coupling constant holds even when $`g`$ is a running, scale dependent coupling. This in turn implies that one can separate the velocity renormalization from the renormalization of the vertex to all orders in $`1/N`$.
Separating the corrections to $`v_x`$ and $`v_y`$ and performing standard RG manipulations, we obtain a set of two RG equations for the running $`v_x^R`$ and $`v_y^R`$
$`{\displaystyle \frac{\text{d}v_x^R}{\text{d}L}}`$ $`=`$ $`{\displaystyle \frac{12}{\pi N}}{\displaystyle \frac{(v_x^R)^2v_y^R}{(v_x^R)^2+(v_y^R)^2}}`$ (8)
$`{\displaystyle \frac{\text{d}v_y^R}{\text{d}L}}`$ $`=`$ $`{\displaystyle \frac{12}{\pi N}}{\displaystyle \frac{(v_y^R)^2v_x^R}{(v_x^R)^2+(v_y^R)^2}}`$ (9)
where $`L=\mathrm{log}\xi `$. The solution of these equations is straightforward, and yields
$$v_x^R=v_xZ;v_y^R=v_yZ^1;Z=\left(1+\frac{24L}{\pi N}\frac{v_y}{v_x}\right)^{1/2}$$
(10)
where, we remind, $`v_x`$ and $`v_y`$ are the bare values of the velocities (the ones which appear in the Hamiltonian).
We see that $`v_y^R`$ vanishes logarithmically at $`\xi \mathrm{}`$. This implies that right at the QCP, the renormalized velocities at $`k_{hs}`$ and $`k_{hs}+Q`$ are antiparallel to each other, i.e. the Fermi surface becomes nested at hot spots (see Fig 1b). This nesting creates a “bottle neck effect” immediately below the criticality as the original and the shadow Fermi surfaces approach hot spots with equal derivatives (see Fig. 1b). This obviously helps developing a SDW gap at $`k_{hs}`$ below the magnetic instability. However, above the transition, no SDW precursors appear at $`T=0`$.
Another feature of the RG equations (9) is that they leave the product $`v_xv_y`$ unchanged. This is a combination in which velocities appear in $`\omega _{sf}`$. The fact that $`v_xv_y`$ is not renormalized implies that, without vertex renormalization, $`\omega _{sf}\xi ^2`$ remains finite at $`\xi =\mathrm{}`$, i.e., spin fluctuations preserve a simple $`z=2`$ relaxational dynamics.
We now consider vertex renormalization. Using again the fact that $`g^2\omega _{sf}`$ does not depend on the running coupling constant, one can straightforwardly extent the second-order result for the vertex renormalization, Eqn (6), to the one-loop RG equation
$$\frac{\text{d}g^R}{\text{d}L}=\frac{Q(v)}{N}g^R$$
(11)
where $`g^R`$ is a running coupling constant, and $`Q(v)`$ is the same as in (6) but contain renormalized velocities $`v_x^R`$ and $`v_y^R`$. At the QCP, the dependence on $`\xi `$ obviously transforms into the dependence on frequency ($`L=\mathrm{log}\xi (1/2)\mathrm{log}|\omega _0/\omega |`$, where $`\omega _0`$ is the upper cutoff). Using the fact that for $`\xi \mathrm{}`$, $`v_y^R/v_x^RN\pi /24L`$ and expanding $`Q(v)`$ near $`v_y^R=0`$, we find $`Q(v)2(1(2/\pi )v_y^R/v_x^R)=2N/3L`$. Substituting this result into (11) and solving the differential equation we obtain ($`\overline{\omega }=\omega /\omega _0`$)
$$g^R=g|\overline{\omega }|^{1/N}|\mathrm{log}\overline{\omega }|^{1/6}$$
(12)
We see that at the QCP, running coupling constant diverges as $`\omega 0`$ roughly as $`|\omega |^{1/N}`$. Substituting this result into the spin polarization operator and using the fact that $`\omega _{sf}(g^R)^2`$ we find that at the QCP,
$$\mathrm{\Pi }_\mathrm{\Omega }|\omega |^{\frac{N2}{N}}|\mathrm{log}\omega |^{\frac{1}{3}}$$
(13)
This result implies that vertex corrections change the dynamical exponent $`z`$ from its mean-field value $`z=2`$ to $`z=2N/(N2)`$. For $`N=8`$, this yields $`z2.67`$ and $`\chi (Q,\omega )|\omega |^{1\alpha }`$ where $`\alpha =0.75`$.
Singular vertex corrections also renormalize the fermionic self-energy as $`\mathrm{\Sigma }(\omega )g^R\sqrt{|\omega |}/v_F`$. Using the results for $`g^R`$ and $`v_Fv_x`$ we obtain at criticality
$$\mathrm{\Sigma }(\omega )|\omega |^{\frac{N2}{2N}}|\mathrm{log}\omega |^{\frac{2}{3}}$$
(14)
Eqs. (10), (13) and (14) are the central results of the paper. We see that the singular corrections to the Fermi velocity cause nesting but do not affect the spin dynamics. The corrections to the vertex on the other hand do not affect velocities, but change the dynamical critical exponent for spin fluctuations.
We now briefly discuss the form of the susceptibility at finite $`T`$. Previous studies have demonstrated that the scattering of a given spin fluctuation by classical, thermal spin fluctuations yields, up to logarithmical prefactors, $`\xi ^2uT`$, where $`u`$ is the coefficient in the $`\varphi ^4`$ term in the Ginsburg-Landau potential. This implies that at the QCP, $`\chi (Q,\omega )Ti|\omega |`$.
We, however, argue that the linear in $`T`$ and the linear in $`\omega `$ terms have completely different origin: the linear in $`\omega `$ term comes from low-energies and is universal, while the linear in $`T`$ term comes from high energies and is model dependent. This can be understood by analyzing the particle-hole bubble at finite $`T`$. We found that as long as one restricts with the linear expansion near the Fermi surface, $`\mathrm{\Pi }_\mathrm{\Omega }`$ preserves exactly the same form as at $`T=0`$, to all orders in the perturbation theory. The temperature dependence of $`\mathrm{\Pi }`$ appears only due to a nonzero curvature of the electronic dispersion and is obviously sensitive to the details of the dispersion at energies comparable to the bandwidth. Similarly, the derivation of the Landau-Ginsburg potential from (2) shows that $`u`$ vanishes for linearized $`ϵ_k`$, and is finite only due to a nonzero curvature of the fermionic dispersion.
The different origins of $`T`$ and $`\omega `$ dependences in $`\chi (Q,\omega )`$ imply that the anomalous $`\omega ^{1\alpha }`$ frequency dependence of $`\chi (Q,\mathrm{\Omega })`$ is not accompanied by the anomalous temperature dependence of $`\chi (Q,0)`$ simply because for high energy fermions, vertex corrections are non-singular. This result implies, in particular, that our theory does not explain anomalous spin dynamics observed in heavy fermion despite the similarity in the exponent for the frequency dependence of $`\mathrm{\Pi }_\mathrm{\Omega }`$, because the experimental data imply the existence of the $`\mathrm{\Omega }/T`$ scaling in $`CeCu_{6x}Au_x`$ . More likely, the explanation should involve the local Kondo physics .
Finally, we consider how anomalous vertex corrections affect the superconducting problem. We and Finkel’stein argued recently that at $`\xi =\mathrm{}`$, the kernel $`K(\omega ,\mathrm{\Omega })`$ of the Eliashberg-type gap equation for the $`d`$wave anomalous vertex $`F(\mathrm{\Omega })=(\pi T/2)_\omega K(\omega ,\mathrm{\Omega })F(\omega )`$ behaves as $`K(\omega ,\mathrm{\Omega })g^2/(v_F^2\mathrm{\Sigma }^2(\omega )\mathrm{\Pi }_{\mathrm{\Omega }\omega })^{1/2}`$ At $`N=\mathrm{}`$, this yields (including prefactor) $`K(\omega ,\mathrm{\Omega })=|\omega (\mathrm{\Omega }\omega )|^{1/2}`$. Although this kernel is qualitatively different from the one in the BCS theory because it depends on both frequencies, it still scales as inverse frequency due to an interplay between a non-Fermi liquid form of the fermionic self-energy and the absence of the gap in the spin susceptibility which mediates pairing. We demonstrated in that this inverse frequency dependence gives rise to a finite pairing instability temperature even when $`\xi =\mathrm{}`$.
To check how the kernel is affected by vertex corrections, we substitute the results for $`g^R`$, $`v_F`$, $`\mathrm{\Sigma }(\omega )`$ and $`\mathrm{\Pi }_\mathrm{\Omega }`$ into $`K(\omega ,\mathrm{\Omega })`$. We find after simple manipulations that despite singular vertex corrections, the kernel in the gap equation still scales inversely proportional to frequency. A simple extension of the analysis in then shows that the system still possesses a pairing instability at $`\xi =\mathrm{}`$ at a temperature which differs from that without vertex renormalization only by $`1/N`$ corrections.
To summarize, in this paper we considered the properties of the antiferromagnetic quantum critical point for itinerant electrons by expanding in the inverse number of hot spots in the Brillouin zone $`N=8`$. We went beyond a self-consistent $`N=\mathrm{}`$ theory and found two new effects: (i) Fermi surface becomes nested at hot spots which is a weak SDW precursor effect, and (ii) vertex corrections account for anomalous spin dynamics and change the dynamical critical exponent from $`z=2`$ to $`z>2`$. To first order in $`1/N`$ we found $`z=2N/(N2)2.67`$. We argued that anomalous frequency dependence is not accompanied by anomalous $`T`$ dependence.
It is our pleasure to thank G. Blumberg, P. Coleman, M. Grilli, A. Finkel’stein, D. Khveshchenko, A. Millis, H. von Löhneysen, J. Schmalian, Q. Si, and A. Tsvelik for useful conversations. The research was supported by NSF DMR-9979749.
|
warning/0002/quant-ph0002068.html
|
ar5iv
|
text
|
# Conditions for manipulation of a set of entangled pure states
## Abstract
We derive a sufficient condition for a set of pure states, each entangled in two remote $`N`$-dimensional systems, to be transformable to $`k`$-dimensional-subspace equivalent entangled states ($`kN`$) by same local operations and classical communication. If $`k=N`$, the condition is also necessary. This condition reveals the function of the relative marginal density operators of the entangled states in the entanglement manipulation without sufficient information of the initial states.
PACS numbers: 03.67-a, 03.65.Bz, 89.70.+c
The deep ways that quantum information differs from classical information involve the properties, implications, and uses of quantum entanglement $`[1]`$. As a useful physical resource of quantum information, entanglement plays a key role for quantum computation $`[2]`$, quantum teleportation $`[3]`$, quantum superdense coding $`[4]`$ and certain types of quantum cryptography $`[5]`$, etc. To accomplish these tasks, transformation between the input entanglement we possess and the target entanglement we require is necessary. Attempts $`\left[618\right]`$ have been made to uncover the fundamental laws of the transformations under local quantum operations and classical communication (LQCC), that is, the different entangled parties may do whatever they wish to in their local system, and may communicate classically, but they cannot use quantum communication.
All previous entanglement manipulation protocols only consider a definite entangled state shared by distant observers. However, quantum information processing often has to work with insufficiently known initial states. It is therefore important to understand which processes work without full knowledge of initial states. In this letter, we address the question whether it is possible to manipulate a set of entangled pure states only by one LQCC protocol, just like quantum clone $`[1922]`$. This problem is fundamentally and also practically important. An example may be in the disentangled process of quantum clone $`\left[2324\right]`$. In deterministic state-dependent cloning process $`\left[21\right]`$, although according to Nielsen Theorem $`\left[11\right]`$ each of the two final states can be transformed to the disentangled state by LQCC respectively, it is impossible to separate the output by LQCC without knowing which one the initial state is $`\left[23\right]`$. The same result also exists in probabilistic telecloning process $`\left[24\right]`$. In Ref. $`\left[18\right]`$, we showed that a local operation can enhance the entanglement of a set of two-level entangled states simultaneously. In this letter, we investigate the problem with some restrictions of the final states. The investigations here are for the finite (nonasymptotic) case, from which asymptotic results may be recovered by taking limits. The transformation process may be probabilistic, but not approximate. To present our questions and results, we first collect some useful Facts:
1. An arbitrary bipartite entangled pure state $`|\mathrm{\Omega }`$ that Alice and Bob share can be written as $`\left[25\right]`$ $`|\mathrm{\Omega }=\left(U_AU_B\right)\underset{i=1}{\overset{N}{}}\sqrt{\lambda _i}|i_A|i_B`$, where $`U_A`$and $`U_B`$ are local unitary transformations by Alice and Bob respectively, $`\underset{i=1}{\overset{N}{}}\lambda _i=1`$, and $`\left\{|i_A\right\}`$ ($`\left\{|i_B\right\}`$) form an orthogonal basis for system $`A`$ ($`B`$). In this letter, we take $`N`$ as the maximum dimensions of the subsystem. If $`|\mathrm{\Omega }`$ has $`m`$ no-zero eigenvalues, we call $`|\mathrm{\Omega }`$ as $`m`$-dimensional entangled state. The $`m`$-dimensional maximally entangled state can be generally written as $`|\mathrm{\Phi }=\left(U_AU_B\right)\frac{1}{\sqrt{m}}\underset{i=1}{\overset{m}{}}|i_A|i_B`$. The marginal density operator for Alice’s (Bob’s) subsystem is defined as $`\rho _{A\left(B\right)}\left(|\mathrm{\Omega }\right)=Tr_{B(A)}|\mathrm{\Omega }\mathrm{\Omega }|`$. Obviously $`\lambda _i`$ is the eigenvalue of $`\rho _{A\left(B\right)}\left(|\mathrm{\Omega }\right)`$. Furthermore, we denote $`|\alpha |\beta `$ if $`|\alpha `$ and $`|\beta `$ are the same up to local unitary operations by Alice and Bob. The Schmidt decomposition implies that $`|\alpha |\beta `$ if and only if $`\rho _A\left(|\alpha \right)`$ and $`\rho _A\left(|\beta \right)`$ have the same spectrum of eigenvalues.
2. Denote $`Q`$ as an index set. We call that a set of entangled states $`\left\{|\alpha _{\mathrm{}},\mathrm{}Q\right\}`$ are $`k`$-dimensional-subspace equivalent if and only if there exist no-zero constant $`C_\alpha _{\mathrm{}}`$ and no-zero Schmidt coefficients $`\mu _t\left(|\alpha _{\mathrm{}}\right)`$ ($`1tk`$) making $`\mu _t\left(|\alpha _\mathrm{}_0\right)=C_\alpha _{\mathrm{}}\mu _t\left(|\alpha _{\mathrm{}}\right)`$. Suppose $`|\alpha `$ and $`|\beta `$ are $`N`$ and $`N^{^{}}`$ ($`N^{^{}}N`$) dimensional entangled states respectively, we denote $`\widehat{F}_{A\left(B\right)}\left(|\alpha |\beta \right)=\rho _{A\left(B\right)}^{\frac{1}{2}}\left(|\alpha \right)\rho _{A\left(B\right)}\left(|\beta \right)\rho _{A\left(B\right)}^{\frac{1}{2}}\left(|\alpha \right)`$ as the relative marginal density operator of states $`|\alpha `$ and $`|\beta `$ for Alice’s (Bob’s) subsystem. $`\widehat{F}_{A(B)}`$ describes the relation of the marginal density operators of the two states. For a set of relative marginal density operators $`\left\{\widehat{F}_{A\left(B\right)}^{\mathrm{}},\mathrm{}Q\right\}`$, we denote that $`\widehat{F}_{A\left(B\right)}^{\mathrm{}}`$ are similar about $`I_k`$ if and only if $`\widehat{F}_{A\left(B\right)}^{\mathrm{}}`$ can be represented on the orthogonal basis $`|i_A`$ as
$$\widehat{F}_{A\left(B\right)}^{\mathrm{}}=Vdiag(s_{\mathrm{}}I_k,D_{\mathrm{}})V,$$
(1)
where $`s_{\mathrm{}}>0`$, $`I_k`$ is the $`k\times k`$ unit matrix, $`V`$ is unitary and $`D_{\mathrm{}}`$ is a symmetric matrix.
3. Any operation $`P`$ Alice performs on the maximally entangled state $`\frac{1}{\sqrt{N}}\underset{k=1}{\overset{N}{}}|k_A|k_B`$ is equal to the transposed operation $`P^+`$ performed by Bob $`\left[26\right]`$, that is, $`\left(PI\right)\underset{k=1}{\overset{N}{}}|k_A|k_B=\left(IP^+\right)\underset{i=1}{\overset{N}{}}|k_A|k_B`$.
4. Given any pure bipartite state $`|\mathrm{\Psi }_{AB}=\underset{i=1}{\overset{N}{}}\sqrt{\lambda _i}|i_A|i_B`$ shared by Alice and Bob and any complete set of projection operators $`\left\{P_l^{Bob}\right\}`$’s by Bob, there exists a complete set of projection operators $`\left\{P_l^{Alice}\right\}`$’s by Alice and, for each outcome $`l`$, a direct product of local unitary transformations $`U_l^AU_l^B`$ such that, for each $`l`$ $`\left[8\right]`$
$$\left(IP_l^{Bob}\right)|\mathrm{\Psi }_{AB}=\left(U_l^AU_l^B\right)\left(P_l^{Alice}I\right)|\mathrm{\Psi }_{AB}.$$
(2)
5. The most general scheme of entanglement manipulation of a bipartite entangled pure state involves local operations of respective system and two-way communication between Alice and Bob $`\left[6\right]`$. The local operations can be represented as generalized measurements, described by operators $`A_k`$ and $`B_l`$ on each system, satisfying the condition $`_kA_k^+A_kI_N`$ ($`I_N_kA_k^+A_k`$ is positive semidefinite) and $`_lB_l^+B_lI_N`$, where $`I_N`$ is the unit operator of Alice’s or Bob’s subsystem. The LQCC protocol we consider maps the initial state $`|\varphi \varphi |`$ to the target state $`\left[9\right]`$,
$$|\phi \phi |=\frac{_{kl}A_kB_l|\varphi \varphi |A_k^+B_l^+}{Tr\left(_{kl}A_kB_l|\varphi \varphi |A_k^+B_l^+\right)}.$$
(3)
The initial and final states are pure, it follows that
$$A_kB_l|\varphi =\sqrt{p_{kl}}|\phi ,$$
(4)
with non-negative success probability $`p_{kl}`$ satisfying $`p_{kl}=Tr\left(A_kB_l|\varphi \varphi |A_k^+B_l^+\right)`$.
Suppose Alice and Bob share a pure bipartite $`N`$-dimensional entangled state $`|\varphi _1`$ that they can convert to another entangled pure state $`|\phi _1`$ by a LQCC process with no-zero probability $`\left[12\right]`$. Denote $`S`$ as an index set, our question is what property characterizes the set of entangled pure state $`\left\{|\varphi _1,|\varphi _\nu ,\nu S\right\}`$ that can be transformed to the final states $`\left\{|\phi _1,|\phi _\nu ,\nu S\right\}`$ by the same LQCC process if $`|\phi _\nu `$ are $`k`$-dimensional-subspace equivalent to state $`|\phi _1`$ ($`kN`$). In this letter, we derive a sufficient condition for such manipulation. If $`k=N`$, we show that the condition is also necessary.
Theorem 1: A set of bipartite entangled pure states $`\left\{|\phi _1,|\phi _\nu ,\nu S\right\}`$ can be probabilistic transformed to $`k`$-dimensional-subspace equivalent states by one LQCC protocol if the relative marginal density operators of states $`|\phi _1`$ and $`|\phi _\nu `$ are similar about $`I_k`$.
As a simple application of the result, suppose Alice and Bob each possess a four-dimensional quantum system, with respectively orthonormal bases denoted by $`|1`$, $`|2`$, $`|3`$ and $`|4`$. The initial entangled state may be one of the following states
$`|\alpha `$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{4}}}|11+\sqrt{{\displaystyle \frac{1}{4}}}|22+\sqrt{{\displaystyle \frac{1}{16}}}|33+\sqrt{{\displaystyle \frac{7}{16}}}|44,`$ ()
$`|\beta `$ $`=`$ $`\sqrt{{\displaystyle \frac{1}{4}}}|11+\sqrt{{\displaystyle \frac{1}{4}}}|22+\sqrt{{\displaystyle \frac{1}{2}}}|33.`$ (5)
Obviously the relative marginal density operator $`F_A\left(|\alpha |\beta \right)=diag(1,1,8,0)`$ has two same eigenvalues. Alice can transform above two states to $`2`$-dimensional maximally entangled state $`|\mathrm{{\rm Y}}=\sqrt{\frac{1}{2}}\left(|11+|22\right)`$ with local generalized measurement $`P_1=|11|+|22|`$ satisfying $`P_1^+P_1I_4`$.
Proof of Theorem 1:
Generally, the states to be transformed can be represented as $`|\varphi _1=\underset{i=1}{\overset{N}{}}\sqrt{\lambda _i}|i_A|i_B`$ and $`|\varphi _\nu =\left(U_A^\nu U_B^\nu \right)\underset{i=1}{\overset{N}{}}\sqrt{\mu _i^\nu }|i_A|i_B`$ with $`\lambda _i>0`$. Suppose $`|\varphi _1`$ is transformed to the state $`|\phi _1=\underset{i=1}{\overset{N}{}}\sqrt{\gamma _i}|i_A|i_B`$ by a LQCC process, the same LQCC should transform state $`|\varphi _\nu `$ to state $`|\phi _\nu `$ that has Schmidt coefficients $`\eta _i^\nu =c_\nu \gamma _i`$, $`i=1,2,\mathrm{},k`$, where $`c_\nu `$ is a no-zero real number.
Denote $`\lambda =diag(\lambda _1,\lambda _2,\mathrm{},\lambda _N)`$, and $`\mu ^\nu `$, $`\gamma `$, $`\eta ^\nu `$ are of similar definitions. Obviously $`\rho _A\left(|\varphi _1\right)=\lambda `$, $`\rho _A\left(|\varphi _\nu \right)=U_A^\nu \mu ^\nu \left(U_A^\nu \right)^+`$. Since the relative marginal density operators $`\left\{\widehat{F}_A^\nu \left(|\phi _1|\phi _\nu \right),\nu S\right\}`$ are similar about $`I_k`$, applying Fact 2, we obtain
$$\lambda ^{\frac{1}{2}}U_A^\nu \sqrt{\mu ^\nu }=V\left(\begin{array}{cc}\sqrt{s_\nu }I_k& 0\\ 0& \sqrt{D_\nu }\end{array}\right)G_\nu ,$$
(6)
where $`G_\nu `$ is a unitary matrix. Suppose $`P_l=\sqrt{\epsilon _l}\sqrt{\gamma }V^+\sqrt{\lambda ^1}`$. Since $`P_l^+P_l=\epsilon _l\sqrt{\lambda ^1}V\gamma V^+\sqrt{\lambda ^1}`$, suitable choice of $`\epsilon _l`$ can make $`P_l^+P_lI_N`$, which means $`P_l`$ is a generalized measurement (Fact 5) independent of the initial states $`|\varphi _\nu `$. According to Fact 3, $`P_l`$ acts on the states $`|\varphi _1`$ and $`|\varphi _\nu `$ as follows:
$`\left(P_lI\right)|\varphi _1`$ $`=`$ $`\sqrt{\epsilon _l}\left(IV\right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\gamma _i}|i_A|i_B`$ ()
$`=`$ $`\sqrt{\epsilon _l}\left(IV\right)|\phi _1,`$ (6)
$`\left(P_lI\right)|\varphi _\nu `$ ()
$`=`$ $`\sqrt{\epsilon _ls_\nu }\left(IU_B^\nu G_\nu ^+V\right)\left(IH_\nu \right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\gamma _i}|i_A|i_B`$ (7)
$`=`$ $`\sqrt{\epsilon _ls_\nu }\left(IU_B^\nu G_\nu ^+V\right)|\phi _2^{^{}},`$ (8)
where the corresponding matrix of $`H_\nu =diag(I_k,s_\nu ^1\sqrt{D_\nu })`$, $`|\phi _\nu ^{^{}}=\left(IH_\nu \right)`$ $`\underset{i=1}{\overset{N}{}}\sqrt{\gamma _i}|i_A|i_B`$. Denote the normalized states of $`|\phi _\nu ^{^{}}`$ as $`|\phi _\nu `$, they are $`k`$-dimensional-subspace equivalent to $`|\phi _1`$. Thus we finish the proof of Theorem 1.
If the final states are $`N`$-dimensional-subspace equivalent, the above sufficient condition can be expressed in a more simple form with clear physical meaning. In fact, since $`Tr\rho _A\left(\varphi _1\right)=Tr\rho _A\left(|\varphi _\nu \right)=1`$, the above sufficient condition means that the relative marginal density operators $`F_A^\nu \left(|\phi _1|\phi _\nu \right)=I`$, i.e., the marginal density operators $`\rho _A\left(\varphi _1\right)=\rho _A\left(|\varphi _\nu \right)`$ and $`|\varphi _\nu `$ are also $`N`$-dimensional entangled pure states. In this case, such condition is also necessary.
Theorem 2: A set of $`N`$-dimensional entangled pure states $`\left\{|\varphi _1,|\varphi _\nu ,\nu S\right\}`$ can be probabilistic transformed to $`N`$-dimensional-subspace equivalent states $`\left\{|\phi _1,|\phi _\nu ,\nu S\right\}`$ by same LQCC protocol if and only if they share same marginal density operators for Alice’s or Bob’s subsystem.
Proof of Theorem 2:
We need only prove the necessity. Consider that a generalized measurement $`A_kB_l`$ can transform the $`N`$-dimensional states $`|\varphi _1`$ and $`|\varphi _\nu `$ to $`N`$-dimensional-subspace equivalent states $`|\phi _1`$ and $`|\phi _\nu `$. Obviously the Schmidt coefficients of states $`|\phi _1`$ and $`|\phi _\nu `$ are greater than zero and $`|\phi _1`$ $`|\phi _\nu `$. We first prove the necessity in the condition that only one side generalized measurement is performed. The one-side generalized measurement acts on the initial states as follows:
$`\left(P_lI\right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\lambda _i}|i_A|i_B`$ ()
$`=`$ $`\sqrt{\varsigma }\left(E_1F_1\right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\kappa _i}|i_A|i_B,`$ (10)
$`\left(P_lU_A^\nu U_B^\nu \right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\mu _i^\nu }|i_A|i_B`$
$`=`$ $`\sqrt{\tau _\nu }\left(E_\nu F_\nu \right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\kappa _i}|i_A|i_B,`$ (11)
where $`E_1F_1`$ and $`E_\nu F_\nu `$ are local unitary operations, $`\varsigma `$ and $`\tau _\nu `$ are the probabilities of success, $`\kappa =diag(\kappa _1,\kappa _2,\mathrm{},\kappa _N)`$ is the eigenvalue matrix of the final states. Since the final states are $`N`$-dimensional-subspace equivalent, they must be $`N`$-dimensional entangled states (Fact 2) and $`\kappa _i>0`$ for $`i=1,2,\mathrm{},N`$. According to Fact 3, the above two equations can be represented with matrices as
$`P_l\sqrt{\lambda }`$ $`=`$ $`\sqrt{\varsigma }E_1\sqrt{\kappa }F_1^+,`$ ()
$`P_lU_A^\nu \sqrt{\mu ^\nu }\left(U_B^\nu \right)^+`$ $`=`$ $`\sqrt{\tau _\nu }E_\nu \sqrt{\kappa }F_\nu ^+.`$ (12)
Substituting $`P_l\sqrt{\lambda }`$ of the first equation into the second, we obtain
$$T_\nu ^+\kappa T_\nu =\kappa ,$$
(13)
where $`T_\nu =\sqrt{\frac{\varsigma }{\tau _\nu }}F_1^+\sqrt{\lambda ^1}U_A^\nu \sqrt{\mu ^\nu }\left(U_B^\nu \right)^+F_\nu `$. Since $`\kappa _i>0`$, Eq. (11) means that $`T_\nu `$ is unitary, it follows $`\frac{\varsigma }{\tau _\nu }U_A^\nu \mu ^\nu \left(U_A^\nu \right)^+=\lambda `$. Since $`_i\lambda _i=1`$, $`_i\mu _i^\nu =1`$, we get $`\tau _\nu =\varsigma `$, $`\mu ^\nu =\lambda `$, and
$$\rho _A\left(|\varphi _1\right)=\lambda =U_A^\nu \mu ^\nu \left(U_A^\nu \right)^+=\rho _A\left(|\varphi _\nu \right).$$
(13)
Now we consider that a two-side generalized measurement $`A_kB_l`$ transforms the input states $`|\varphi _1`$ and $`|\varphi _\nu `$ to $`N`$-dimensional-subspace equivalent states. With Fact 4, we get
$`A_kB_l|\varphi _1`$ ()
$`=`$ $`\left(A_kV_l^AB_lV_l^B\right)|\varphi _1,`$ (15)
$`\left(A_kB_l\right)|\varphi _\nu `$
$`=`$ $`\left(A_kU_A^\nu H_l^{\nu A}B_lU_B^\nu H_l^{\nu B}\right){\displaystyle \underset{i=1}{\overset{N}{}}}\sqrt{\mu _i^\nu }|i_A|i_B,`$ (16)
where $`V_l^A`$, $`V_l^B`$, $`H_l^{\nu A}`$ and $`H_l^{\nu B}`$ are local unitary operations. The above two equations means that one-side generalized measurement $`A_kI`$ can transform the initial states $`\left(V_l^AB_lI\right)\underset{i=1}{\overset{N}{}}\sqrt{\lambda _i}|i_A|i_B`$ and $`\left(U_A^\nu H_l^{\nu A}B_lU_B^\nu I\right)\underset{i=1}{\overset{N}{}}\sqrt{\mu _i^\nu }|i_A|i_B`$ to $`N`$-dimensional-subspace equivalent states. Therefore $`\left(B_lI\right)\underset{i=1}{\overset{N}{}}\sqrt{\lambda _i}|i_A|i_B`$ and $`\left(B_lU_B^\nu I\right)\underset{i=1}{\overset{N}{}}\sqrt{\mu _i^\nu }|i_A|i_B`$ must also be $`N`$-dimensional-subspace equivalent states, which means the marginal density operators for Bob’s side of the input states must satisfy $`\rho _B\left(|\varphi _1\right)=\rho _B\left(|\varphi _\nu \right)`$. So one of the two subsystems of the initial states must have same marginal density operators.
So far we have proven Theorem 1 and Theorem 2. In Theorem 1, we give a sufficient condition for that a set of entangled pure states can be probabilistic transformed to $`k`$-dimensional-subspace equivalent states by same LQCC protocol. We conjecture that this condition is also necessary. In fact, it is true if the generalized measurement is restricted in one side. In this case, the eigenvalue matrix of the final states in Eq. (10) is substituted by $`\kappa ^\nu `$. $`\kappa ^\nu `$ should have at least $`k_\nu `$ ($`k_\nu k`$) no-zero eigenvalues $`\kappa _i^\nu =d_\nu \kappa _i`$, $`1ik_\nu `$, where $`d_\nu `$ is a constant dependent on $`\nu `$. Eq. (11) should be rewritten as
$$T_\nu ^+\kappa T_\nu =\kappa ^^\nu ,$$
(17)
where $`T_\nu `$ is the same as that in Eq. (11). Eq. (14) means $`T_\nu `$ can be represented as
$$T_\nu =\left(\begin{array}{cc}\sqrt{d_\nu }M_{k_\nu }^\nu & 0\\ 0& R_\nu \end{array}\right),$$
(17)
where $`M_{k_\nu }^\nu `$ is a $`k_\nu \times k_\nu `$ unitary matrix and $`R_\nu `$ may be any possible matrix. The unitarity of $`M_{k_\nu }^\nu `$ yields
$`\widehat{F}_A\left(|\varphi _1|\varphi _\nu \right)`$ $`=`$ $`\lambda ^{\frac{1}{2}}U_A^\nu \mu ^\nu \left(U_A^\nu \right)^+\lambda ^{\frac{1}{2}}`$ ()
$`=`$ $`{\displaystyle \frac{d_\nu \tau _\nu }{\varsigma }}F_1\left(\begin{array}{cc}I_{k_\nu }& 0\\ 0& \frac{1}{d}R_\nu R_\nu ^+\end{array}\right)F_1^+.`$ (20)
Since $`k_\nu k`$ and $`F_1`$ is independent of index $`\nu `$, Eq. (16) means that the relative marginal density operators $`\widehat{F}_A\left(|\varphi _1|\varphi _\nu \right)`$ of the initial states $`|\varphi _1`$ and $`|\varphi _\nu `$ are similar about $`I_k`$.
Theorem 2 shows that the sufficient condition in Theorem 1 is also necessary in a special case. The result means that Alice (Bob) cannot probabilistically transform $`N`$-dimensional entangled states that are different in her (his) local observation to $`N`$-dimensional-subspace equivalent states. Generally the ordered Schmidt coefficients of the states to be transformed must be the same, but these states need not be the same, there exist unitary transformations on both Alice’s and Bob’s sides. While arbitrary on Bob’s (Alice’s) side, the unitary operators on Alice’s (Bob’s) side must preserve the density matrix $`\rho _A`$ ($`\rho _B`$), which means that only when there exist some coefficients satisfying $`\lambda _i=\lambda _{i+1}`$, the unitary operators $`U_A^\nu `$ ($`U_B^\nu `$) can be non-unit.
The above results can be directly applied to concentration of entanglement $`[6,8]`$, that is, transforming partial entanglement to maximally entanglement. Theorem 1 gives a sufficient condition for the concentration of a set of partial entanglement to the maximally entanglement (not necessary $`N`$-dimensional), while Theorem 2 shows that Alice (Bob) can probabilistic concentrate several different $`N`$-dimensional partial entangled states to $`N`$-dimensional maximally entangled states by same LQCC process if and only if the marginal density operators of these states are the same for her or his subsystem. In the proof of Theorem 2 we also showed the following important result:
Proposition 1: Different $`N`$-dimensional entangled states cannot be transformed to one $`N`$-dimensional entangled state by same LQCC protocol on individual pairs.
However, such result does not prohibit us from transforming different entangled states to one of lower dimension. An example is the states $`|\alpha `$ and $`|\beta `$ in Eqs. (5).
While one can always, with finite probability, bring an individual entangled pure state to a maximally entangled state using only LQCC $`\left[8\right]`$, Linden et al. $`[9]`$ have shown that it is impossible to purify a two-level mixed state to a maximally entangled state by any combination of LQCC acting on individual pairs. In this letter we generalize it to $`N`$-level mixed state as
Theorem 3 : It is impossible to purify a $`N`$-dimensional mixed state to a $`N`$-dimensional maximally entangled state by LQCC on individual pairs.
Proof of Theorem 3:
Consider a given mixed state $`\rho `$, generally we can use the spectral decomposition $`\left[25\right]`$ of the state $`\rho =_ip_i|\psi _i\psi _i|`$. Proposition 1 indicates that different decomposition terms $`|\psi _i`$ of the mixed state $`\rho `$ can never be transformed to one $`N`$-dimensional pure state by same LQCC, which means $`\rho `$ cannot be concentrated into a $`N`$-dimensional maximally entangled pure state by LQCC on individual pairs. This result is surprising because we expect entanglement to be a property of each pair individually rather than a global property of many pairs.
However, Theorem 3 does not mean that we cannot obtain lower dimensional entangled pure state from a mixed state by LQCC. For example, we can concentrate the mixed state $`\rho =\frac{1}{4}|\alpha \alpha \left|+\frac{3}{4}\right|\beta \beta |`$ to the maximally entangled pure state $`|\mathrm{{\rm Y}}=\sqrt{\frac{1}{2}}\left(|11+|22\right)`$, where $`|\alpha `$ and $`|\beta `$ are the states in Eqs. (5).
Another interesting application may be probabilistic quantum superdense coding. Suppose Bob has four choices to perform $`U_B`$ i.e. $`\{I,\sigma _x,i\sigma _y,\sigma _z\}`$ on the initial possessed partial entangled states, just like that in Ref. $`[4]`$. Alice still can transform the partial entangled state to the maximally entangled state with no-zero probability, although she does not know which $`U_B`$ Bob performs. Bob sends his particle to Alice after he has performed $`U_B`$. Alice’s task is then to identify the four Bell states and obtain the information.
The further application of these results need to be explored. Similar to quantum cloning process, although we lack sufficient information about the initial states, we still can make operations on them and extract information at the end. The indefinite initial entanglement may contain quantum information and our results may be useful in quantum cryptography and quantum communication.
In summary, we have shown that a set of entangled pure states $`\left\{|\phi _1,|\phi _\nu ,\nu S\right\}`$ can be probabilistic transformed to $`k`$-dimensional-subspace equivalent states by same LQCC protocol if the relative marginal density operators $`\widehat{F}_{A(B)}\left(|\phi _1|\phi _\nu \right)`$ are similar about $`I_k`$. In the case of that the final states are $`N`$-dimensional-subspace equivalent, the condition can be expressed as that the input states must share the same marginal density operators for Alice’s or Bob’s subsystem and it is both sufficient and necessary. As the application, we showed that it is impossible to purify a mixed state to a maximally entangled state of same dimension by LQCC on individual pairs and presented the probabilistic superdense coding.
Acknowledgment: This work was supported by the National Natural Science Foundation of China.
|
warning/0002/hep-lat0002023.html
|
ar5iv
|
text
|
# Domain-Wall Induced Quark Masses in Topologically-Nontrivial Background
## A EIGENVALUES OF FREE DOMAIN-WALL FERMION
The domain-wall induced fermion mass in the free case was first calculated by Shamir using Green’s function approach. Vranas stated in his paper that he obtained the same result by diagonalizing the domain-wall Dirac operator without showing the actual calculation. An explicit derivation of $`m_{\mathrm{eff}}`$ in the $`m_f=0`$ case was later provided by Neuberger . Here we show a complete derivation with the inclusion of $`m_f`$.
The domain-wall Dirac operator in the free field limit in momentum space can be written as:
$`D=i\overline{p}+M^+P_++M^{}P_{},`$ (A1)
where $`P_\pm =\frac{1}{2}(1\pm \gamma _5)`$ and $`\overline{p}=\gamma _\mu \mathrm{sin}p^\mu `$. Mass matrices $`M^\pm `$ are defined as
$`(M^+)_{ss^{}}=\delta _{s+1,s^{}}b(p)\delta _{s,s^{}}m_f\delta _{s,N_s}\delta _{s^{},1}`$ $`,`$ (A2)
$`(M^{})_{ss^{}}=\delta _{s1,s^{}}b(p)\delta _{s,s^{}}m_f\delta _{s,1}\delta _{s^{},N_s}`$ $`(s,s^{}=1,\mathrm{}N_s),`$ (A3)
where $`b(p)=1m_0+_\mu (1\mathrm{cos}p_\mu )`$ and $`m_f`$ is the explicit fermion mass. Our goal is to calculate the smallest eigenvalue of the bilinear hermitian domain-wall Dirac operator
$`DD^{}=\overline{p}^2+M^+M^{}P_++M^{}M^+P_{}.`$ (A4)
Since $`M^+M^{}`$ and $`M^{}M^+`$ have the same eigenvalue spectrum, it is sufficient to consider
$`M^+M^{}=\left(\begin{array}{ccccc}b^2+1\hfill & b\hfill & 0\hfill & \mathrm{}\hfill & m_fb\hfill \\ b\hfill & b^2+1\hfill & b\hfill & \mathrm{}\hfill & 0\hfill \\ \mathrm{}\hfill & & & \mathrm{}\hfill & \\ m_fb\hfill & \mathrm{}\hfill & 0\hfill & b\hfill & b^2+m_f^2\hfill \end{array}\right).`$ (A9)
The second to $`(N_s1)`$-th row of the secular equation $`(M^+M^{})_{ss^{}}\mathrm{\Psi }_s^{}=\lambda ^2\mathrm{\Psi }_s`$, or,
$`(b^2+1)\mathrm{\Psi }_sb(\mathrm{\Psi }_{s1}+\mathrm{\Psi }_{s+1})=\lambda ^2\mathrm{\Psi }_s,`$ (A10)
is solved by $`\mathrm{\Psi }_s=\mathrm{exp}[\pm \alpha s](s=1,\mathrm{}N_s),`$ provided $`\lambda `$ and $`\alpha `$ satisfy
$`2b\mathrm{cosh}(\alpha )+(b^2+1\lambda ^2)=0.`$ (A11)
The first and the last rows of the secular equation can be satisfied by a linear combination of exponential solutions, $`\mathrm{\Psi }_s=\mathrm{exp}[\alpha (s1)]+A\mathrm{exp}[\alpha (N_ss)]`$, where $`A`$ is a constant to be determined:
$`(b^2+1\lambda ^2)(1+Ae^{\alpha (N_s1)})b(e^\alpha +Ae^{\alpha (N_s2)})+m_fb(e^{\alpha (N_s1)}+A)=0,`$ (A12)
$`(b^2+m_f^2\lambda ^2)(e^{\alpha (N_s1)}+A)b(e^{\alpha (N_s2)}+Ae^\alpha )+m_fb(1+Ae^{\alpha (N_s1)})=0.`$ (A13)
Using Eq. (A11) to eliminate $`\lambda `$ from the above, we get
$`2\mathrm{cosh}(\alpha )(1+Ae^{\alpha (N_s1)})(e^\alpha +Ae^{\alpha (N_s2)}`$ $`)+m_f(e^{\alpha (N_s1)}+A)=0,`$ (A14)
$`(2\mathrm{cosh}(\alpha )(1m_f^2)/2b)(e^{\alpha (N_s1)}+A)`$ (A15)
$`(e^{\alpha (N_s2)}`$ $`+Ae^\alpha )+m_f(1+Ae^{\alpha (N_s1)})=0.`$ (A16)
Eliminating A from Eq. (A15) and rearranging terms, we have
$`e^{2\alpha }e^\alpha /bm_f^2(1e^\alpha /b)`$ $`+2m_fe^{\alpha N_s}(e^{2\alpha }1)`$ (A17)
$`+m_f^2e^{\alpha (2N_s2)}(e^\alpha `$ $`1/b)e^{\alpha 2N_s}(1e^\alpha /b)=0.`$ (A18)
In the $`N_s=\mathrm{},m_f=0`$ limit, $`e^\alpha =1/b`$. Assuming $`e^\alpha =1/b+\delta `$ and keeping terms linear in $`\delta `$, we find
$`\delta (m_f+b^{N_s})^2(1b^2)/b.`$ (A19)
Finally, substituting $`e^\alpha =1/b+\delta `$ into Eq. (A11), we get the eigenvalue,
$`\lambda ^2=b^2+1b(e^\alpha +e^\alpha )b(1b^2)\delta =(1b^2)^2(m_f+b^{N_s})^2,`$ (A20)
or $`\lambda =(1b^2)(m_f+b^{N_s})`$, as quoted in Ref. .
|
warning/0002/cond-mat0002134.html
|
ar5iv
|
text
|
# Nature of the Spin Glass State
\[
## Abstract
The nature of the spin glass state is investigated by studying changes to the ground state when a weak perturbation is applied to the bulk of the system. We consider short range models in three and four dimensions and the infinite range Sherrington-Kirkpatrick (SK) and Viana-Bray models. Our results for the SK and Viana-Bray models agree with the replica symmetry breaking picture. The data for the short range models fit naturally a picture in which there are large scale excitations which cost a finite energy but whose surface has a fractal dimension, $`d_s`$, less than the space dimension $`d`$. We also discuss a possible crossover to other behavior at larger length scales than the sizes studied.
\]
The nature of ordering in spin glasses below the transition temperature, $`T_c`$, remains a controversial issue. Two theories have been extensively discussed: the “droplet theory” proposed by Fisher and Huse (see also Refs. ), and the replica symmetry breaking (RSB) theory of Parisi. An important difference between these theories concerns the number of large-scale, low energy excitations. In the RSB theory, which follows the exact solution of the infinite range SK model, there are excitations which involve turning over a finite fraction of the spins and which cost only a finite energy even in the thermodynamic limit. Furthermore, the surface of these excitations is argued to be space filling, i.e. the fractal dimension of their surface, $`d_s`$, is equal to the space dimension, $`d`$. By contrast, in the droplet theory, the lowest energy excitation which involves a given spin and which has linear spatial extent $`L`$ typically costs an energy of order $`L^\theta `$, where $`\theta `$ is a (positive) exponent. Hence, in the thermodynamic limit, excitations which flip a finite fraction of the spins cost an infinite energy. Also, the surface of these excitations is not space filling, i.e. $`d_s<d`$.
Recently we investigated this issue by looking at how spin glass ground states in two and three dimensions change upon changing the boundary conditions. Extrapolating from the range of sizes studied to the thermodynamic limit, our results suggest that the low energy excitations have $`d_s<d`$. Similar results were found in two dimensions by Middleton. In this paper, following a suggestion by Fisher, we apply a perturbation to the ground states in the bulk rather than at the surface. The motivation for this is two-fold: (i) We can apply the same method both to models with short range interactions and to infinite range models, like the SK model, and so can verify that the method is able to distinguish between the RSB picture, which is believed to apply to infinite range models, and some other picture which may apply to short range models. (ii) It is possible that there are other low energy excitations which are not excited by changing the boundary conditions.
We consider the short-range Ising spin glass in three and four dimensions, and, in addition, the SK and Viana-Bray models. The latter is infinite range but with a finite average coordination number $`z`$, and is expected to show RSB behavior. All these models have a finite transition temperature.
Our results for the SK and Viana-Bray models show clearly the validity of the RSB picture. However, for the short range models, our data is consistent with a picture suggested by Krzakala and Martin where there are extensive excitations with finite energy, i.e. their energy varies as $`L^\theta ^{}`$ with $`\theta ^{}=0`$, but $`d_s<d`$. In three dimensions, this picture is difficult to differentiate from the droplet picture where the energy varies as $`L^\theta `$, because of the small value of $`\theta `$ ($`0.2`$, obtained from the magnitude of the change of the ground state energy when the boundary conditions are changed from periodic to anti-periodic). It is easier to distinguish the two pictures in 4-D, even though the range of $`L`$ is less, because $`\theta `$ is much larger ($`0.7`$).
The Hamiltonian is given by
$$=\underset{i,j}{}J_{ij}S_iS_j,$$
(1)
where, for the short range case, the sites $`i`$ lie on a simple cubic lattice in dimension $`d=3`$ or 4 with $`N=L^d`$ sites ($`L8`$ in 3-D, $`L5`$ in 4-D), $`S_i=\pm 1`$, and the $`J_{ij}`$ are nearest-neighbor interactions chosen from a Gaussian distribution with zero mean and standard deviation unity. Periodic boundary conditions are applied. For the SK model there are interactions between all pairs chosen from a Gaussian distribution of width $`1/\sqrt{N1}`$, where $`N199`$. For the Viana-Bray model each spin is connected with $`z=6`$ spins on average, chosen randomly, the width of the Gaussian distribution is unity, and the range of sizes is $`N399`$. To determine the ground state we use a hybrid genetic algorithm introduced by Pal, as discussed elsewhere.
Let $`S_i^{(0)}`$ be the spin configuration in the ground state for a given set of bonds. Having found $`S_i^{(0)}`$, we then add a perturbation to the Hamiltonian designed to increase the energy of the ground state relative to the other states, and so possibly induce a change in the ground state. This perturbation, which depends upon a positive parameter $`ϵ`$, changes the interactions $`J_{ij}`$ by an amount proportional to $`S_i^{(0)}S_j^{(0)}`$, i.e.
$$\mathrm{\Delta }(ϵ)=ϵ\frac{1}{N_b}\underset{i,j}{}S_i^{(0)}S_j^{(0)}S_iS_j,$$
(2)
where $`N_b`$ is the number of bonds in the Hamiltonian. The energy of the ground state will thus increase exactly by an amount $`\mathrm{\Delta }E^{(0)}=ϵ.`$ The energy of any other state, $`\alpha `$ say, will increase by the lesser amount $`\mathrm{\Delta }E^{(\alpha )}=ϵq_l^{(0,\alpha )},`$ where $`q_l^{(0,\alpha )}`$ is the “link overlap” between the states “0” and $`\alpha `$, defined by
$$q_l^{(0,\alpha )}=\frac{1}{N_b}\underset{i,j}{}S_i^{(0)}S_j^{(0)}S_i^{(\alpha )}S_j^{(\alpha )},$$
(3)
in which the sum is over all the $`N_b`$ pairs where there are interactions. Note that the total energy of the states is changed by an amount of order unity.
The decrease in the energy difference between a low energy excited state and the ground state is given by
$$\delta E^{(\alpha )}=\mathrm{\Delta }E^{(0)}\mathrm{\Delta }E^{(\alpha )}=ϵ(1q_l^{(0,\alpha )}).$$
(4)
If this exceeds the original difference in energy, $`E^{(\alpha )}E^{(0)}`$, for at least one of the excited states, then the ground state will change due to the perturbation. We denote the new ground state spin configuration by $`\stackrel{~}{S}_i^{(0)}`$, and indicate by $`q_l`$ and $`q`$, with no indices, the link- and spin-overlap between the new and old ground states.
Next we discuss the expected behavior of $`q`$ and $`q_l`$ for the various models. For the SK model, it is easy to derive the trivial relation, $`q_l=q^2`$ (for large $`N`$). Since RSB theory is expected to be correct, there are some excited states which cost a finite energy and which have an overlap $`q`$ less than unity. According to Eq. (4), these have a finite probability of becoming the new ground state. Hence the average value of $`q`$ and $`q_l`$ over many samples, denoted by $`[\mathrm{}]_{\mathrm{av}}`$, should tend to a constant less than unity in the thermodynamic limit. This behavior is shown in the inset of Fig. 1. For the Viana-Bray model, where there is no trivial connection between $`q`$ and $`q_l`$, we show in Fig. 1 data for $`R=(1[q_l]_{\mathrm{av}})/(1[q]_{\mathrm{av}})`$ for several values of $`ϵ`$. This also appears to saturate. We plot this ratio rather than $`[q]_{\mathrm{av}}`$ or $`[q_l]_{\mathrm{av}}`$ for better comparison with the short range case below. For both models we took $`ϵ`$ to be a multiple of the transition temperature (the mean field approximation to it, $`T_c^{MF}=\sqrt{z}`$, for the Viana-Bray model), so that a perturbation of comparable magnitude was applied in both cases.
What do we expect for the short range models? In the RSB theory, $`1[q]_{av}`$ and $`1[q_l]_{av}`$ (and hence the ratio $`R`$) should saturate to a finite value for large $`L`$. To derive the prediction of the droplet theory, suppose that the energy to create an excitation of linear dimension, $`l`$, has a characteristic scale of $`l^\theta ^{}`$ (we use $`\theta ^{}`$ rather than $`\theta `$ to allow for the possibility that this exponent is different from the one found by changing the boundary conditions). Let us assume that large clusters ($`lL`$) dominate and ask for the probability that a large cluster is excited. The energy gained from the perturbation is $`ϵ(1q_l)ϵ/L^{(dd_s)}`$ since $`1/L^{(dd_s)}`$ is the fraction of the system containing the surface (i.e. the broken bonds) of the cluster. Generally this will not be able to overcome the $`L^\theta ^{}`$ energy cost to create the cluster. However, there is a distribution of cluster energies and if we make the plausible hypothesis that this distribution has a finite weight at the origin, then the probability that the cluster is excited is proportional to $`1/L^{dd_s+\theta ^{}}`$. In other words
$$1[q]_{\mathrm{av}}ϵ/L^\mu \mathrm{w}here\mu =\theta ^{}+dd_s.$$
(5)
As discussed above, $`1q_l`$ is of order $`1/L^{(dd_s)}`$ and so
$$1[q_l]_{\mathrm{av}}ϵ/L^{\mu _l}\mathrm{w}here\mu _l=\theta ^{}+2(dd_s).$$
(6)
Similar expressions have been derived by Drossel et al. in another context. Eqs. (5) and (6) are expected to be valid only asymptotically in the limit $`ϵ0`$. In order to include data for a range of values of $`ϵ`$ we note that the data is expected to scale as
$`1[q]_{\mathrm{av}}`$ $`=`$ $`F_q(ϵ/L^\mu ),`$ (7)
$`1[q_l]_{\mathrm{av}}`$ $`=`$ $`L^{(dd_s)}F_{q_l}(ϵ/L^\mu ),`$ (8)
where the scaling functions $`F_q(x)`$ and $`F_{q_l}(x)`$ both vary linearly for small $`x`$. Note that the above discussion applies also to a picture in which $`\theta ^{}=0`$ and $`d_s<d`$.
A scaling plot of our results for $`1[q]_{\mathrm{av}}`$ in 3D is shown in Fig. 2. We consider a range of $`ϵ`$ from $`\sqrt{6}/4`$ to $`4\sqrt{6}`$ (note that $`T_c^{MF}=\sqrt{6}`$) and find that the data collapse well onto the form expected in Eq. (8) with $`\mu =0.44\pm 0.02.`$
It is also convenient to plot the ratio $`R`$, which represents the surface to volume ratio of the excited clusters. This has a rather weak dependence on $`ϵ`$ and, as shown in Fig. 3, the data for each of the values of $`ϵ`$ fits well the power law behavior $`L^{(dd_s)}`$, expected from Eqs. (5) and (6), with $`dd_s`$ between 0.40 and 0.41 (the goodness of fit parameter, $`Q`$, is $`0.07,0.03,0.85,0.23,0.10`$, in order of increasing $`ϵ`$). The inset to Fig. 3 shows that there are small deviations from the asymptotic behaviour, which can be accounted for by a scaling function with the same value of $`\mu `$ as in Fig. 2 and with
$$dd_s=0.42\pm 0.02(3D).$$
(9)
From this value of $`\mu `$ and Eqs. (5) and (9) we find
$$\theta ^{}=0.02\pm 0.03(3D).$$
(10)
In order to test the RSB prediction, we tried fits of the form $`R=a+b/L^c`$, which give $`a=0.28\pm 0.18,0.01\pm 0.14,0.04\pm 0.11`$, and $`0.28\pm 0.18`$ ($`Q=0.08,0.01,0.72`$, and 0.52) for $`ϵ/\tau =0.25,0.5,1`$ and $`2`$. These are consistent with $`a=0`$ though a fairly small positive value, which would imply $`d_s=d`$, cannot be ruled out. For $`ϵ/\tau =4`$ the fit gives a small positive value, $`0.18\pm 0.07`$ ($`Q=0.79`$), but this is likely too large a value of $`ϵ`$ to be in the asymptotic regime for these sizes (see the inset of Fig. 3). The form $`R=a+b/L+c/L^2`$ also fits reasonably well the data and gives $`a`$ between 0.41 and 0.48 ($`Q=0.16,0.03,0.82,0.80,0.16`$). However, for both forms the data are very far from the asymptotic limit $`Ra`$ for the sizes considered, unlike for the Viana-Bray model (compare the main parts of Figs. 1 and 3). By contrast, the deviation from the asymptotic behavior $`RL^{(dd_s)}`$ is quite small (see the inset of Fig. 3).
In Fig. 4 we show analogous results in 4-D. The calculations were performed for two different values $`ϵ=\sqrt{8}/4`$ and $`\sqrt{8}`$ ($`=T_c^{MF}`$). The exponents are essentially the same for these two values of the perturbation and the fits give $`\mu =0.26\pm 0.04`$, $`dd_s=0.23\pm 0.02`$ , and so from Eq. (5) we get our main results for 4D:
$$\theta ^{}=0.03\pm 0.05,dd_s=0.23\pm 0.02(4D).$$
(11)
The data in Fig. 4 is consistent with the scaling form in Eq. (8) but the data for the two values of $`ϵ`$ are too widely separated to demonstrate scaling.
Interestingly, our results in both 3-D and 4-D are consistent with $`\theta ^{}=0`$, and, within the error bars, (which are purely statistical) incompatible with the relation $`\theta ^{}=\theta `$, since $`\theta 0.20`$ in 3-D and $`\theta 0.7`$ in 4-D. In 3-D, $`\theta \theta ^{}`$ is small, but in 4-D this difference is larger and hence the conclusion that $`\theta ^{}\theta `$ is stronger. However, the conclusion that $`dd_s>0`$ is less strong in 4-D because our value for $`dd_s`$ is quite small and the range of sizes is smaller than in 3-D.
It would be interesting, in future work, to study the nature of these excitations to see how they differ from the excitation of energy $`L^\theta `$ (with $`\theta >0`$) induced by boundary condition changes. In particular, if their volume is space filling, one would expect a non-trivial order parameter distribution, $`P(q)`$, at finite temperatures.
To conclude, an interpretation of our results for short range models which is natural, in that it fits the data with a minimum number of parameters and with small corrections to scaling, is that there are large-scale low energy excitations which cost a finite energy, and whose surface has fractal dimension less than $`d`$. This picture differs from the one suggested by Houdayer and Martin, in which $`d_s=d`$. Furthermore, the results for short range models appear quite different from those of the mean-field like Viana-Bray model for samples with a similar coordination number and a similar number of spins. Other scenarios, such as the droplet theory (with $`\theta ^{}=\theta (>0)`$) or an RSB picture (where $`\theta ^{}=0,dd_s=0`$), require larger corrections to scaling, but we cannot rule out the possibility of crossover to one of these behaviors at larger sizes.
We would like to thank D. S. Fisher for suggesting this line of enquiry, and for many stimulating comments. We also acknowledge useful discussions and correspondence with G. Parisi, E. Marinari, O. Martin, M. Mézard and J.-P. Bouchaud. We are grateful to D. A. Huse, M. A. Moore and A. J. Bray for suggesting the scaling plot in Fig. 2 and one of the referees for suggesting plotting the ratio $`R`$. This work was supported by the National Science Foundation under grant DMR 9713977. M.P. also is supported in part by a fellowship of Fondazione Angelo Della Riccia. The numerical calculations were supported by computer time from the National Partnership for Advanced Computational Infrastructure.
|
warning/0002/hep-th0002177.html
|
ar5iv
|
text
|
# References
SINP-TNP/00-05
SU-4240-715
Quantum Field Theories on Null Surfaces
Kumar S. Gupta<sup>1</sup><sup>1</sup>1E-mail: gupta@tnp.saha.ernet.in, Badis Ydri<sup>+</sup><sup>2</sup><sup>2</sup>2E-mail: idri@suhep.phy.syr.edu.
Saha Institute of Nuclear Physics, 1/AF Bidhannagar,
Calcutta - 700 064, India.
<sup>+</sup>Department of Physics , Syracuse University ,
Syracuse, NY 13244-1130, U.S.A.
## Abstract
We study the behaviour of quantum field theories defined on a surface $`S`$ as it tends to a null surface $`S_n`$. In the case of a real, free scalar field theory the above limiting procedure reduces the system to one with a finite number of degrees of freedom. This system is shown to admit a one parameter family of inequivalent quantizations. A duality symmetry present in the model can be used to remove the quantum ambiguity at the self-dual point . In the case of the non-linear $`\sigma `$-model with the Wess-Zumino-Witten term a similar limiting behaviour is obtained. The quantization ambiguity in this case however cannot be removed by any means.
PACS Numbers : 11.10Kk, 11.15Tk, 11.10.Ef
1. Introduction
Quantum Field Theory (QFT) on null surfaces have been studied in different contexts for a long time. One example of such theories consists of QFT on the light cone . Analysis of these systems has led to the discovery of a rich underlying structure of field theories and gauge theories on null surfaces . Another system of interest in this context involves the dynamics of the degrees of freedom on the horizon of a black hole, the horizon being a null surface. Analysis of the boundary field theories on black hole horizons has recently led to a new understanding of black hole entropy and other related issues .
Suppose that a field theory is defined on a surface $`S`$ which is embedded in a flat Minkowskian manifold $``$. Let the embedding of $`S`$ in $``$ be parametrized by a quantity $`v`$ whose limiting value $`v_n`$ corresponds to a null surface $`S_n`$ . It is natural to ask the question as to how a field theory defined on $`S`$ evolves as the parameter $`v`$ varies. In particular, a field theory defined on a null surface $`S_n`$ can be thought of as a limit of the theory defined on $`S`$ as $`vv_n`$. It is this limiting case that we propose to investigate in this paper.
The metric $`h`$ induced on $`S`$ from $``$ is a function of the parameter $`v`$. When $`v`$ is away from its limiting value $`v_n`$, the induced metric on $`S`$ is well defined. However, as $`vv_n`$ the induced metric $`h`$ tends to become degenerate. Correspondingly, any regular metric based action defined on a surface $`S`$ fails to have a well defined limit as $`S`$ tends to $`S_n`$ .
A physical example of where this scenario may occur can be described as follows. Considering a gravitationally collapsing sperical shell $`S`$ of dust on which some field theory is defined. The collapsing surface $`S`$ at any stage can be parametrised by a quantity $`v`$. If the situation is such that the system eventually tends to a black hole, the surface $`S`$ finally would tend to a null surface $`S_n`$. Field theories on the surface of such a black hole could be studied using the above mentioned limiting procedure.
In this paper we analyze the behaviour of field theories on a surface $`S`$ as $`SS_n`$. The analysis is based on specific examples where all the issues involved can be seen in an explicit fashion. In Section 2 we study the case of an abelian, free scalar field theory. In the limiting case this model reduces to a system with finite number of degrees of freedom that admits a one parameter family of inequivalent quantizations. This model also exhibits a type of duality symmetry. The quantization ambiguity can be removed if the system is at the self-dual point. Section 3 describes the analysis as applied to the $`SU(l)`$ Wess-Zumino-Witten (WZW) model. The parameter in front of the action in this case is constrained from topological considerations. In the limit of $`SS_n`$, the quantum theory in this case is described by a finite degrees of freedom and is characterized by an arbitrary parameter just as in the scalar field theory . We conclude the paper in Section 4 with a summary and outlook.
2. Scalar Field
Let $``$ be a flat Minkowskian manifold in 2+1 dimensions whose spatial slice has the topology of a cylinder $`S^1\times R`$ . Let $`r`$ be the radius of $`S^1`$ and let $`\theta `$ be the angle spanning it. Consider the following flat metric in $``$ given by
$$ds^2=dt^2dz^2r^2d\theta ^2,$$
(2.1)
Let $`S`$ given by $`z=vt`$ be a surface embedded in $``$ where $`v`$ is the parameter defining the embedding. The limiting value of this parameter is given by $`v=1`$. The pull-back of the above metric in $``$ to the time-like surface $`S`$ can be written as
$$ds^2|_{z=vt}=(1v^2)dt^2r^2d\theta ^2=h_{ab}dy^ady^b.$$
(2.2)
As $`v1`$, the surface $`S`$ tends to the null surface $`S_n`$ given by $`z=t`$ . From Eqn. (2.2) it is easilly seen that the metric $`h_{ab}`$ inudced on $`S`$ is degenerate as $`SS_n`$.
Consider a single real scalar field $`\varphi `$ which is defined on the surface $`S`$. We will assume that the scalar field is valued in a circle. As we shall see later, the degeneracy of the metric in the limit of $`v1`$ leads to a Hamiltonian that is ill-defined. In order to address this problem we consider a renormalized field $`f\varphi `$ where $`f`$ is the renormalization parameter. The action for such a real scalar field can be written as
$$𝒮=_S\sqrt{h}d^2y=\frac{f^2}{8\pi }_S\sqrt{h}d^2yh^{ab}_a\varphi _b\varphi .$$
(2.3)
As is evident from Eqn. (2.3), $`f`$ can also be interpreted as the coupling constant of this model.
The field $`\varphi `$ obeys the equation of motion
$$h^{ab}_a_b\varphi =0.$$
(2.4)
In terms of a variable $`x=r\gamma \theta `$ where $`\gamma =\frac{1}{\sqrt{1v^2}}`$ , the above action has the form
$`𝒮`$ $`=`$ $`{\displaystyle 𝑑t_0^{x_0}}`$
$`=`$ $`{\displaystyle \frac{f^2}{8\pi }}{\displaystyle 𝑑t_0^{x_0}𝑑x[(_t\varphi )^2(_x\varphi )^2]}.`$
where $`x_0=2\pi r\gamma `$ is the period of the variable $`x`$. In terms of the variables $`x`$ and $`t`$ the action $`𝒮`$ is that of a free scalar field in 1+1 dimensions with a diagonal metric of signature (1,-1) . The equation of motion following from the action $`𝒮`$ is
$$[(_t)^2(_x)^2]\varphi =0.$$
(2.6)
2.1 Canonical Quantization
The mode expansion for the real field $`\varphi `$ defined on a circle has the form
$$\varphi (t,x)=\varphi _0+\varphi _{\mathrm{osc}}(t,x)$$
(2.7)
where
$$\varphi _0=Q+\frac{N}{r\gamma }x+\frac{2P}{f^2}t,$$
(2.8)
and
$`\varphi _{\mathrm{osc}}(t,x)`$ $`=`$ $`{\displaystyle \frac{1}{f}}{\displaystyle \underset{k>0}{}}[{\displaystyle \frac{A_k}{\sqrt{k}}}e^{ikx_+}+{\displaystyle \frac{A_{k}^{}{}_{}{}^{+}}{\sqrt{k}}}e^{ikx_+}`$
$`+{\displaystyle \frac{B_k}{\sqrt{k}}}e^{ikx_{}}+{\displaystyle \frac{B_{k}^{}{}_{}{}^{+}}{\sqrt{k}}}e^{ikx_{}}].`$
In Eqn. (2.9) $`x_\pm =t\pm x`$ and $`B_k=A_k`$ .
As mentioned before, the field $`\varphi `$ is assuemd to be valued in a circle for all time $`t`$. It therefore satisfies the consistency condition
$$\varphi _0(t,x+2\pi r\gamma )=\varphi _0(t,x)+2\pi m,$$
(2.10)
where m is an integer. From the above mode expansion of the field $`\varphi `$ and Eqn. (2.10) it follows that
$$kr\gamma =n(\mathrm{n}\mathrm{is}\mathrm{an}\mathrm{integer}),$$
(2.11)
and
$$N=m.$$
(2.12)
The canonical momentum conjugate to the field $`\varphi `$ is defined by
$$\pi (t,x)=\frac{f^2}{4\pi }_t\varphi .$$
(2.13)
Using Eqns. (7), (8) and (13) we get that
$$\pi (t,x)=\frac{f^2}{4\pi }_t\varphi _{\mathrm{osc}}+\frac{P}{2\pi }.$$
(2.14)
In the quantum theory, the wave-functional $`\psi `$ is a function of the field $`\varphi `$. Since $`\varphi (x)`$ is identified with $`\varphi (x)+2\pi `$, the $`\varphi _0`$ dependency of the wave-functional $`\psi `$ satisfies the condition
$$\psi (\varphi _0+2\pi )=\mathrm{e}^{i2\pi \alpha }\psi (\varphi _0),$$
(2.15)
where $`\alpha `$ is a real number between 0 and 1. Since we are dealing with bosonic variables alone, it is natural to choose $`\alpha =0`$ corresponding to periodic boundary condition. It therefore follows from Eqn. (2.15) that
$$\psi _m(\varphi _0)=\mathrm{e}^{ip\varphi _0},(\mathrm{p}\mathrm{is}\mathrm{an}\mathrm{integer})$$
(2.16)
are the eigenfunctions of the operator
$$P=𝑑x\pi (x)=i\frac{}{\varphi _0}$$
(2.17)
and the corresponding eigenvalues are $`p`$. The spectrum of the operator $`P`$ therefore consists of integers $`p`$.
The canonical commutaion relations of the basic field variables are given by
$$[\varphi (t,x),\pi (t,y)]=i\delta (xy)$$
(2.18)
and
$$[\varphi (t,x),\varphi (t,y)]=[\pi (t,x),\pi (t,y)]=0.$$
(2.19)
It follows that
$$[Q,P]=i,$$
(2.20)
$$[A_k,A_{k}^{^{}}{}_{}{}^{+}]=\delta _{kk^{^{}}}$$
(2.21)
and all other commutation relations are zero .
2.2 Ground State Energy
The Hamiltonian of the system is given by
$`H`$ $`=`$ $`{\displaystyle _0^{x_0}}𝑑x[\pi (t,x)_t\varphi (t,x)]`$
$`=`$ $`{\displaystyle \frac{f^2}{8\pi }}{\displaystyle _0^{x_0}}𝑑x[(_t\varphi )^2+(_x\varphi )^2]`$
$`=`$ $`H_0+H_{\mathrm{osc}},`$
where $`H_0`$ and $`H_{\mathrm{osc}}`$ are the Hamiltonians for the zero (or winding) and oscillating modes respectively. They are given by
$$H_0=\frac{r\gamma }{f^2}P^2+\frac{f^2}{4r\gamma }m^2$$
(2.23)
and
$$H_{\mathrm{osc}}=r\gamma \underset{k0}{}|k|[A_k^+A_k+\frac{1}{2}]$$
(2.24)
A given zero mode sector is characterized by the integers $`p`$ and $`m`$. Let the ground state (or the vacuum) in this sector be denoted by $`|0>_{pm}`$. The vacuum satisfies the condition
$$H_{\mathrm{osc}}|0_{pm}=0$$
(2.25)
where in the above $`H_{\mathrm{osc}}`$ is assumed to have been normal ordered and the zero-point energy has been subtracted. The ground state energy in a given zero mode sector characterised by the integers $`p`$ and $`m`$ satisfies the equation
$$H|0_{pm}=(H_0+H_{\mathrm{osc}})|0_{pm}=E_G|0_{pm}$$
(2.26)
where
$$E_G=\frac{r\gamma }{f^2}p^2+\frac{f^2}{4r\gamma }m^2$$
(2.27)
and p is the eigenvalue of the operator P in the ground state under consideration.
Let us now define the quantity $`\stackrel{~}{r}`$ by
$$\stackrel{~}{r}=\frac{2\gamma r}{f^2}$$
(2.28)
which is an effective radius for the system. Then the ground state energy can be written as
$$E_G=\frac{1}{2}[\stackrel{~}{r}p^2+\frac{1}{\stackrel{~}{r}}m^2].$$
(2.29)
In the limit when $`v1`$, the induced metric $`h_{ab}`$ in Eqn. (2.2) tends to blow up. In this limit $`\stackrel{~}{r}`$ and the ground state energy $`E_G`$ also become undefined. It may thus seem that there is now smooth way of taking the aforementioned limit.
We can however use the following “renormalization group inspired” prescription to make this limit well defined. Let us first note that the quantity $`v`$ used to define the embedding of $`S`$ in $``$ could be thought as a regulator. We are really interested in the situation $`SS_n`$ or $`v1`$, which can be thought of as the regulator being removed. In order for the ground state energy to have a smooth behaviour in this limit we postulate that the coupling constant $`f`$ is a function of the regulator $`v`$. The functional dependence of $`f`$ on $`v`$ is to be determined from the physical condition that as $`v1`$, the ground state energy should be independent of $`v`$. In other words, as $`v1`$
$$\frac{dE_G}{dv}=0.$$
(2.30)
Using the equation $`(2.27)`$ we find that
$$\frac{dE_G}{dv}=\frac{1}{2\stackrel{~}{r}}\frac{d\stackrel{~}{r}}{dv}(\stackrel{~}{r}p^2\frac{1}{\stackrel{~}{r}}m^2)$$
(2.31)
The second term cannot be zero for all values of $`v`$ because $`p`$ and $`m`$ are fixed numbers . We are hence left with the condition
$$\frac{1}{2\stackrel{~}{r}}\frac{d\stackrel{~}{r}}{dv}=0$$
(2.32)
This is possible only if in the limit $`v1`$,
$$f(v)=F\sqrt{\gamma }$$
(2.33)
where $`F`$ is a constant parameter and can be thought of as the “renormalized” coupling constant.
Using Eqns. (28), (29) and (33), the ground state energy has the form
$$E_G=\frac{r}{F^2}p^2+\frac{F^2}{4r}m^2.$$
(2.34)
In any given zero-mode sector, different choices of the parameter $`F`$ would lead to different values of the ground state energy and hence to inequivalent quantum field theories. The parameter $`F`$ is not determined by the above analysis and can presumably be obtained from empirical considerations.
It should be noted that only the zero mode sector has information about the coupling constant $`f`$ and consequently of the parameter $`F`$. The expression for $`H_{\mathrm{osc}}`$ (cf. Eqn. (2.24)) is independent of $`f`$. As $`v1`$ , we see from Eqn. (2.11) that for any given $`n`$, $`k0`$. However, from Eqn. (2.24), only nonzero $`k`$ contributes to $`H_{\mathrm{osc}}`$. Hence the contribution to the Hamiltonian coming from the the oscillatory modes become energetically unfavourable as $`SS_n`$. We therefore arrive at the conclusion that as $`SS_n`$, the model under consideration reduces to a quantum mechanical system with a finite number of degrees of freedom given by the zero modes of the original problem. The Hamiltonian of this reduced system is still given by Eqn. (2.23) where $`f`$ is given by Eqn. (2.33) and the corresponding eigenavlues are given by Eqn. (2.34).
2.3 Duality
A real scalar field system valued in a circle has a well known duality symmetry . A remnant of that can be seen in the reduced system obtained above. For a fixed $`r`$, our system is characterized by the numbers $`p`$, $`m`$ and $`F`$. Under the transformations
$`p`$ $``$ $`m`$
$`m`$ $``$ $`p`$
$`{\displaystyle \frac{r}{F^2}}`$ $``$ $`{\displaystyle \frac{F^2}{4r}}`$ (2.35)
$`E_G`$ belonging to two different configurations get interchanged. This is analogous to a T-duality.
The duality symmetry by itself imposes no restriction on $`F`$. There is however a special configuration, namely the “self-dual” point where the duality symmetry can be used to fix the arbitrariness in $`F`$. The “self-dual” point is given by
$$\frac{r}{F^2}=\frac{F^2}{4r}.$$
(2.36)
At this special point $`F`$ therefore satisfies the condition
$$F^2=2r.$$
(2.37)
Using Eqns. (34) and (37), the ground state energy can them be expressed as
$$E_G=\frac{1}{2}[p^2+m^2]$$
(2.38)
which is independent of $`F`$. The limiting procedure described above thus leads to a unique quantization only at the self-dual point.
3. Non Linear $`\sigma `$ \- Model
Let $`𝒞`$ be a three dimensional manifold whose boundary $`𝒞`$ is a two dimensional Minkowskian manifold with a topology of $`S^1\mathrm{x}R`$. We identify $`𝒞`$ with a surface $`S`$ on which the induced metric $`h^{ab}`$ is given by Eqn(2.2) . The action for the non linear $`\sigma `$-model with a Wess-Zumino-Witten term (WZW) for a group G is given by
$`S`$ $`=`$ $`S_0+S_{\mathrm{wzw}},`$
$`S_0`$ $`=`$ $`A{\displaystyle _𝒞}d^2y\sqrt{h}h^{\mu \nu }Tr_\mu g_\nu g^1,`$
$`S_{\mathrm{wzw}}`$ $`=`$ $`{\displaystyle _𝒞}\mathrm{\Omega }=B{\displaystyle _𝒞}Tr(g^1dg)^3,`$ (3.1)
where $`A`$ and $`B`$ are constants and $`g`$ takes values in the group $`G`$. For simplicity we will assume in this section that $`G=SU(l)`$ with $`l>1`$ . The coefficient $`B`$ of the second term in the action is not arbitrary. From topological considerations $`B=\frac{n}{24\pi }`$ , where $`n`$ is an integer .
Let $`T_a,a=1,..r`$ be the generators of $`G`$ satisfying the comutation relations
$$[T_a,T_b]=if_{abc}T_c.$$
(3.2)
These generators are normalised in such a way that
$$TrT_aT_b=2\delta _{ab}.$$
(3.3)
The group $`G`$ can act on $`g`$ either from the left or from the right . The left action is given by
$$gg^{^{}}=g+igx_aT_a.$$
(3.4)
where $`x_a`$ are small parameters . The right action of the group is similarly given by
$$gg^{^{}}=g+ix_aT_ag.$$
(3.5)
For both cases, the variation of $`S`$ is
$`\delta S`$ $`=`$ $`\delta S_0+\delta S_{wzw},`$
$`\mathrm{where}\delta S_0`$ $`=`$ $`2A{\displaystyle d^2x\sqrt{h}h^{\mu \nu }Trg^1\delta g_\mu (g^1_\nu g)}+\mathrm{Total}\mathrm{derivatives}`$
$`\mathrm{and}\delta S_{wzw}`$ $`=`$ $`3B{\displaystyle d^2xϵ^{\mu \nu }Trg^1\delta g_\mu (g^1_\nu g)}.`$
By setting $`\delta S=0`$ we get as equations of motion
$$2A\sqrt{h}h^{\mu \nu }_\mu (g^1_\nu g)3Bϵ^{\mu \nu }_\mu (g^1_\nu g)=0.$$
(3.7)
In terms of the light cone coordinates
$$x_\pm =t\pm x,$$
(3.8)
Eqn. (3.7) can be expressed as
$$(2A3B)_+(g^1_{}g)+(2A+3B)_{}(g^1_+g)=0.$$
(3.9)
For the choice of $`A=\pm \frac{3B}{2}`$ the left and right movers decouple from each other and the equations of motion reduces to
$$_{}(g^1_\pm g)=0.$$
(3.10)
The currents arising from the left action of the group are given by
$`J_a^\mu `$ $`=`$ $`h^{\mu \nu }TrT_ag^1_\nu g,`$
$`\mathrm{with}J_a^0`$ $`=`$ $`\gamma ^2TrT_ag^1\dot{g}`$
$`\mathrm{and}J_a^1`$ $`=`$ $`{\displaystyle \frac{1}{r^2}}TrT_ag^1g^{^{}},`$
where $`\dot{g}=_tg`$ and $`g^{^{}}=_\theta g`$. Using Eqn. (3.11) the light-cone components of the currents can be written as
$`J_a^\pm `$ $`=`$ $`TrT_ag^1_\pm g`$ (3.12)
$`=`$ $`{\displaystyle \frac{1}{2\gamma ^2}}J_a^0{\displaystyle \frac{r}{2\gamma }}J_a^1.`$
The currents resulting from the right action of the group can also be found in a similar fashion. They are
$`\overline{J}_a^\pm `$ $`=`$ $`TrT_a(_\pm g)g^1`$ (3.13)
$`=`$ $`{\displaystyle \frac{1}{2\gamma ^2}}\overline{J}_a^0{\displaystyle \frac{r}{2\gamma }}\overline{J}_a^1,`$
where
$`\overline{J}_a^0`$ $`=`$ $`\gamma ^2TrT_a\dot{g}g^1`$
$`\overline{J}_a^1`$ $`=`$ $`{\displaystyle \frac{1}{r^2}}TrT_ag^{^{}}g^1.`$ (3.14)
3.1 The Canonical Formalism
Let $`\xi _i,i=1,\mathrm{},\mathrm{dim}G`$ be a set of local coordinates parametrizing the elements $`gG`$.In terms of these local coordinates $`S_0`$ can be expressed as
$`S_0`$ $`=`$ $`{\displaystyle d^2x_0},`$
$`_0`$ $`=`$ $`Ar\gamma Tr{\displaystyle \frac{g}{\xi _i}}{\displaystyle \frac{g^1}{\xi _j}}[\dot{\xi _i}\dot{\xi _j}{\displaystyle \frac{1}{r^2\gamma ^2}}\xi _i^{^{}}\xi _j^{^{}}].`$ (3.15)
We would also like to express the WZW part of the action in terms of the coordinates $`\xi `$. The WZW terms cannot be written globally in terms of a single set of local coordinates. To proceed we assume that the group manifold consists of a number of patches labelled by a parameter $`u`$. The restriction of $`\mathrm{\Omega }=BTr(g^1dg)^3`$ to any of these patches can be written as
$$\mathrm{\Omega }^u=d\omega ^u,$$
(3.16)
where $`\omega ^u`$ is a two form defined by
$$\omega ^u=\frac{1}{2}\omega _{ij}^ud\xi ^id\xi ^j.$$
(3.17)
From Eqns. (3.16) and (3.17), $`\mathrm{\Omega }^u`$ has the form
$$\mathrm{\Omega }^u=\frac{1}{6}\mathrm{\Omega }_{ijk}^ud\xi ^id\xi ^jd\xi ^k,$$
(3.18)
where $`\mathrm{\Omega }_{ijk}^u`$ is given by
$$\mathrm{\Omega }_{ijk}^u=\frac{\omega _{jk}^u}{\xi ^i}+\frac{\omega _{ki}^u}{\xi ^j}+\frac{\omega _{ij}^u}{\xi ^k}.$$
(3.19)
$`\mathrm{\Omega }_{ijk}^u`$ can also be expressed however in terms of the group element $`g`$ as
$$\mathrm{\Omega }_{ijk}^u=3BTr[g^1\frac{g}{\xi ^i},g^1\frac{g}{\xi ^j}]g^1\frac{g}{\xi ^k}.$$
(3.20)
In terms of the local coordinates the WZW action then takes the form
$`S_{wzw}`$ $`=`$ $`{\displaystyle \underset{u}{}}{\displaystyle _𝒞}_{wzw}^ud^2x,`$
$`\mathrm{where}_{wzw}^u`$ $`=`$ $`{\displaystyle \frac{1}{2}}\omega _{ij}^u_a\xi ^i_b\xi ^jϵ^{ab}.`$
We would next like to compute the canonical momentum $`P_i`$ conjugate to the coordinate $`\xi _i`$. Using Eqns. (3.15) and (3.21) $`P_i`$ can be written as
$`P_i`$ $`=`$ $`{\displaystyle \frac{_0}{\dot{\xi }_i}}+{\displaystyle \frac{_{wzw}}{\dot{\xi }_i}}`$
$`=`$ $`Ar\gamma Tr[{\displaystyle \frac{g}{\xi _i}}{\displaystyle \frac{g^1}{\xi _j}}\dot{\xi _j}+{\displaystyle \frac{g}{\xi _j}}{\displaystyle \frac{g^1}{\xi _i}}\dot{\xi _j}]\omega _{ij}^u_\theta \xi ^j`$
$`=`$ $`2Ar\gamma Trg^1\dot{g}g^1{\displaystyle \frac{g}{x^i}}\omega _{ij}^u_\theta \xi ^j.`$
The Hamiltonian density therefore has the form
$``$ $`=`$ $`P_i\xi _i_0_{wzw}`$
$`=`$ $`Ar\gamma Tr[(\dot{g}g^1)^2+{\displaystyle \frac{1}{r^2\gamma ^2}}(g^{^{}}g^1)^2].`$
As expected, there is no contribution from the WZW term to the Hamiltonian. In terms of the light cone currents defined in Eqn. (3.12) , the Hamiltonian density can be written as
$$=Ar\gamma [J_{a}^{+}{}_{}{}^{2}+J_{a}^{}{}_{}{}^{2}].$$
(3.24)
Next we turn to the commutation relations for this system. The basic commutation relations are given by
$$[\xi _a(\theta ),\xi _b(\theta ^{^{}})]=[P_a(\theta ),P_b(\theta ^{^{}})]=0$$
(3.25)
and
$$[\xi _a(\theta ),P_b(\theta ^{^{}})]=i\delta _{ab}\delta (\theta \theta ^{^{}})$$
(3.26)
From these commutation relations it follows that
$$[g(\theta ),P_b(\theta ^{^{}})]=i\frac{g(\theta )}{\xi _b}\delta (\theta \theta ^{^{}}).$$
(3.27)
Similarly we have
$$[g(\theta )^1,P_b(\theta ^{^{}})]=i\frac{g(\theta )^1}{\xi _b}\delta (\theta \theta ^{^{}}).$$
(3.28)
It is however useful to rewrite these commutation relations in a form that is independent of the local coordinates $`\xi `$ . To this end, we introduce a new set of functions $`\xi (x)`$ with the condition that $`\xi (0)=\xi `$ . The field $`g(\xi )`$ can be understood as the value of a field $`g(\xi (x))`$ at $`\xi (0)`$ . The field $`g(\xi (x))`$ is defined by
$$g(\xi (x))=g(\xi (0))\mathrm{exp}(ix_aT_a)$$
(3.29)
Differentiating with respect to $`x_a`$ and then setting $`x=0`$ we get the identity
$$N_b^a\frac{g}{\xi ^a}=ig(\xi )T_b$$
(3.30)
where $`N_b^a=\frac{\xi ^a}{x_b}|_{x=0}`$ can be proven to be nondegenerate . Using Eqn. (3.30), we can replace the phase space variables $`P_a`$ with new variables $`\mathrm{\Pi }_a`$ defined as
$`\mathrm{\Pi }_a`$ $`=`$ $`N_a^bP_b`$
$`=`$ $`2i{\displaystyle \frac{Ar}{\gamma }}J_a^0N_a^b\omega _{bc}^u_\theta \xi _c.`$
Using Eqns. (3.27), (3.28) ,(3.30) and $`(3.31)`$ it follows that
$$[g(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]=\delta (\theta \theta ^{^{}})g(\theta )T_b$$
(3.32)
and
$$[g^1(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]=\delta (\theta \theta ^{^{}})T_bg^1(\theta ).$$
(3.33)
The above commutators between $`g`$, $`g^1`$ and $`\mathrm{\Pi }`$ carry no explicit dependence on the local coordinates $`\xi `$. Finally, the commutator of the $`\mathrm{\Pi }`$’s is given by (see the appendix for the proof)
$$[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]=if_{abc}\mathrm{\Pi }_c(\theta )\delta (\theta \theta ^{^{}})$$
(3.34)
and all other commutation relations are trivial. Eqns (3.32), (3.33) and (3.34) embody the fundamental commutators for this system.
3.2 Current Algebra
We are now ready to calculate the current commutators. First we note that the expression for $`J_a^1(\theta )`$ contains no time derivative. It therefore follows that
$$[J_a^1(\theta ),J_b^1(\theta ^{^{}}]=0.$$
(3.35)
Next, by differentiating Eqn. (3.32) with respect to $`\theta `$ we get
$$[_\theta g(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]=\delta (\theta \theta ^{^{}})_\theta g(\theta )T_b_\theta \delta (\theta \theta ^{^{}})g(\theta )T_b.$$
(3.36)
Using the above relation and the definition of the current $`J_a^1`$ (cf. Eqn. (3.11)), it can be shown that
$$[\mathrm{\Pi }_a(\theta ),J_b^1(\theta ^{^{}})]=if_{abc}J_c^1(\theta )+\frac{TrT_aT_b}{r^2}_\theta \delta (\theta \theta ^{^{}}).$$
(3.37)
From Eqn. (3.31) we have $`J_a^0=i\gamma /2Ar[\mathrm{\Pi }_a+N_a^b\omega _{bc}^u_\theta \xi _c]`$. Using this expression for $`J_a^0`$ and Eqn. (3.37) we get the second current commutator as
$$[J_a^0(\theta ),J_b^1(\theta ^{^{}})]=\frac{\gamma }{2Ar}f_{abc}J_c^1(\theta )\delta (\theta \theta ^{^{}})+\frac{i\gamma }{2Ar^3}TrT_aT_b_\theta \delta (\theta \theta ^{^{}}).$$
(3.38)
Finally, a tedious calculation (the details are shown in the appendix) gives the last current commutator as
$$[J_a^0(\theta ),J_b^0(\theta ^{^{}})]=\frac{\gamma }{2Ar}f_{abc}J_c^0(\theta )\delta (\theta \theta ^{^{}})\frac{3B\gamma ^2}{4A^2}\delta (\theta \theta ^{^{}})f_{abc}J_c^1(\theta )$$
(3.39)
We would next like to obtain the commutators for the light-cone components of the currents . Using Eqns. (3.35), (3.38) and (3.39) we get the following currents algebra
$`[J_a^+(\theta ),J_b^+(\theta ^{^{}})]`$ $`=`$ $`{\displaystyle \frac{1}{8Ar\gamma ^3}}f_{abc}\delta (\theta \theta ^{^{}})J_c^0(\theta )`$
$`+`$ $`{\displaystyle \frac{1}{4A\gamma ^2}}(1{\displaystyle \frac{3B}{4A}})f_{abc}J_c^1(\theta )\delta (\theta \theta ^{^{}})`$
$`+`$ $`{\displaystyle \frac{1}{4iAr^2\gamma ^2}}TrT_aT_b_\theta \delta (\theta \theta ^{^{}}).`$
Let us now proceed by considering the two cases $`A=\frac{3B}{2}`$ and $`A=\frac{3B}{2}`$ separately . For the first case where $`\frac{3B}{2A}=1,`$ Eqn. (3.40) reduces to the Kac-Moody algebra
$$[J_a^+(\theta ),J_b^+(\theta ^{^{}})]=\frac{1}{4Ar\gamma }f_{abc}\delta (\theta \theta ^{^{}})J_c^+\frac{i}{2Ar^2\gamma ^2}\delta _{ab}_\theta \delta (\theta \theta ^{^{}}).$$
(3.41)
The currents $`\overline{J}_a^{}=TrT_a_{}gg^1`$ coming from the right action of the group would similarly generate another Kac-Moody algebra. To get the algebra generated by $`\overline{J}_a^{}`$ first note that the action is invariant under the transformations $`\theta \theta `$ and $`gg^1`$ . Under these transformations $`J_a^+\overline{J}_a^{}`$ and therefore the current commutator in Eqn. (3.41) becomes
$$[\overline{J}_a^{}(\theta ),\overline{J}_b^{}(\theta ^{^{}})]=\frac{1}{4Ar\gamma }f_{abc}\delta (\theta \theta ^{^{}})\overline{J}_c^{}+\frac{i}{2Ar^2\gamma ^2}\delta _{ab}_\theta \delta (\theta \theta ^{^{}}).$$
(3.42)
We also have
$$[J_a^+(\theta ),\overline{J}_b^{}(\theta ^{^{}})]=0$$
(3.43)
as the two currents come from the two commuting actions of the group on itself .
For the second point $`\frac{3B}{2A}=1`$ exactly the same arguments will lead to the two other commuting Kac-Moody algebras given by the currents $`J_a^{}`$ and $`\overline{J}_a^+`$ . The first currents algebra generated by $`J_a^{}`$ has the form $`(3.41)`$ with the substitution $`J_a^+J_a^{}`$ and $`AA`$ . The currents algebra corresponding to $`\overline{J}_a^+`$ is aobtained from $`(3.42)`$ by a similar substitution $`\overline{J}_a^{}\overline{J}_a^+`$ and $`AA`$ .
3.3 Mode Expansion
Let us consider the case when $`\frac{3B}{2A}=1`$ (the treatement of the case $`\frac{3B}{2A}=1`$ is exactly similar). We first express the the Hamiltonian density $``$ in terms of the two commuting set of currents $`J_a^+`$ and $`\overline{J}_a^{}`$ which are relevant to the case under consideration. To this end we note that a given element $`L`$ in the Lie algebra of $`G`$ can be written as $`L=\frac{T_a}{2}Tr(T_aL)`$ Using this we can then check that $`J_{a}^{}{}_{}{}^{2}=(\overline{J}_a^{})^2`$. The Hamiltonian density in Eqn. (3.24) can then be expressed as
$``$ $`=`$ $`Ar\gamma [(J_a^+)^2+(J_a^{})^2]`$ (3.44)
$`=`$ $`Ar\gamma [(J_a^+)^2+(\overline{J}_a^{})^2]`$
$`=`$ $`{\displaystyle \frac{1}{16Ar\gamma }}[(K_a^+)^2+(\overline{K}_a^{})^2],`$
where $`K_a^+`$ and $`\overline{K}_a^{}`$ are defined as
$`K_a^+`$ $`=`$ $`4Ar\gamma J_a^+,`$
$`\overline{K}_a^{}`$ $`=`$ $`4Ar\gamma \overline{J}_a^{}.`$ (3.45)
They satify the commutation relations
$$[K_a^+(\theta ),K_b^+(\theta ^{^{}})]=f_{abc}\delta (\theta \theta ^{^{}})K_c^+(\theta )8iA\delta _{ab}_\theta \delta (\theta \theta ^{^{}})$$
(3.46)
and
$$[\overline{K}_a^{}(\theta ),\overline{K}_b^{}(\theta ^{^{}})]=f_{abc}\delta (\theta \theta ^{^{}})\overline{K}_c^{}(\theta )+8iA\delta _{ab}_\theta \delta (\theta \theta ^{^{}}).$$
(3.47)
Next we proceed with the mode expansion of the currents. Let us first note that in terms of $`gG`$, the current $`K_a^+(x)`$ has the expression
$$K_a^+(x)=2Ar\gamma TrT_ag^1\dot{g}2ATrT_ag^1g^{^{}}.$$
(3.48)
A similar expression will hold for the current $`\overline{K}^{}`$ . The mode expansion for the two terms in the rhs of Eqn (3.48) are given by
$`TrT_ag^1\dot{g}`$ $`=`$ $`{\displaystyle \underset{k0}{}}J_a^0(k)e^{i(\omega tkx)}+J_a^0(0),`$
$`TrT_ag^1g^{^{}}`$ $`=`$ $`{\displaystyle \underset{k0}{}}J_a^1(k)e^{i(\omega tkx)}+J_a^1(0).`$ (3.49)
Next we can check using the periodicity requirement
$$J_a^\mu (t,x+x_0)=J_a^\mu (t,x)$$
(3.50)
that
$$kr\gamma =q$$
(3.51)
where $`q`$ is an integer . Using the equations of motion (3.10), we see that $`\omega =k`$ for $`J_a^+`$ and $`\omega =k`$ for $`J_a^{}`$.
Now by using Eqn. (3.49) in (3.48), we get
$`K_a^+(x_+)`$ $`=`$ $`{\displaystyle \frac{1}{2i\pi }}[{\displaystyle \underset{k0}{}}K_a^+(k)e^{ikx_+}+𝒫_a]`$
$`\mathrm{where}\mathrm{K}_\mathrm{a}^+(\mathrm{k})`$ $`=`$ $`4i\pi Ar\gamma [J_a^0(k)+{\displaystyle \frac{1}{r\gamma }}J_a^1(k)]`$
$`\mathrm{and}𝒫_\mathrm{a}`$ $`=`$ $`4i\pi Ar\gamma [J_a^0(0)+{\displaystyle \frac{1}{r\gamma }}J_a^1(0)].`$ (3.52)
Similarly ge wet that,
$`\overline{K}_a^{}(x_{})`$ $`=`$ $`{\displaystyle \frac{1}{2i\pi }}[{\displaystyle \underset{k0}{}}\overline{K}_a^{}(k)e^{ikx_{}}+_a]`$
$`\mathrm{where}\overline{\mathrm{K}}_\mathrm{a}^{}(\mathrm{k})`$ $`=`$ $`4i\pi Ar\gamma [\overline{J}_a^0(k){\displaystyle \frac{1}{r\gamma }}\overline{J}_a^1(k)]`$
$`\mathrm{and}_\mathrm{a}`$ $`=`$ $`4i\pi Ar\gamma [\overline{J}_a^0(0){\displaystyle \frac{1}{r\gamma }}\overline{J}_a^1(0)].`$ (3.53)
In above, $`\overline{J}_a^0(k)`$ and $`\overline{J}_a^1(k)`$ are the modes corresponding to $`TrT_a\dot{g}g^1`$ and $`TrT_ag^{^{}}g^1`$ respectively .
Using the above currents and the Kac-Moody algebra $`(3.46)`$ , we get that
$$[𝒫_a,𝒫_b]=if_{abc}𝒫_c$$
(3.54)
and
$$[K_a^+(p),K_b^+(k)]=if_{abc}K_c^+(p+k)+16\pi Ar\gamma p\delta _{ab}\delta _{p+k,0}.$$
(3.55)
In the same way we get from $`(3.47)`$ ,
$$[_a,_b]=if_{abc}_c$$
(3.56)
and
$$[\overline{K}_a^{}(p),\overline{K}_b^{}(k)]=if_{abc}\overline{K}_c^{}(p+k)16\pi Ar\gamma p\delta _{ab}\delta _{p+k,0}.$$
(3.57)
From $`(3.54)`$ and $`(3.56)`$ we immediately see that $`\{𝒫_a\}`$ and $`\{_a\}`$ are two representations of $`SU(l)`$ generators .
We can now compute the Hamiltonian
$$H=_0^{2\pi }𝑑\theta $$
(3.58)
in terms of the oscillation modes $`K_a^+(k)`$ , $`\overline{K}_a^{}(k)`$ and the zero modes $`𝒫_a`$ , $`_a`$ . The answer turns out to be
$$H=H_{\mathrm{osc}}+H_0$$
(3.59)
where
$$H_{\mathrm{osc}}=\frac{\pi }{8Ar\gamma }\underset{k0}{}[:K_a^+(k)K_a^+(k):+:\overline{K}_a^{}(k)\overline{K}_a^{}(k):]$$
(3.60)
and
$$H_0=\frac{\pi }{8Ar\gamma }(𝒫^2+^2).$$
(3.61)
The contribution to the Hamiltonian from the oscillatory mode has been normel ordered. $`𝒫^2`$ and $`^2`$ are simply the $`SU(l)`$ Casimirs $`𝒫^2=_a𝒫_a^2`$ and $`^2=_a_a^2`$ respectively and are given by
$`𝒫^2`$ $`=`$ $`{\displaystyle \frac{N_{adj}}{N_p}}p`$
$`^2`$ $`=`$ $`{\displaystyle \frac{N_{adj}}{N_m}}m`$ (3.62)
where $`N_{adj}`$ is the dimension of the adjoint representation of $`SU(l)`$ . $`N_p`$ and $`N_m`$ above are the dimensions of the representaions $`\{𝒫_a\}`$ and $`\{_a\}`$ respectively and $`p`$ $`(m)`$ is the index of the representations $`\{P_a\}`$ $`(\{M_a\})`$. A given zero mode will be characterized by two integers $`p`$ and $`m`$ and it will be denoted by $`|pm`$ . The state $`|pm`$ will be annihilated by $`H_{\mathrm{osc}}`$ as the latter is normal ordered. The ground state energy of the system would therefore be given by
$$E_{pm}=\frac{\pi }{8Ar\gamma }(𝒫^2+^2).$$
(3.63)
where now $`𝒫^2`$ and $`^2`$ are being understood to be equal to the numbers given by the equation $`(3.62)`$.
We want now to investigate the behaviour of the currents algebras and the Hamiltonian as $`v1`$. The currents in the equations $`(3.52)`$ $`(3.53)`$ as well as the Hamiltonians $`(3.60)`$ and $`(3.61)`$ are functions of the parameter $`v`$ and tend to become ill defined as $`v1`$ . As in Section 2.2 , we can again use a “renormalization group inspired” technique to get a well defined theory in this limit . The constant $`B`$ in this case has the allowed values given by $`\frac{n}{24\pi }`$ where $`n`$ is an integer. Furthermore $`A`$ is constrained by the condition $`A=\pm \frac{3B}{2}`$ . We will however assume that $`A`$ is a function of $`v`$, and its dependence on $`v`$ is to be determined from the condition that in the limit of $`v1`$, the ground state energy $`E_{pm}`$ becomes independent of $`v`$, i.e.
$$\frac{dE_{pm}}{dv}=0.$$
(3.64)
However by using $`(3.63)`$ it immediately follows that
$$\frac{\pi }{8r\gamma A}=\frac{1}{C}$$
(3.65)
where $`C`$ is a constant. $`A`$ in the above equation is constrained as mentioned above. It therefore cannot run continuously with $`v`$ and changes only in discrete steps always satisfying the constraint. Suppose when $`v=0`$, $`A`$ was given by $`A_0`$. The constant $`C`$ was then given by $`C=\frac{8r}{\pi }A_0`$. As $`v1`$, it follows from Eqn. (3.65) that the limiting value of $`A`$ actually tends to zero. $`C`$ however is finite in this limit and is given by the same constant value as mentioned above.
In view of the above, as $`v1`$, the ground state energy of the system is given by
$$E_{pm}=\frac{1}{C}(𝒫^2+^2)$$
(3.66)
corresponding to the Hamiltonian
$$H_0=\frac{1}{C}(𝒫^2+^2).$$
(3.67)
where now $`𝒫_a`$ and $`_a`$ are given by :
$`𝒫_a`$ $`=`$ $`{\displaystyle \frac{i\pi ^2C}{2}}J_a^0(0)`$
$`_a`$ $`=`$ $`{\displaystyle \frac{i\pi ^2C}{2}}\overline{J}_a^0(0),`$ (3.68)
and they still do satisfy $`(3.54)`$ and $`(3.56)`$ respectively . The value of the constant $`C`$ is related the the value of $`A`$ when $`v=0`$. This value is not determined by the theory and must be obtained from empirical considerations. As in the scalar field case, this system also therefore admits a one parameter family of inequivalent quantizations.
Let us now turn our attention to the oscillatory modes. As $`v1`$, $`A`$ satisfies Eqn. (3.65) and the expressions for the oscillatory modes are given by
$`K_a^+(k)`$ $`=`$ $`{\displaystyle \frac{i\pi ^2C}{2}}J_a^0(k)`$
$`\overline{K}_a^{}(k)`$ $`=`$ $`{\displaystyle \frac{i\pi ^2C}{2}}\overline{J}_a^0(k).`$ (3.69)
The current algebra satisfied by these modes are now given by
$$[K_a^+(p),K_b^+(k)]=if_{abc}K_c^+(p+k)+2\pi ^2Cp\delta _{ab}\delta _{p+k,0}$$
(3.70)
and
$$[\overline{K}_a^{}(p),\overline{K}_b^{}(k)]=if_{abc}\overline{K}_c^{}(p+k)2\pi ^2Cp\delta _{ab}\delta _{p+k,0}.$$
(3.71)
However, as $`v1`$, it is clear from Eqn. (3.51) that $`k`$ must go to zero for any value of the integer $`q`$. From Eqn. (3.60), we see that the oscillatory part of the Hamiltonian has contributions only from those modes for which $`k`$ is not equal to zero. We therefore conclude that as $`v1`$, the oscillatory modes becomes energetically unfavourable and do not contribute to the Hamiltonian. The entire theory in this limit, just as in the scalar field case, is described by a finite number of degrees of freedom given only by the zero modes.
4. Conclusion
In this paper we have investigated the limitng behaviour of quantum field theories defined on a surface $`S`$ as the latter tends to a null surface $`S_n`$.
In the case of a scalar field theory the above limiting procedure reveals several interesting features. First, as $`SS_n`$, the excitation of the oscillatory degrees of freedom of the system becomes energetically unfavourable. In this situation, the model reduces to a quantum mechanical system with the winding modes as the only degrees of freedom. Second, in the limit when $`SS_n`$, the renormalized Hamiltonian of the system contains an arbitrary parameter. Hamiltonians with different values of this parameter cannot be related via a unitary transformation. The limiting case of this system therefore admits a one-parameter family of inequivalent quantizations. Finally, this model exhibits a type of T-duality symmetry. This feature can be used to remove the quantization ambiguity only at the self-dual point.
In the case of a non-linear $`\sigma `$-model with a Wess-Zumino-Witten term a similar result is obtained. The parameters of this model are however constrained by topological considerations. However, in the limit when $`SS_n`$, the oscillatory modes of this system also have the same behaviour as in the scalar field case. The renormalized Hamiltonian is described only in terms of a finite number of degrees of freedom given by the zero modes. It is also seen to contain an arbitrary parameter that can be related to one of the constants of the theory when $`v=0`$. This observation however is not enough to fix a unique value of this parameter. We can hence say that this system also admits a one-parameter family of inequivalent quantizations.
There seems to be a degree of universality associated with the results obtained above. The suppression of the oscillatory modes in the limit of $`v1`$ can be traced to Eqn. (2.11) and (3.51) for the scalar field and the non-linear $`\sigma `$-model cases respectively. Such equations would always occur whenever there is periodicity condition on the basic variables of the theory concerned. We therefore conclude that the suppression of the oscillatory modes would be a generic phenomenon in this type of a scenario. This would in turn mean that as $`SS_n`$, the resulting theory on the null surface would generically be described by a finite number of degrees of freedom related to the zero modes of the system.
Both the models considered in this paper admits a one-parameter family of inequivalent quantizations. We have however not found any general argument supporing the universality of this phenomenon.
It would be interesting to perform similar analysis to more realistic models of physical interest, e.g. boundary field theories on a black hole horizon which is a null surface in space-time. The analysis presented in this paper could be adapted to study the dynamics of field theories on such a surface which is currently under investigation.
Acknowledgments
We would like to thank A.P.Balachandran for suggesting the problem and for his critical comments while the work was in progress. The second author would like also to thank Djamel Dou , Denjoe O’Connor and Garnik Alexanian for helpful discussions . The work of B.Y was supported in part by the DOE under contract number DE-FG02-85ER40231.
A. Appendix
Here we give some of the identities necessary to derive Eqns. (3.34) and (3.39).
To find the commutation relations among the conjugate momenta $`\mathrm{\Pi }_a`$ we proceed as follows. Consider the Jacobi identity
$$[[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})],g(\theta ^{^{\prime \prime }})]=[[\mathrm{\Pi }_b(\theta ^{^{}}),g(\theta ^{^{\prime \prime }})],\mathrm{\Pi }_a(\theta )][[g(\theta ^{^{\prime \prime }}),\mathrm{\Pi }_a(\theta )],\mathrm{\Pi }_b(\theta ^{^{}})].$$
(A.1)
Using Eqns. (3.2) and (3.22), the above Jacobi identity gives
$$[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]=if_{abc}\mathrm{\Pi }_c(\theta )\delta (\theta \theta ^{^{}})+F$$
(A.2)
where $`[F,g]=0`$ . The fact that $`[F,g]=0`$ implies that $`F`$ does not depend on $`P_i`$ but only on $`g`$ . Setting $`P_i=0`$ in Eqn. (A.2) gives $`F=0`$. This proves Eqn. (3.34).
Next we sketch the steps leading to Eqn. (3.39). First we prove the identity
$$f_{abd}N_d^c=N_a^d\frac{N_b^c}{\xi ^d}+N_b^d\frac{N_a^c}{\xi ^d}$$
(A.3)
which will be used in the proof of the commutator . Using the definition (3.31) of $`\mathrm{\Pi }_a(\theta )`$ we get
$`[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]`$ $`=`$ $`[N_a^c(\theta )P_c(\theta ),N_b^d(\theta ^{^{}})P_d(\theta ^{^{}})]`$
$`=`$ $`N_a^c(\theta )[P_c(\theta ),N_b^d(\theta ^{^{}})]P_d(\theta ^{^{}})+N_b^d(\theta ^{^{}})[N_a^c(\theta ),P_d(\theta ^{^{}})]P_c(\theta ).`$
From Eqn. (A.2) we get,
$`[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]`$ $`=`$ $`if_{abc}\delta (\theta \theta ^{^{}})\mathrm{\Pi }_c(\theta )`$
Using Eqns. (A.4), (A.5) and (3.27), Eqn. (A.3) follows easily.
We are now ready to compute $`[J_a^0(\theta ),J_b^0(\theta ^{^{}}]`$ . From Eqn. (3.31) we get
$`[{\displaystyle \frac{2iAR}{\gamma }}J_a^0(\theta ),{\displaystyle \frac{2iAR}{\gamma }}J_b^0(\theta ^{^{}})]`$ $`=`$ $`[\mathrm{\Pi }_a(\theta ),\mathrm{\Pi }_b(\theta ^{^{}})]`$ (A.6)
$`+`$ $`[\mathrm{\Pi }_a(\theta ),N_b^i(\theta ^{^{}})\omega _{ij}(\theta ^{^{}})_\theta ^{^{}}\xi ^j]+[N_a^i(\theta )\omega _{ij}(\theta )_\theta \xi ^j,\mathrm{\Pi }_b(\theta ^{^{}}]`$
$`+`$ $`[N_a^i(\theta )\omega _{ij}(\theta )_\theta \xi ^j,N_b^i(\theta ^{^{}})\omega _{ij}(\theta ^{^{}})_\theta ^{^{}}\xi ^j].`$
The last commutator is zero as it has no time derivative . The first commutator is given by $`(A.5)`$ . The third commutator can be obtained from the second by interchanging $`a`$ with $`b`$ and $`\theta `$ with $`\theta ^{^{}}`$ then putting an overall minus sign . Let us then compute the second comutator
$`[\mathrm{\Pi }_a(\theta ),N_b^i(\theta ^{^{}})\omega _{ij}(\theta ^{^{}})_\theta ^{^{}}\xi ^j]`$ $`=`$ $`[\mathrm{\Pi }_a(\theta ),N_b^i(\theta ^{^{}})]\omega _{ij}(\theta ^{^{}})_\theta ^{^{}}\xi ^j+N_b^i(\theta ^{^{}})\omega _{ij}(\theta ^{^{}})[\mathrm{\Pi }_a(\theta ),_\theta ^{^{}}\xi ^j]`$ (A.7)
$`+`$ $`N_b^i(\theta ^{^{}})[\mathrm{\Pi }_a(\theta ),\omega _{ij}(\theta ^{^{}})]_\theta ^{^{}}\xi ^j]`$
$`=`$ $`i\delta (\theta \theta ^{^{}})N_a^c{\displaystyle \frac{N_b^i}{\xi ^c}}\omega _{ij}_\theta \xi ^ji_\theta ^{^{}}\delta (\theta \theta ^{^{}})N_b^i(\theta ^{^{}})N_a^j(\theta )\omega _{ij}(\theta ^{^{}})`$
$``$ $`i\delta (\theta \theta ^{^{}})N_b^iN_a^c{\displaystyle \frac{\omega _{ij}}{\xi ^c}}_\theta \xi ^j`$
where we have made use of Eqns. $`(3.27)`$ and $`(3.31)`$ . The sum of the second and the third commutator in Eqn. $`(A.6)`$ is then given by
$`2+3`$ $`=`$ $`a+b+c`$
$`\mathrm{where}\mathrm{a}`$ $`=`$ $`i\delta (\theta \theta ^{^{}})[N_a^c{\displaystyle \frac{N_b^i}{\xi ^c}}N_b^c{\displaystyle \frac{N_a^i}{\xi ^c}}]\omega _{ij}_\theta \xi ^j`$
$`b`$ $`=`$ $`i_\theta ^{^{}}\delta (\theta \theta ^{^{}})N_b^i(\theta ^{^{}})N_a^j(\theta )\omega _{ij}(\theta ^{^{}})+i_\theta \delta (\theta \theta ^{^{}})N_a^i(\theta )N_b^j(\theta ^{^{}})\omega _{ij}(\theta )`$
$`c`$ $`=`$ $`i\delta (\theta \theta ^{^{}})[N_b^iN_a^cN_a^iN_b^c]{\displaystyle \frac{\omega _{ij}}{\xi ^c}}_\theta \xi ^j.`$ (A.8)
By using $`(A.3)`$ we find that
$$a=i\delta (\theta \theta ^{^{}})f_{abc}N_c^i\omega _{ij}_\theta \xi ^j$$
(A.9)
Careful manipulations with $`b`$ will give
$`b`$ $`=`$ $`i_\theta ^{^{}}\delta (\theta \theta ^{^{}})N_b^i(\theta ^{^{}})N_a^j(\theta )\omega _{ij}(\theta ^{^{}})+i_\theta \delta (\theta \theta ^{^{}})N_a^i(\theta )N_b^j(\theta ^{^{}})\omega _{ij}(\theta )`$ (A.10)
$`=`$ $`iN_a^j(\theta )N_b^i(\theta ^{^{}})[_\theta ^{^{}}\delta (\theta \theta ^{^{}})\omega _{ij}(\theta ^{^{}})+_\theta \delta (\theta \theta ^{^{}})\omega _{ij}(\theta )]`$
$`=`$ $`i\delta (\theta \theta ^{^{}})N_a^j(\theta )N_b^i(\theta ^{^{}})_\theta \omega _{ij}`$
$`=`$ $`i\delta (\theta \theta ^{^{}})N_a^cN_b^i{\displaystyle \frac{\omega _{ic}}{\xi ^j}}_\theta \xi ^j.`$
Finally $`c`$ can be rewitten as
$$c=i\delta (\theta \theta ^{^{}})N_a^cN_b^i[\frac{\omega _{cj}}{\xi ^i}+\frac{\omega _{ji}}{\xi ^c}]_\theta \xi ^j.$$
(A.11)
Putting Eqns. $`(A.9)`$ , $`(A.10)`$ and $`(A.11)`$ together and using Eqns. $`(3.19)`$ and $`(3.20)`$, we get
$`2+3`$ $`=`$ $`i\delta (\theta \theta ^{^{}})f_{abc}N_c^i\omega _{ij}_\theta \xi ^j+i\delta (\theta \theta ^{^{}})N_a^cN_b^i[{\displaystyle \frac{\omega _{ic}}{\xi ^j}}+{\displaystyle \frac{\omega _{cj}}{\xi ^i}}+{\displaystyle \frac{\omega _{ji}}{\xi ^c}}]_\theta \xi ^j`$ (A.12)
$`=`$ $`i\delta (\theta \theta ^{^{}})f_{abc}N_c^i\omega _{ij}_\theta \xi ^j+3iB\delta (\theta \theta ^{^{}})N_a^cN_b^iTr[g^1{\displaystyle \frac{g}{\xi ^i}},g^1{\displaystyle \frac{g}{\xi ^c}}]g^1g^{^{}}`$
$`=`$ $`i\delta (\theta \theta ^{^{}})f_{abc}N_c^i\omega _{ij}_\theta \xi ^j+\delta (\theta \theta ^{^{}})3r^2Bf_{abc}J_a^1.`$
Using Eqns. $`(A.12)`$ and $`(A.5)`$ in Eqn. $`(A.6)`$ will lead to the commutation relations
$$[J_a^0(\theta ),J_b^0(\theta ^{^{}})]=\frac{\gamma }{2AR}f_{abc}J_c^0(\theta )\delta (\theta \theta ^{^{}})\frac{3B\gamma ^2}{4A^2}\delta (\theta \theta ^{^{}})f_{abc}J_c^1(\theta ).$$
(A.13)
|
warning/0002/math-ph0002025.html
|
ar5iv
|
text
|
# Integrating factors for second order ODEs
## 1 Introduction
Although in principle it is always possible to determine whether a given ODE is exact (a total derivative), there is no known method which is always successful in making arbitrary ODEs exact. For $`\mathrm{n}^{\mathrm{th}}`$order ODEs - as in the case of symmetries - integrating factors ($`\mu `$) are determined as solutions of an $`\mathrm{n}^{\mathrm{th}}`$order linear PDE in $`n`$+1 variables, and to solve this determining PDE is a major problem in itself.
Despite the fact that the determining PDE for $`\mu `$ naturally splits into a PDE system, the problem is - as a whole - too general, and to solve it a restriction of the problem in the form of a more concrete ansatz for $`\mu `$ is required. For example, in a recent work by the authors explore possible ansatzes depending on the given ODE, which are useful when this ODE has known symmetries of certain type. In another work, explores the use of computer algebra to try various ansatzes for $`\mu `$, no matter the ODE input, but successively increasing the order of the derivatives (up to the $`\mathrm{n}^{\mathrm{th}}`$$`1`$ order) on which $`\mu `$ depends; the idea is to try to maximize the splitting so as to increase the chances of solving the resulting PDE system by first simplifying it using differential Groebner basis techniques.
Bearing this in mind, this paper presents a method, for second order explicit ODEs<sup>1</sup><sup>1</sup>1We say that a second order ODE is in explicit form when it appears as $`y^{\prime \prime }\mathrm{\Phi }(x,y,y^{})=0`$. Also, we exclude from the discussion the case of a linear ODE and an integrating factor of the form $`\mu (x)`$, already known to be the solution to the adjoint ODE., which systematically determines the existence and the explicit form of integrating factors when they depend on only two variables, that is: when they are of the form $`\mu (x,y)`$, $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$. The approach works without solving any auxiliary differential equations - except for a linear ODE in one subcase of the $`\mu (x,y)`$ problem - and is based on the use of the forms of the ODE families admitting such integrating factors. It turns out that with this restriction - $`\mu `$ depends on only two variables - the use of differential Groebner basis techniques is not necessary; these integrating factors, when they exist, can be given directly by identifying the input ODE as a member of one of various related ODE families.
The exposition is organized as follows. In sec. 2, the standard formulation of the determination of integrating factors is briefly reviewed and the method we used for obtaining the aforementioned integrating factors $`\mu (x,y)`$, $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$ is presented. In sec. 3, some aspects of the integrating factor and symmetry approaches are discussed, and their complementariness is illustrated with two ODE families not having point symmetries. Sec. 4 contains some statistics concerning the new solving method and the second order non-linear ODEs found in Kamke’s book, as well as a comparison of performances of some popular computer algebra packages in solving a related subset of these ODEs. Finally, the conclusions contain some general remarks about the work.
Aside from this, in the Appendix, a table containing extra information concerning integrating factors for some of Kamke’s ODEs is presented.
## 2 Integrating Factors and ODE patterns
In this paper we use the term “integrating factor” in connection with the explicit form of an $`\mathrm{n}^{\mathrm{th}}`$order ODE
$$y^{(n)}\mathrm{\Phi }(x,y,y^{},\mathrm{},y^{(n1)})=0$$
(1)
so that $`\mu (x,y,y^{},\mathrm{},y^{(n1)})`$ is an integrating factor if
$$\mu \left(y^{(n)}\mathrm{\Phi }\right)=\frac{d}{dx}R(x,y,y^{},\mathrm{},y^{(n1)})$$
(2)
for some function $`R`$. The knowledge of $`\mu `$ is - in principle - enough to determine $`R`$ by using standard formulas (see for instance Murphy’s book). To determine $`\mu `$, one can try to solve for it in the exactness condition, obtained applying Euler’s operator to the total derivative $`H\mu \left(y^{(n)}\mathrm{\Phi }\right)`$:
$$\frac{H}{y}\frac{d}{dx}\left(\frac{H}{y^{}}\right)+\frac{d^2}{dx^2}\left(\frac{H}{y^{\prime \prime }}\right)+\mathrm{}+(1)^n\frac{d^n}{dx^n}\left(\frac{H}{y^{(n)}}\right)=0$$
(3)
Eq.(3) is of the form
$$A(x,y,y^{},\mathrm{},y^{(2n3)})+y^{(2n2)}B(x,y,y^{},\mathrm{},y^{(n1)})=0$$
(4)
where $`A`$ is of degree $`n1`$ in $`y^{(n)}`$ and linear in $`y^{(k)}`$ for $`n<k(2n3)`$, so that Eq.(4) can be split into a PDE system for $`\mu `$. In the case of second order ODEs - the subject of this work - Eq.(3) is of the form
$$A(x,y,y^{})+y^{\prime \prime }B(x,y,y^{})=0$$
(5)
and the PDE system is obtained by taking $`A`$ and $`B`$ equal to zero<sup>2</sup><sup>2</sup>2In a recent work by , the authors arrive at Eq.(5) and Eq.(7) departing from the adjoint linearized system corresponding to a given ODE; the possible splitting of Eq.(4) into an overdetermined system for $`\mu `$ is also mentioned. However, in that work, $`y^{\prime \prime }`$ of Eq.(5) above appears replaced by $`\mathrm{\Phi }(x,y,y^{})`$, and the authors discuss possible alternatives to tackle Eqs.(5) and (7) instead of Eqs.(6) and (7).:
$`A`$ $``$ $`\left(y^{}\mu _{y^{}y}\mu _y+\mu _{y^{}x}\right)\mathrm{\Phi }+\left(\mathrm{\Phi }_{y^{}x}+y^{}\mathrm{\Phi }_{y^{}y}\mathrm{\Phi }_y\right)\mu +y_{}^{}{}_{}{}^{2}\mu _{yy}`$ (6)
$`+\left(\mu _y\mathrm{\Phi }_y^{}+\mu _y^{}\mathrm{\Phi }_y+2\mu _{xy}\right)y^{}+\mu _y^{}\mathrm{\Phi }_x+\mu _x\mathrm{\Phi }_y^{}+\mu _{xx}=0`$
$`B`$ $``$ $`y^{}\mu _{y^{}y}+\mathrm{\Phi }\mu _{y^{}y^{}}+\mu \mathrm{\Phi }_{y^{}y^{}}+2\mu _y+2\mu _y^{}\mathrm{\Phi }_y^{}+\mu _{y^{}x}=0`$ (7)
Regarding the solvability of these equations, unless a more concrete ansatz for $`\mu (x,y,y^{})`$ is given, the problem is in principle as difficult as solving the original ODE. We then studied the solution for $`\mu `$ of Eqs.(6) and (7) when $`\mu `$ depends only on two variables, that is: for $`\mu (x,y)`$, $`\mu (x,y^{})`$ and $`\mu (y,y^{})`$. Concretely, we searched for the existence conditions for such integrating factors, expressed as a set of equations in $`\mathrm{\Phi }`$, plus an algebraic expression for $`\mu `$ as a function of $`\mathrm{\Phi }`$, valid when the existence conditions hold. Formulating the problem in that manner and taking into account the integrability conditions of the system, Eqs.(6) and (7) turned out to be solvable for $`\mu (x,y)`$, but appeared to us untractable when $`\mu `$ depends on two variables one of which is $`y^{}`$.
We then considered a different approach, taking into account from the beginning the form of the ODE family admitting a given integrating factor. As shown in the following sections, it turns out that, using that piece of information (Eq.(11) below), when $`\mu `$ depends only on two variables the existence conditions and the integrating factors themselves can be systematically determined; and in the cases $`\mu (x,y^{})`$ and $`\mu (y,y^{})`$, this can be done without solving any differential equations.
Concerning the ODE families admitting given integrating factors, we note that, from Eq.(2)
$$\mu (x,y,y^{},\mathrm{},y^{(n1)})=\frac{R\text{ }}{y^{(n1)}}$$
(8)
and hence the first integral $`R`$ is of the form
$$R=G(x,y,\mathrm{},y^{(n2)})+\mu 𝑑y^{(n1)}$$
(9)
for some function $`G`$. In turn, since $`R`$ is a first integral, it satisfies
$$R_x+y^{}R_y+\mathrm{}+\mathrm{\Phi }R_{y^{(n1)}}=0$$
(10)
Inserting Eq.(9) into the above and solving for $`y^{(n)}`$ leads to the general form of an ODE admitting a given integrating factor:
$$y^{(n)}=\frac{1}{\mu }\left[\frac{}{x}\left(\mu 𝑑y^{(n1)}+G\right)+\mathrm{}+y^{(n1)}\frac{}{y^{(n2)}}\left(\mu 𝑑y^{(n1)}+G\right)\right]$$
(11)
### 2.1 Second order ODEs and integrating factors of the form $`\mu (x,y)`$
We consider first the determination of integrating factors of the form $`\mu (x,y)`$, which turns out to be straightforward<sup>3</sup><sup>3</sup>3The result for Case A presented in this subsection is also presented as lemma 3.8 in .. The determining equations (6) and (7) for this case are given by:
$$\begin{array}{c}y_{}^{}{}_{}{}^{2}\mu _{yy}+2\mu _{xy}y^{}+\mu \mathrm{\Phi }_{y^{}x}+\mu \mathrm{\Phi }_{y^{}y}y^{}\mu \mathrm{\Phi }_y\mu _y\mathrm{\Phi }+\mu _yy^{}\mathrm{\Phi }_y^{}+\mu _{xx}+\mu _x\mathrm{\Phi }_y^{}=0\\ \mu \mathrm{\Phi }_{y^{}y^{}}+2\mu _y=0\end{array}$$
(12)
Although the use of integrability conditions is enough to tackle this problem, the solving of Eqs.(12) can be directly simplified if we take into account the ODE family admitting an integrating factor $`\mu (x,y)`$. From Eq.(11), that ODE family takes the form
$$y^{\prime \prime }=a(x,y)y_{}^{}{}_{}{}^{2}+b(x,y)y^{}+c(x,y),$$
(13)
where
$$a(x,y)=\frac{\mu _y}{\mu },b(x,y)=\frac{G_y+\mu _x}{\mu },c(x,y)=\frac{G_x}{\mu }$$
(14)
and $`G(x,y)`$ is an arbitrary function of its arguments. Hence, as a shortcut to solving Eqs.(12), one can take Eq.(13) as an existence condition - $`\mathrm{\Phi }`$ must be a polynomial of degree two in $`y^{}`$ \- and directly solve Eqs.(14) for $`\mu `$. The calculations are straightforward; there are two different cases.
Case A: $`2a_xb_y0`$
Defining the two auxiliary quantities
$$\phi c_yacb_x,\mathrm{{\rm Y}}a_{xx}+a_xb+\phi _y$$
(15)
an integrating factor of the form $`\mu (x,y)`$ exists only when
$$\mathrm{{\rm Y}}_ya_x=0,\mathrm{{\rm Y}}_x+\phi +b\mathrm{{\rm Y}}\mathrm{{\rm Y}}^2=0$$
(16)
and is then given in solved form, in terms of $`a`$, $`b`$ and $`c`$ by
$$\mu (x,y)=\mathrm{exp}\left(\left(\mathrm{{\rm Y}}+\frac{}{x}a𝑑y\right)𝑑xa𝑑y\right)$$
(17)
So, in this case, when an integrating factor of this type exists there is only one<sup>4</sup><sup>4</sup>4We recall that if $`\mu `$ is an integrating factor leading to a first integral $`\psi `$, then the product $`\mu F(\psi )`$ \- where $`F`$ is an arbitrary function - is also an integrating factor, which however does not lead to a first integral independent of $`\psi `$. and it can be determined without solving any differential equations.
Case B: $`2a_xb_y=0`$
Redefining $`\phi c_yac`$, an integrating factor of the form $`\mu (x,y)`$ exists only when
$$a_{xx}a_xb\phi _y=0,$$
(18)
and then $`\mu (x,y)`$ is given by
$$\mu (x,y)=\nu (x)\mathrm{e}^{^{{\displaystyle a𝑑y}}}$$
(19)
where $`\nu (x)`$ is either one of the independent solutions of the second order linear ODE<sup>5</sup><sup>5</sup>5When the given ODE is linear, Eq.(20) is just the corresponding adjoint equation.
$$\nu ^{\prime \prime }=A(x)\nu ^{}+B(x)\nu ,$$
(20)
and
$$A(x)2b,B(x)\phi +\left(\frac{}{x}\right)\left(b\right),\frac{}{x}a𝑑y$$
(21)
So in this case, to transform Eq.(19) into an explicit expression for $`\mu `$ we first need to solve a second order linear ODE. When the attempt to solve Eq.(20) is successful, using each of its two independent solutions as integrating factors leads to the general solution of Eq.(13), instead of just a reduction of order.
### 2.2 Second order ODEs and integrating factors of the form $`\mu (x,y^{})`$
When the integrating factor is of the form $`\mu (x,y^{})`$, the determining equations (6) and (7) become
$$\begin{array}{c}\left(\mathrm{\Phi }_yy^{}+\mathrm{\Phi }_x\right)\mu _y^{}+\left(\mathrm{\Phi }_y+\mathrm{\Phi }_{y^{}x}+\mathrm{\Phi }_{y^{}y}y^{}\right)\mu +\mu _{xx}+\mu _x\mathrm{\Phi }_y^{}+\mu _{y^{}x}\mathrm{\Phi }=0\\ \mathrm{\Phi }\mu _{y^{}y^{}}+\mu \mathrm{\Phi }_{y^{}y^{}}+2\mu _y^{}\mathrm{\Phi }_y^{}+\mu _{y^{}x}=0\end{array}$$
(22)
As in the case $`\mu (x,y)`$, the solution we are interested in is an expression for $`\mu (x,y^{})`$ in terms of $`\mathrm{\Phi }`$, as well as existence conditions for such an integrating factor expressed as equations in $`\mathrm{\Phi }`$. However, differently than the case $`\mu (x,y)`$, we didn’t find a way to solve the $`\mu (x,y^{})`$ problem just using integrability conditions, neither working by hand nor using the specialized computer algebra packages diffalg and standard\_form . We then considered approaching the problem as explained in the previous subsection, departing from the form of Eq.(9) for $`\mu =\mu (x,y^{})`$:
$$y^{\prime \prime }=\mathrm{\Phi }(x,y,y^{})\frac{F_x+G_x+G_yy^{}}{F_y^{}}$$
(23)
where $`G(x,y)`$ and $`F(x,y^{})`$ are arbitrary functions of their arguments and
$$\mu (x,y^{})=F_y^{}$$
(24)
Now, Eq.(23) is not polynomial in either $`x`$, $`y`$ or $`y^{}`$, and hence its use to simplify and solve the problem is less straightforward than in the case $`\mu (x,y)`$. However, in Eq.(23), all the dependence on $`y`$ comes from $`G(x,y)`$ in the numerator, and as it is shown below, this fact is a key to solving the problem. Considering ODEs for which $`\mathrm{\Phi }_y0`$<sup>6</sup><sup>6</sup>6ODEs missing y may also have integrating factors of the form $`\mu (x,y^{})`$. Such an ODE however can always be reduced to first order by a change of variables, so that the determination of a $`\mu (x,y^{})`$ for it is equivalent to solving a first order ODE problem - not the focus of this work., the approach we used can be summarized in the following three lemmas whose proofs are developed separately for convenience.
Lemma 1. For all linear ODEs of the family Eq.(23), an integrating factor of the form $`\mu (x,y^{})`$ such that $`\mu _y^{}0`$, when it exists, can be determined directly from the coefficient of $`y`$ in the input ODE.
Lemma 2. For all non-linear ODEs of Eq.(23), the knowledge of $`\mu (x,y^{})`$ up to a factor depending on $`x`$, that is, of $`(x,y^{})`$ satisfying
$$(x,y^{})=\frac{\mu (x,y^{})}{\stackrel{}{\mu }(x)}$$
(25)
is enough to determine $`\stackrel{}{\mu }(x)`$ by means of an integral.
Lemma 3. For all non-linear ODEs members of Eq.(23), it is always possible to determine a function $`(x,y^{})`$ satisfying Eq.(25).
Corollary. For all second order ODEs such that $`\mathrm{\Phi }_y0`$, the determination of $`\mu (x,y^{})`$ (if the ODE is linear we assume $`\mu _y^{}0`$), when it exists, can be performed systematically and without solving any differential equations.
#### 2.2.1 Proof of Lemma 1
For Eq.(23) to be linear and not missing $`y`$, either $`G_x`$ or $`G_y`$ must be linear in $`y`$. Both $`G_x`$ and $`G_y`$ cannot simultaneously be linear in $`y`$ since, in such a case, $`G_x/F_y^{}`$ or $`y^{}G_y/F_y^{}`$ would be non-linear in $`\{y,y^{}\}`$<sup>7</sup><sup>7</sup>7We are only interested in the case $`\mu _y^{}=F_{y^{}y^{}}0`$.; therefore, either $`G_{yy}=0`$ or $`G_{xy}=0`$.
Case A: $`G_y`$ is linear in $`y`$ and $`G_{xy}=0`$
Hence, $`G`$ is given by
$$G=C_2y^2+C_1y+g(x)$$
(26)
where $`g(x)`$ is arbitrary. From Eq.(23), in order to have $`y^{}G_y/F_y^{}`$ linear in $`\{y,y^{}\}`$, $`F_y^{}`$ must of the form $`\nu (x)y^{}`$ for some function $`\nu (x)`$. Also, $`F_x/F_y^{}`$ can have a term linear in $`y^{}`$, and a term proportional to $`1/y^{}`$ to cancel with the one coming from $`G_x/F_y^{}=g^{}/F_y^{}`$, so that
$$F_y^{}=\nu y^{},F_x=\frac{\nu ^{}y_{}^{}{}_{}{}^{2}}{2}g^{}$$
(27)
where the coefficient $`\nu ^{}/2`$ in the second equation above arises from the integrability conditions between both equations. Eq.(23) is then of the form
$$y^{\prime \prime }=\frac{\nu ^{}}{2\nu }y^{}\frac{2C_2}{\nu }y\frac{C_1}{\nu }$$
(28)
and hence, a linear ODE $`y^{\prime \prime }=a(x)y^{}+b(x)y`$ has an integrating factor $`\mu (x,y^{})=y^{}/b`$ when $`b^{}/b2a=0`$. $`\mathrm{}`$
Case B: $`G_x`$ is linear in $`y`$ and $`G_{yy}=0`$
In this case, in order to have Eq.(23) linear, $`F_y^{}`$ cannot depend on $`y^{}`$, so that the integrating factor is of the form $`\mu (x)`$ and hence the case is of no interest: we end up with the standard search for $`\mu (x)`$ as the solution to the adjoint of the original linear ODE. $`\mathrm{}`$
#### 2.2.2 Proof of Lemma 2
It follows from Eqs.(23) and (24) that, given $``$ satisfying (25),
$$\frac{}{y}\left(\mathrm{\Phi }(x,y,y^{})(x,y^{})\right)=\frac{G_{yx}(x,y)+G_{yy}(x,y)y^{}}{\stackrel{}{\mu }(x)}$$
(29)
Hence, by taking coefficients of $`y^{}`$ in the above,
$`\phi _1`$ $``$ $`\mathrm{\Phi }_y(x,y,y^{})(x,y^{})y^{}{\displaystyle \frac{}{y^{}}}\left(\mathrm{\Phi }_y(x,y,y^{})(x,y^{})\right)={\displaystyle \frac{G_{yx}(x,y)}{\stackrel{}{\mu }(x)}}`$
$`\phi _2`$ $``$ $`{\displaystyle \frac{}{y^{}}}\left(\mathrm{\Phi }_y(x,y,y^{})(x,y^{})\right)={\displaystyle \frac{G_{yy}(x,y)}{\stackrel{}{\mu }(x)}}`$ (30)
where the left-hand-sides can be calculated explicitly since they depend only on $`\mathrm{\Phi }`$ and the given $``$. Similarly,
$`\phi _3`$ $``$ $`{\displaystyle \frac{}{y^{}}}\left(\mathrm{\Phi }(x,y,y^{})(x,y^{})\right)={\displaystyle \frac{F_{y^{}x}(x,y^{})+G_y(x,y)}{\stackrel{}{\mu }(x)}}`$
$`\phi _4`$ $``$ $`{\displaystyle \frac{}{y^{}}}(x,y^{})={\displaystyle \frac{F_{y^{}y^{}}(x,y^{})}{\stackrel{}{\mu }(x)}}`$ (31)
Now, since in this case the ODE family Eq.(23) is nonlinear by hypothesis, either $`\phi _2`$ or $`\phi _4`$ is different from zero, so that at least one of the pairs $`\{\phi _1,\phi _2\}`$ or $`\{\phi _3,\phi _4\}`$ can be used to determine $`\stackrel{}{\mu }(x)`$ as the solution of a first order linear ODE. For example, if $`\phi _20`$,
$$\frac{}{y}\left(\phi _1(x,y)\stackrel{}{\mu }(x)\right)=\frac{}{x}\left(\phi _2(x,y)\stackrel{}{\mu }(x)\right)$$
(32)
from where
$$\stackrel{}{\mu }(x)=e^{^{{\displaystyle \frac{1}{\phi _2}\left(\frac{\phi _1}{y}\frac{\phi _2}{x}\right)𝑑x}}}$$
(33)
If $`\phi _2=0`$ then $`\phi _40`$ and we obtain
$$\stackrel{}{\mu }(x)=e^{^{{\displaystyle \frac{1}{\phi _4}\left(\frac{\phi _3}{y^{}}\frac{\phi _4}{x}\right)𝑑x}}}$$
(34)
When combined with Eq.(25), Eqs.(33) and (34) alternatively give both an explicit solution to the problem and an existence condition, since a solution $`\stackrel{}{\mu }(x)`$ \- and hence an integrating factor of the form $`\mu (x,y^{})`$ \- exists if the integrand in Eq.(33) or Eq.(34) only depends on $`x`$. $`\mathrm{}`$
#### 2.2.3 Proof of Lemma 3
We start from Eq.(23) by considering the expression
$$\mathrm{{\rm Y}}\mathrm{\Phi }_y=\frac{G_{xy}(x,y)+G_{yy}(x,y)y^{}}{F_y^{}(x,y^{})}$$
(35)
and develop the proof below splitting the problem into different cases. For each case we show how to find $`(x,y^{})`$ satisfying Eq.(25). $``$ will then lead to the required integrating factor when, in addition to the conditions explained below, the existence conditions for $`\stackrel{}{\mu }(x)`$ mentioned in the previous subsection are satisfied.
Case A: $`G_{xy}/G_{yy}`$ depends on $`y`$
To determine whether this is the case, we cannot just analyze the ratio $`G_{xy}/G_{yy}`$ itself since it is unknown. However, from Eq.(35), in this case the factors of $`\mathrm{{\rm Y}}`$ depending on $`y`$ will also depend on $`y^{}`$, and this condition can be formulated as
$$\frac{}{y^{}}\left(\frac{\mathrm{{\rm Y}}_y}{\mathrm{{\rm Y}}}\right)0$$
(36)
When this inequation holds, we determine $`F_y^{}(x,y^{})`$ up to a factor depending on $`x`$, that is, the required $`(x,y^{})`$, as the reciprocal of the factors of $`\mathrm{{\rm Y}}`$ which depend on $`y^{}`$ but not $`y`$. $`\mathrm{}`$
Example: Kamke’s ODE 226
This ODE is presented in Kamke’s book already in exact form, so we start by rewriting it in explicit form as
$$y^{\prime \prime }=\frac{x^2yy^{}+xy^2}{y^{}}$$
(37)
We determine $`\mathrm{{\rm Y}}`$ (Eq.(35)) as
$$\mathrm{{\rm Y}}=\frac{x(xy^{}+2y)}{y^{}}$$
(38)
The only factor of $`\mathrm{{\rm Y}}`$ containing $`y`$ is:
$$xy^{}+2y$$
(39)
and since this also depends on $`y^{}`$, $`(x,y^{})`$ is given by
$$(x,y^{})=y^{}$$
(40)
Case B: either $`G_{xy}=0`$ or $`G_{yy}=0`$
When the expression formed by all the factors of $`\mathrm{{\rm Y}}`$ containing $`y`$ does not contain $`y^{}`$, in Eq.(36) we will have $`\frac{}{y^{}}(\frac{\mathrm{{\rm Y}}_y}{\mathrm{{\rm Y}}})=0`$, and it is impossible to determine a priori whether one of the functions $`\{G_{xy},G_{yy}\}`$ is zero, or alternatively their ratio does not depend on $`y`$. We then proceed by assuming the former, build an expression for $`(x,y^{})`$ as in Case A, and check for the existence of $`\stackrel{}{\mu }(x)`$ as explained in the previous subsection. If $`\stackrel{}{\mu }(x)`$ exists, the problem is solved; otherwise we proceed as follows.
Case C: $`G_{xy}/G_{yy}0`$ and does not depend on $`y`$
In this case, neither $`G_{xy}`$ nor $`G_{yy}`$ is zero and their ratio is a function of just $`x`$, so that
$`G_{xy}`$ $`=`$ $`v_1(x)w(x,y)`$
$`G_{yy}`$ $`=`$ $`v_2(x)w(x,y)`$ (41)
for some unknown functions $`v_1(x)`$ and $`v_2(x)`$. Eq.(35) is then given by
$$\mathrm{{\rm Y}}=w(x,y)\frac{\left(v_1(x)+v_2(x)y^{}\right)}{F_y^{}(x,y^{})}$$
(42)
for some function $`w(x,y)`$, which is made up of the factors of $`\mathrm{{\rm Y}}`$ depending on $`y`$ and not on $`y^{}`$. To determine $`F_y^{}(x,y^{})`$ up to a factor depending on $`x`$, we need to determine the ratio $`v_1(x)/v_2(x)`$. For this purpose, from Eq.(41) we build a PDE for $`G_y(x,y)`$,
$$G_{xy}=\frac{v_1(x)}{v_2(x)}G_{y,y}$$
(43)
The general solution of Eq.(43) is
$$G_y=𝒢\left(y+p(x)\right)$$
(44)
where $`𝒢`$ is an arbitrary function of its argument and for convenience we introduced
$$p^{}(x)v_1(x)/v_2(x)$$
(45)
We now determine $`p^{}(x)`$ as follows. Taking into account Eq.(41),
$$v_2(x)w(x,y)=𝒢^{}(y+p(x))$$
(46)
By taking the ratio between this expression and its derivative w.r.t $`y`$ we obtain
$$(y+p(x))\frac{\mathrm{ln}(w)}{y}=\frac{𝒢^{\prime \prime }(y+p(x))}{𝒢^{}(y+p(x))}$$
(47)
that is, a function of $`y+p(x)`$ only, which we can determine since we know $`w(x,y)`$. If $`^{}0`$, $`p^{}(x)`$ is given by
$$p^{}(x)=\frac{_x}{_y}=\frac{w_{xy}ww_xw_y}{w_{yy}ww_{y}^{}{}_{}{}^{2}}$$
(48)
In summary, the conditions for this case are
$$\mathrm{{\rm Y}}_y0,\frac{}{y^{}}\left(\frac{\mathrm{{\rm Y}}_y}{\mathrm{{\rm Y}}}\right)=0,\frac{^2}{yx}\mathrm{ln}(w)0,\frac{^2}{yy}\mathrm{ln}(w)0$$
(49)
and then, from Eq.(42), $`(x,y^{})`$ is given by
$$(x,y^{})=\frac{(p^{}+y^{})w}{\mathrm{{\rm Y}}}$$
(50)
where at this point $`\mathrm{{\rm Y}}`$, $`w(x,y)`$ and $`p^{}(x)`$ are all known. $`\mathrm{}`$
Example: Kamke’s ODE 136.
We begin by writing the ODE in explicit form as
$$y^{\prime \prime }=\frac{h(y^{})}{xy}$$
(51)
This example is interesting since the standard search for point symmetries is made difficult by the presence of an arbitrary function of $`y^{}`$. $`\mathrm{{\rm Y}}`$ (Eq.(35)) is determined as
$$\mathrm{{\rm Y}}=\frac{h(y^{})}{(xy)^2}$$
(52)
and $`w(x,y)`$ as
$$w(x,y)=\frac{1}{(xy)^2}$$
(53)
Then $`(y+p(x))`$ (Eq.(47)) becomes
$$=\frac{2}{xy}$$
(54)
and hence, from Eq.(48), $`p^{}(x)`$ is
$$p^{}(x)=1$$
(55)
so from Eq.(50):
$$(x,y^{})=\frac{1y^{}}{h(y^{})}$$
(56)
Case D: $`=0`$
We now discuss how to obtain $`p^{}(x)`$ when $`^{}(y+p(x))=0`$. We consider first the case in which $`=0`$. Then, $`𝒢^{\prime \prime }=0`$ and the condition for this case is
$$\mathrm{{\rm Y}}_y=0$$
(57)
Recalling Eq.(44), $`G`$ is given by
$$G(x,y)=C_1(y+p(x))^2+C_2(y+p(x))+g(x)$$
(58)
for some function $`g(x)`$ and some constants $`C_1`$, $`C_2`$. From Eq.(23), $`\mathrm{\Phi }(x,y,y^{})`$ takes the form
$$\mathrm{\Phi }(x,y,y^{})=\frac{F_x(x,y^{})+g^{}(x)+(2C_1(y+p(x))+C_2)(y^{}+p^{}(x))}{F_y^{}(x,y^{})}$$
(59)
We now determine $`p^{}(x)`$ as follows. First, from the knowledge of $`\mathrm{{\rm Y}}`$ and $`\mathrm{\Phi }`$ we build the two explicit expressions:
$$\mathrm{\Lambda }\frac{1}{\mathrm{{\rm Y}}}=\frac{F_y^{}}{2C_1(y^{}+p^{}(x))}$$
(60)
and
$$\mathrm{\Psi }\frac{\mathrm{\Phi }(x,y,y^{})}{\mathrm{{\rm Y}}}y=\frac{F_x+g^{}(x)}{2C_1(y^{}+p^{}(x))}+p(x)+\frac{C_2}{2C_1}$$
(61)
From Eq.(60) and Eq.(61) $`\mathrm{\Lambda }`$ and $`\mathrm{\Psi }`$ are related by:
$$\frac{}{x}\left((y^{}+p^{}(x))\mathrm{\Lambda }\right)+\frac{}{y^{}}\left((y^{}+p^{}(x))\mathrm{\Psi }\right)=p(x)+\frac{C_2}{2C_1}$$
(62)
where the only unknowns are $`p(x)`$, $`C_1`$, and $`C_2`$. By differentiating the equation above w.r.t $`y^{}`$ and $`x`$ we obtain two equations where the only unknown is $`p^{}(x)`$:
$`\mathrm{\Lambda }_y^{}p^{\prime \prime }(x)+(\mathrm{\Lambda }_{xy^{}}+\mathrm{\Psi }_{y^{}y^{}})(y^{}+p^{}(x))+\mathrm{\Lambda }_x+2\mathrm{\Psi }_y^{}=0`$ (63)
$`\mathrm{\Lambda }p^{\prime \prime \prime }(x)+(\mathrm{\Lambda }_{xx}+\mathrm{\Psi }_{y^{}x})(y^{}+p^{}(x))+(\mathrm{\Lambda }_x+\mathrm{\Psi }_y^{})p^{\prime \prime }(x)+\mathrm{\Psi }_x=p^{}(x)`$ (64)
from where we obtain $`p^{}(x)`$ by solving a linear algebraic equation built by eliminating $`p^{\prime \prime }(x)`$ between Eq.(63) and Eq.(64)<sup>8</sup><sup>8</sup>8From Eq.(60), $`\mathrm{\Lambda }0`$, so that Eq.(64) always depends on $`p^{\prime \prime \prime }(x)`$, and solving Eq.(63) for $`p^{\prime \prime }(x)`$ and substituting twice into Eq.(64) will lead to the desired equation for $`p^{}(x)`$. If Eq.(63) depends on $`p^{}(x)`$ but not on $`p^{\prime \prime }(x)`$, then Eq.(63) itself is already a linear algebraic equation for $`p^{}(x)`$.. Also, as a shortcut, if $`(\mathrm{\Lambda }_{xy^{}}+\mathrm{\Psi }_{y^{}y^{}})/\mathrm{\Lambda }_y^{}`$ depends on $`y^{}`$, then we can build a linear algebraic equation for $`p^{}(x)`$ by solving for $`p^{\prime \prime }(x)`$ in Eq.(63) and differentiating w.r.t. $`y^{}`$. $`\mathrm{}`$
Remark
If Eq.(63) depends neither on $`p^{}(x)`$ nor on $`p^{\prime \prime }(x)`$ this scheme will not succeed. However, in that case the original ODE is actually linear and given by Eq.(28). To see this, we set to zero the coefficients of $`p^{}(x)`$ and $`p^{\prime \prime }(x)`$ in Eq.(63), obtaining:
$$\mathrm{\Lambda }_y^{}=\mathrm{\Lambda }_{xy^{}}+\mathrm{\Psi }_{y^{}y^{}}=\mathrm{\Lambda }_x+2\mathrm{\Psi }_y^{}=0$$
(65)
which implies that $`\mathrm{\Lambda }`$ is a function of $`x`$ only, and then
$$\mathrm{\Psi }_{y^{}y^{}}=0$$
(66)
If we now rewrite $`F(x,y^{})`$ as
$$F(x,y^{})=Z(x,y^{})g(x)\mathrm{\Lambda }(y^{}+p^{})^2C_1$$
(67)
and introduce this expression in Eq.(60), we obtain $`Z_y^{}=0`$; similarly, using this result, Eq.(61), Eq.(66) and Eq.(67) we obtain $`Z_x=0`$. Hence, $`Z`$ is a constant, and taking into account Eq.(67) and Eq.(59), the ODE which led us to this case is just a non-homogeneous linear ODE of the form
$$(y+p)^{\prime \prime }+(\mathrm{\Lambda }^{}(y+p)^{}2(y+p)C_2/C_1)/2\mathrm{\Lambda }=0$$
(68)
whose homogeneous part does not depend on $`p(x)`$:
$$y^{\prime \prime }+\frac{\mathrm{\Lambda }^{}(x)}{2\mathrm{\Lambda }(x)}y^{}\frac{y}{\mathrm{\Lambda }(x)}=0$$
(69)
and as mentioned, it is the same as Eq.(28).
Example: Kamke’s ODE 66.
This ODE is given by
$$y^{\prime \prime }=a\left(c+bx+y\right)\left(y_{}^{}{}_{}{}^{2}+1\right)^{3/2}$$
(70)
Proceeding as in Case A, we determine $`\mathrm{{\rm Y}}`$, $`w(x,y)`$, and $`(y+p(x))`$ as
$$\mathrm{{\rm Y}}=a\left(y_{}^{}{}_{}{}^{2}+1\right)^{3/2};w(x,y)=1;=0$$
(71)
From the last equation we realize that we are in Case D. We determine $`\mathrm{\Lambda }`$ and $`\mathrm{\Psi }`$ (Eqs. (60), (61)) as:
$`\mathrm{\Lambda }`$ $`=`$ $`{\displaystyle \frac{1}{\left(y_{}^{}{}_{}{}^{2}+1\right)^{3/2}a}}`$
$`\mathrm{\Psi }`$ $`=`$ $`c+bx`$ (72)
We then build Eq.(62) for this ODE:
$$\frac{p^{\prime \prime }(x)}{\left(y_{}^{}{}_{}{}^{2}+1\right)^{3/2}a}+c+bx=p(x)+\frac{C_2}{2C_1}$$
(73)
Differentiating w.r.t. $`y^{}`$ leads to Eq.(63):
$$3\frac{p^{\prime \prime }(x)y^{}}{\left(y_{}^{}{}_{}{}^{2}+1\right)^{5/2}a}=0$$
(74)
from which it follows that $`p^{\prime \prime }(x)=0`$. Using this in Eq.(64) we obtain:
$$p^{}(x)=b$$
(75)
after which Eq.(50) becomes
$$(x,y^{})=\frac{y^{}+b}{a\left(y_{}^{}{}_{}{}^{2}+1\right)^{3/2}}$$
(76)
Case E: $`^{}=0`$ and $`0`$
In this case $`(y+p(x))=𝒢^{\prime \prime }/𝒢^{}=C_1`$, so $`𝒢^{}`$ is an exponential function of its argument $`(y+p(x))`$ and hence from Eq.(44)
$$G(x,y)=C_2e^{(y+p(x))C_1}+(y+p(x))C_3+g(x)$$
(77)
for some constants $`C_2`$, $`C_3`$ and some function $`g(x)`$. In this case one of the conditions to be satisfied is
$$\mathrm{{\rm Y}}_y=\text{constant}0$$
(78)
and $`\mathrm{\Phi }(x,y,y^{})`$ will be of the form
$$\mathrm{\Phi }(x,y,y^{})=\frac{F_x(x,y^{})+g^{}(x)+\left(C_2C_1e^{(y+p(x))C_1}+C_3\right)\left(y^{}+p^{}(x)\right)}{F_y^{}(x,y^{})}$$
(79)
Taking advantage of the fact that we explicitly know $`C_1`$, we build a first expression for $`p^{}`$ by dividing $`C_1e^{yC_1}`$ by $`\mathrm{{\rm Y}}`$:
$$\mathrm{\Lambda }\frac{F_y^{}}{C_2e^{p(x)C_1}(y^{}+p^{}(x))}$$
(80)
We obtain a second expression for $`p^{}`$ by multiplying $`\mathrm{\Phi }`$ by $`\mathrm{\Lambda }`$ and subtracting $`C_1e^{C_1y}`$
$$\mathrm{\Psi }\frac{1}{C_2e^{p(x)C_1}}\left(\frac{F_x+g^{}(x)}{y^{}+p^{}(x)}+C_3\right)$$
(81)
As in Case D, $`\mathrm{\Lambda }`$ and $`\mathrm{\Psi }`$ are related by
$$\frac{}{x}\left((y^{}+p^{}(x))\mathrm{\Lambda }\right)+\left(y^{}+p^{}(x)\right)p^{}(x)\mathrm{\Lambda }C_1+\frac{}{y^{}}\left((y^{}+p^{}(x))\mathrm{\Psi }\right)=\frac{C_3}{C_2e^{p(x)C_1}}$$
(82)
where the only unknowns are $`C_2`$, $`C_3`$ and $`p(x)`$. Differentiating Eq.(82) with respect to $`y^{}`$ we have
$$\begin{array}{ccc}\multicolumn{3}{c}{\left(p^{\prime \prime }(x)+p^{}(x)^2C_1\right)\mathrm{\Lambda }_y^{}+p^{}(x)\left(y^{}\mathrm{\Lambda }_y^{}C_1+\mathrm{\Lambda }C_1+\mathrm{\Lambda }_{xy^{}}+\mathrm{\Psi }_{y^{}y^{}}\right)}\\ & \text{ }& +2\mathrm{\Psi }_y^{}+\mathrm{\Lambda }_x+y^{}\mathrm{\Lambda }_{xy^{}}+y^{}\mathrm{\Psi }_{y^{}y^{}}=0\hfill \end{array}$$
The problem now is that, due to the exponential on the RHS of Eq.(82), differently from Case D, we are not able to obtain a second expression for $`p^{}(x)`$ by differentiating w.r.t $`x`$. The alternative we have found can be summarized as follows. We first note that if $`\mathrm{\Lambda }_y^{}=0`$, Eq.(2.2.3) is already a linear algebraic equation<sup>9</sup><sup>9</sup>9We can see this by assuming that $`\mathrm{\Lambda }_y^{}=0`$ and that Eq.(2.2.3) does not contain $`p^{}`$, and then arriving at a contradiction as follows. We first set the coefficients of $`p^{}`$ in Eq.(2.2.3) to zero, arriving at
$$\text{ }0=C_1\mathrm{\Lambda }+\mathrm{\Psi }_{y^{}y^{}}=2\mathrm{\Psi }_y^{}+\mathrm{\Lambda }_x+\mathrm{\Psi }_{y^{}y^{}}y^{}\text{ }(A)$$
Eliminating $`\mathrm{\Psi }_{y^{}y^{}}`$ gives
$$2\mathrm{\Psi }_y^{}=C_1\mathrm{\Lambda }y^{}\mathrm{\Lambda }_x$$
Differentiating the expression above w.r.t $`y^{}`$ and since $`\mathrm{\Lambda }_y^{}=0`$, we have
$$2\mathrm{\Psi }_{y^{}y^{}}=C_1\mathrm{\Lambda }$$
Finally, using Eq.(A), $`0=\mathrm{\Lambda }`$, contradicting $`F_y^{}0`$. for $`p^{}`$, so that we are only worried with the case $`\mathrm{\Lambda }_y^{}0`$. With this in mind, we divide Eq.(2.2.3) by $`\mathrm{\Lambda }_y^{}`$ and, if the resulting expression depends on $`y^{}`$, we directly obtain a linear algebraic equation in $`p^{}(x)`$ by just differentiating w.r.t $`y^{}`$. $`\mathrm{}`$
Example<sup>10</sup><sup>10</sup>10There are no examples of this type in all of Kamke’s set of non-linear second order ODEs.:
$$y^{\prime \prime }=\frac{y^{}\left(xy^{}+1\right)\left(2+e^y\right)}{y^{}x^2+y^{}1}$$
(83)
We determine $`\mathrm{{\rm Y}}`$, $`w(x,y)`$, and $`(y+p(x))`$ as
$$\mathrm{{\rm Y}}=\frac{y^{}(xy^{}+1)e^y}{y^{}x^2+y^{}1};w(x,y)=e^y;=1$$
(84)
From the last equation we know that we are in Case E. We then determine $`\mathrm{\Lambda }`$ and $`\mathrm{\Psi }`$ as in Eqs. (80) and (81):
$`\mathrm{\Lambda }`$ $`=`$ $`{\displaystyle \frac{y^{}x^2+y^{}1}{y^{}(xy^{}+1)}}`$
$`\mathrm{\Psi }`$ $`=`$ $`2`$ (85)
Now, we build Eq.(82):
$$\frac{1}{xy^{}+1}\left(\left(p^{\prime \prime }+p_{}^{}{}_{}{}^{2}+y_{}^{}{}_{}{}^{2}\frac{(xp^{}1)}{xy^{}+1}\right)\left(x^2+1\frac{1}{y^{}}\right)+2xp^{}2\right)=\frac{C_3}{C_2e^p}$$
(86)
and, differentiating w.r.t. $`y^{}`$, we obtain Eq.(2.2.3):
$$\frac{2xy^{}+1(x^3+x)y_{}^{}{}_{}{}^{2}}{y_{}^{}{}_{}{}^{2}(xy^{}+1)^2}\left(p^{\prime \prime }+p_{}^{}{}_{}{}^{2}\right)+\frac{2y^{}12x+xy^{}}{\left(xy^{}+1\right)^3}(xp^{}1)=0$$
(87)
Proceeding as explained, dividing by $`\mathrm{\Lambda }_y^{}`$ and differentiating w.r.t. $`y^{}`$, we have
$$\frac{}{y^{}}\left(y_{}^{}{}_{}{}^{2}\frac{2y^{}12x+xy^{}}{\left(xy^{}+1\right)(2xy^{}+1(x^3+x)y_{}^{}{}_{}{}^{2})}\right)(xp^{}1)=0$$
(88)
Solving for $`p^{}(x)`$ gives $`p^{}(x)=1/x`$, from which Eq.(50) becomes:
$$(x,y^{})=\left(y^{}\frac{1}{x}\right)\frac{y^{}x^2+y^{}1}{y^{}(xy^{}+1)}$$
(89)
Case F
The final branch occurs when Eq.(2.2.3) divided by $`\mathrm{\Lambda }_y^{}`$ does not depend on $`y^{}`$ (so that we will not be able to differentiate w.r.t $`y^{}`$). In this case we can build a linear algebraic equation for $`p^{}(x)`$ as follows. Let us introduce the label $`\beta (x,p^{},p^{\prime \prime })`$ for Eq.(2.2.3) divided by $`\mathrm{\Lambda }_y^{}`$, so that Eq.(2.2.3) becomes:
$$\mathrm{\Lambda }_y^{}(x,y^{})\beta (x,p^{},p^{\prime \prime })=0$$
(90)
Since we obtained Eq.(2.2.3) by differentiating Eq.(82) with respect to $`y^{}`$, Eq.(82) can be written in terms of $`\beta `$ by integrating Eq.(90) with respect to $`y^{}`$:
$$\mathrm{\Lambda }(x,y^{})\beta (x,p^{},p^{\prime \prime })+\gamma (x,p^{},p^{\prime \prime })=\frac{C_3}{C_2e^{p(x)C_1}}$$
(91)
where $`\gamma (x,p^{},p^{\prime \prime })`$ is the constant of integration, and can be determined explicitly in terms of $`x`$, $`p^{}`$ and $`p^{\prime \prime }`$ by comparing Eq.(91) with Eq.(82). Taking into account that $`\beta (x,p^{},p^{\prime \prime })=0`$, Eq.(91) reduces to:
$$\gamma (x,p^{},p^{\prime \prime })=\frac{C_3}{C_2e^{p(x)C_1}}$$
(92)
We can remove the unknowns $`C_2`$ and $`C_3`$ after multiplying Eq.(92) by $`e^{p(x)C_1}`$, differentiating with respect to $`x`$, and then dividing once again by $`e^{p(x)C_1}`$. We now have our second equation for $`p^{}`$, which we can build explicitly in terms of $`p^{}`$, since we know $`\gamma (x,p^{},p^{\prime \prime })`$ and $`C_1`$:
$$\frac{d\gamma }{dx}+C_1p^{}\gamma =0$$
(93)
Eliminating the derivatives of $`p^{}`$ between Eq.(90) and Eq.(93) leads to a linear algebraic equation in $`p^{}`$. Once we have $`p^{}`$, the determination of $`(x,y^{})`$ follows directly from Eq.(50). $`\mathrm{}`$
### 2.3 Integrating factors of the form $`\mu (y,y^{})`$
From Eq.(11), the ODE family admitting an integrating factor of the form $`\mu (y,y^{})`$ is given by
$$y^{\prime \prime }=\frac{y^{}}{\mu }\left(G_y+\frac{}{y}\mu 𝑑y^{}\right)\frac{G_x}{\mu }$$
(94)
where $`\mu (y,y^{})`$ and $`G(x,y)`$ are arbitrary functions of their arguments. For this ODE family, it would be possible to develop an analysis and split the problem into cases as done in the previous section for the case $`\mu (x,y^{})`$. However, it is straightforward to notice that under the transformation $`y(x)x,xy(x)`$, Eq.(94) transforms into an ODE of the form Eq.(23) with integrating factor $`\mu (x,y_{}^{}{}_{}{}^{1})/y_{}^{}{}_{}{}^{2}`$. It follows that an integrating factor for any member of the ODE family above can be found by merely changing variables in the given ODE and calculating the corresponding integrating factor of the form $`\mu (x,y^{})`$.
Example:
$$y^{\prime \prime }\frac{y_{}^{}{}_{}{}^{2}}{y}+\mathrm{sin}(x)y^{}y+\mathrm{cos}(x)y^2=0$$
(95)
Changing variables $`y(x)x,xy(x)`$ we obtain
$$y^{\prime \prime }+\frac{y^{}}{x}\mathrm{sin}(y)y_{}^{}{}_{}{}^{2}x\mathrm{cos}(y)x^2y_{}^{}{}_{}{}^{3}=0$$
(96)
Using the algorithm outlined in the previous section, an integrating factor of the form $`\mu (x,y^{})`$ for Eq.(96) is given by
$$\frac{1}{y_{}^{}{}_{}{}^{2}x}$$
(97)
from where an integrating factor of the form $`\mu (y,y^{})`$ for Eq.(95) is $`1/y`$, leading to the first integral
$$\mathrm{sin}(x)y+\frac{y^{}}{y}+C_1=0,$$
(98)
which is a first order ODE of Bernoulli type. The solution to Eq.(95) then follows directly. This example is interesting since from Eq.(95) has no point symmetries.
## 3 Integrating factors and symmetries
Besides the formulas for integrating factors of the form $`\mu (x,y)`$, the main result presented in this paper is a systematic algorithm for the determination of integrating factors of the form $`\mu (x,y^{})`$ and $`\mu (y,y^{})`$ without solving any auxiliary differential equations or performing differential Groebner basis calculations, and these last two facts constitute the relevant point. Nonetheless, it is interesting to briefly compare the standard integrating factor ($`\mu `$) and symmetry approaches, so as to have an insight of how complementary these methods can be in practice.
To start with, both methods tackle an n<sup>th</sup> order ODE by looking for solutions to a linear n<sup>th</sup> order determining PDE in $`n+1`$ variables. Any given ODE has infinitely many integrating factors and symmetries. When many solutions to these determining PDEs are found, both approaches can, in principle, give a multiple reduction of order.
In the case of integrating factors there is one unknown function, while for symmetries there is a pair of infinitesimals to be found. But symmetries are defined up to an arbitrary function, so that we can always take one of these infinitesimals equal to zero<sup>11</sup><sup>11</sup>11Symmetries $`[\xi (x,y,..y^{(n1)}),\eta (x,y,..y^{(n1)})]`$ of an $`\mathrm{n}^{\mathrm{th}}`$order ODE can always be rewritten as $`[G,(G\xi )y^{}+\eta ]`$, where $`G(x,y,..y^{(n1)})`$ is an arbitrary function (for first order ODEs, $`y^{}`$ must replaced by the right-hand-side of the ODE). Choosing $`G`$ = 0 the symmetry acquires the form $`[0,\overline{\eta }]`$; hence we are facing approaches of equivalent levels of difficulty and actually of equivalent solving power.
Also valid for both approaches is the fact that, unless some restrictions are introduced on the functional dependence of $`\mu `$ or the infinitesimals, there is no hope that the corresponding determining PDEs will be easier to solve than the original ODE. In the case of symmetries, it is usual to restrict the problem to ODEs having point symmetries, that is, to consider infinitesimals depending only on $`x`$ and $`y`$. The restriction to the integrating factors here discussed is similar: we considered $`\mu `$’s depending on only two variables.
At this point it can be seen that the two approaches are complementary: the determining PDEs for $`\mu `$ and for the symmetries are different<sup>12</sup><sup>12</sup>12We are considering here ODEs of order greater than one., so that even using identical restrictions on the functional dependence of $`\mu `$ and the infinitesimals, problems which may be untractable using one approach may be easy or even trivial using the other one.
As an example of this, consider Kamke’s ODE 6.37
$$y^{\prime \prime }+2yy^{}+f(x)\left(y^{}+y^2\right)g(x)=0$$
(99)
For arbitrary $`f(x)`$ and $`g(x)`$, this ODE has an integrating factor depending only on $`x`$, easily determined using the algorithms presented. Now, for non-constant $`f(x)`$ and $`g(x)`$, this ODE has no point symmetries, that is, no infinitesimals of the form $`[\xi (x,y),\eta (x,y)]`$, except for the particular case in which $`g(x)`$ can be expressed in terms of $`f(x)`$ as in<sup>13</sup><sup>13</sup>13To determine $`g(x)`$ in terms of $`f(x)`$ we used the standard form Maple package by Reid and Wittkopf complemented with some basic calculations.
$$g(x)=\frac{f^{\prime \prime }}{4}+\frac{3ff^{}}{8}+\frac{f^3}{16}\frac{C_2\mathrm{exp}\left(3/2{\displaystyle f(x)𝑑x}\right)}{4\left(2C_1+{\displaystyle \mathrm{exp}\left(1/2f(x)𝑑x\right)𝑑x}\right)^3}$$
(100)
Furthermore, this ODE does not have non-trivial symmetries of the form $`[\xi (x,y^{}),\eta (x,y^{})]`$ either, and for symmetries of the form $`[\xi (y,y^{}),\eta (y,y^{})]`$ the determining PDE does not split into a system.
Another ODE example of this type is found in a paper by (1988):
$$y^{\prime \prime }\frac{y_{}^{}{}_{}{}^{2}}{y}g(x)py^py^{}g^{}y^{p+1}=0$$
(101)
In that work it is shown that for constant $`p`$, the ODE above only has point symmetries for very restricted forms of $`g(x)`$. For instance, Eq.(95) is a particular case of the ODE above and has no point symmetries. On the other hand, for arbitrary $`g(x)`$, Eq.(101) has an obvious integrating factor depending on only one variable: $`1/y`$, leading to a first integral of Bernoulli type:
$$\frac{y^{}}{y}g(x)y^p+C_1=0$$
(102)
so that the whole family Eq.(101) is integrable by quadratures.
We note that Eq.(99) and Eq.(101) are respectively particular cases of the general reducible ODEs having integrating factors of the form $`\mu (x)`$:
$$y^{\prime \prime }=\frac{\left(\mu _x+G_y\right)}{\mu (x)}y^{}\frac{G_x}{\mu (x)}$$
(103)
where $`\mu (x)`$ and $`G(x,y)`$ are arbitrary; and $`\mu (y)`$:
$$y^{\prime \prime }=\frac{(\mu _yy^{}+G_y)}{\mu (y)}y^{}\frac{G_x}{\mu (y)}$$
(104)
In turn, these are very simple cases if compared with the general ODE families Eq.(23) and Eq.(94), respectively having integrating factors of the forms $`\mu (x,y^{})`$ and $`\mu (y,y^{})`$, and which can be systematically reduced in order using the algorithms here presented.
It is then natural to conclude that the integrating factor and the symmetry approaches are useful for solving different types of ODEs, and can be viewed as equivalently powerful and general, and in practice complementary. Moreover, if for a given ODE, an integrating factor and a symmetry are known, in principle one can combine this information to build two first integrals and reduce the order by two at once (see for instance ).
## 4 Tests
After plugging the reducible-ODE scheme here presented into the ODEtools package , we tested the scheme and routines using Kamke’s non-linear 246 second order ODE examples<sup>14</sup><sup>14</sup>14Kamke’s ODEs 6.247 to 6.249 cannot be made explicit and are then excluded from the tests.. The purpose was to confirm the correctness of the returned results and to determine which of these ODEs have integrating factors of the form $`\mu (x,y)`$, $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$. The test consisted of determining $`\mu `$ and testing the exactness condition Eq.(3).
In addition, we ran a comparison of performances in solving a subset of Kamke’s examples having integrating factors of the forms $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$, using different computer algebra ODE-solvers (Maple, Mathematica, MuPAD and the Reduce package Convode). The idea was to situate the new scheme in the framework of a sample of relevant packages presently available.
To run the comparison of performances, the first step was to classify Kamke’s ODEs into: missing x, missing y, exact and reducible, where the latter refers to ODEs having integrating factors of the forms $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$. ODEs missing variables were not included in the test since they can be seen as first order ODEs in disguised form, and as such they are not the main target of the algorithm being presented. The classification we obtained for these 246 ODEs is as follows
| Classification | ODE numbers as in Kamke’s book |
| --- | --- |
| 99 ODEs are missing $`x`$ or missing $`y`$ | 1, 2, 4, 7, 10, 12, 14, 17, 21, 22, 23, 24, 25, 26, 28, 30, 31, 32, 40, 42, 43, 45, 46, 47, 48, 49, 50, 54, 56, 60, 61, 62, 63, 64, 65, 67, 71, 72, 81, 89, 104, 107, 109, 110, 111, 113, 117, 118, 119, 120, 124, 125, 126, 127, 128, 130, 132, 137, 138, 140, 141, 143, 146, 150, 151, 153, 154, 155, 157, 158, 159, 160, 162, 163, 164, 165, 168, 188, 191, 192, 197, 200, 201, 202, 209, 210, 213, 214, 218, 220, 222, 223, 224, 232, 234, 236, 237, 243, 246 |
| 13 are in exact form | 36, 42, 78, 107, 108, 109, 133, 169, 170, 178, 226, 231, 235 |
| 40 ODEs are reducible with integrating factor $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$ and missing $`x`$ or $`y`$ | 1, 2, 4, 7, 10, 12, 14, 17, 40, 42, 50, 56, 64, 65, 81, 89, 104, 107, 109, 110, 111, 125, 126, 137, 138, 150, 154, 155, 157, 164, 168, 188, 191, 192, 209, 210, 214, 218, 220, 222, 236 |
| 28 ODEs are reducible and not missing $`x`$ or $`y`$ | 36, 37, 51, 66, 78, 97, 108, 123, 133, 134, 135, 136, 166, 169, 173, 174, 175, 176, 178, 179, 193, 196, 203, 204, 206, 215, 226, 235 |
| Table 1. Missing variables, exact and reducible Kamke’s 246 second order non-linear ODEs. | |
For our purposes, the interesting subset is the one comprised of the 28 ODEs not already missing variables. The results we obtained using the aforementioned computer algebra ODE-solvers<sup>15</sup><sup>15</sup>15Maple R4 is not present in the table since it is not solving any of these 28 ODEs. This situation is being resolved in the upcoming Maple R5, where the ODEtools routines are included in the Maple library, and the previous ODE-solver was replaced by odsolve. However, the scheme here presented was not ready when the development library was closed; the reducible scheme implemented in Maple R5 is able to determine, when they exist, integrating factors only of the form $`\mu (y^{})`$. are summarized as follows<sup>16</sup><sup>16</sup>16Some of these 28 ODEs are given in Kamke in exact form and hence they can be easily reduced after performing a check for exactness; before running the tests all these ODEs were rewritten in explicit form by isolating $`y^{\prime \prime }`$.:
| Kamke’s ODE numbers | | | | |
| --- | --- | --- | --- | --- |
| | Convode | Mathematica 3.0 | MuPAD 1.3 | ODEtools |
| Solved: | 51, 166, 173, 174, 175, 176, 179. | 78, 97, 108, 166, 169, 173, 174, 175, 176, 178, 179, 206. | 78, 97, 108, 133, 166, 169, 173, 174, 175, 176, 179, | 51, 78, 97, 108, 133, 134, 135, 136, 166, 169, 173, 174, 175, 176, 178, 179, 193, 196, 203, 204, 206, 215. |
| Totals: | 7 | 12 | 11 | 22 |
| Reduced: | | | | 36, 37, 66, 123, 226, 235. |
| Totals: | 0 | 0 | 0 | 6 |
| Table 2. Performances in solving 28 Kamke’s ODEs having an integrating factor $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$ | | | | |
As shown above, while the scheme here presented is finding first integrals in all the 28 ODE examples, opening the way to solve 22 of them to the end, the next scores are only 12 and 11 ODEs, respectively solved by Mathematica 3.0 and MuPAD 1.3.
Concerning the six reductions of order returned by odsolve, it must be said that neither MuPAD nor Mathematica provide a way to convey them, so that perhaps their ODE-solvers are obtaining first integrals for these cases but the routines are giving up when they cannot solve the problem to the end.
## 5 Conclusions
In connection with second order ODEs, this paper presented a systematic method for determining the existence of integrating factors and their explicit form, when they have the forms $`\mu (x,y)`$, $`\mu (x,y^{})`$ and $`\mu (y,y^{})`$. The scheme is new, as far as we know, and its implementation in the framework of the computer algebra package ODEtools has proven to be a valuable tool. Actually, the implementation of the scheme solves ODEs not solved by using standard or symmetry methods (see sec. 3) or some other relevant and popular computer algebra ODE-solvers (see sec. 4).
Furthermore, the algorithms presented involve only very simple operations and do not require solving auxiliary differential equations, except in one branch of the $`\mu (x,y)`$ problem. So, even for examples where other methods also work, for instance by solving the related PDE system Eqs.(6) and (7) using ansatzes and differential Groebner basis techniques, the method here presented can return answers faster and avoiding potential explosions of memory<sup>17</sup><sup>17</sup>17Explosions of memory may happen when calculating all the integrability conditions involved at each step in the differential Groebner basis approach..
On the other hand, we have restricted the problem to the universe of second order ODEs having integrating factors depending only on two variables while packages as CONLAW (in REDUCE) can try and in some cases solve the PDE system Eqs.(6) and (7) by using more varied ansatzes for $`\mu `$.
A natural extension of this work would be to develop a scheme for building integrating factors of restricted but more general forms, now for higher order ODEs. We are presently working on these possible extensions<sup>18</sup><sup>18</sup>18See http://lie.uwaterloo.ca/odetools.html, and expect to succeed in obtaining reportable results in the near future.
## Acknowledgments
This work was supported by the State University of Rio de Janeiro (UERJ), Brazil, and by the Symbolic Computation Group, Faculty of Mathematics, University of Waterloo, Ontario, Canada. The authors would like to thank K. von Bülow<sup>19</sup><sup>19</sup>19Symbolic Computation Group of the Theoretical Physics Department at UERJ - Brazil. for a careful reading of this paper.
## Appendix A
We display here both the integrating factors obtained for the 28 Kamke’s ODEs used in the tests (see sec. 4) and the “case” corresponding to each ODE when using just the algorithms for $`\mu (x,y^{})`$ or $`\mu (y,y^{})`$<sup>20</sup><sup>20</sup>20We note that for non-linear ODEs these two algorithms work as well when $`\mu _y^{}=0`$, but in practice these very simple examples are covered by the algorithm for $`\mu (x,y)`$ presented in sec. 2.1.. As explained in sec. 2.2.3, the algorithm presented is subdivided into different cases: A, B, C, D, E and F, and case B is always either A or C.
| Integrating factor | Kamke’s book ODE-number | Case |
| --- | --- | --- |
| $`1`$ | 36 | D |
| $`e^{{\scriptscriptstyle f\left(x\right)𝑑x}}`$ | 37 | A |
| $`y_{}^{}{}_{}{}^{1}`$ | 51, 166, 169, 173, 175, 176, 179, 196, 203, 204, 206, 215 | C |
| $`\frac{b+y^{}}{\left(1+y_{}^{}{}_{}{}^{2}\right)^{3/2}}`$ | 66 | D |
| $`x`$ | 78 | D |
| $`x^1`$ | 97 | A |
| $`y`$ | 108 | D |
| $`y^1`$ | 123 | A |
| $`\frac{1+y^{}}{\left(y^{}1\right)y^{}}`$ | 133 | C |
| $`\frac{y^{}1}{\left(1+y^{}\right)y^{}}`$ | 134 | C |
| $`\frac{y^{}1}{\left(1+y^{}\right)\left(1+y_{}^{}{}_{}{}^{2}\right)}`$ | 135 | C |
| $`\frac{y^{}1}{h\left(y^{}\right)}`$ | 136 | C |
| $`\frac{x}{2xy^{}1}`$ | 174 | C |
| $`\left(1+y^{}\right)^1`$ | 178 | C |
| $`\frac{1}{y^{}\left(1+2yy^{}\right)}`$ | 193 | C |
| $`y^{}`$ | 226 | A |
| $`h\left(y^{}\right)`$ | 235 | C |
| Integrating factors for Kamke’s ODEs which are reducible and not missing $`x`$ or $`y`$. | | |
|
warning/0002/hep-ph0002141.html
|
ar5iv
|
text
|
# Neutrino oscillation, SUSY GUT and 𝐵 decay
## Abstract
Effects of supersymmetric particles on flavor changing neutral current and lepton flavor violating processes are studied in the supersymmetric SU(5) grand unified theory with right-handed neutrino supermultiplets. Using input parameters motivated by neutrino oscillation, it is shown that the time-dependent CP asymmetry of radiative $`B`$ decay can be as large as 30% when the $`\tau \mu \gamma `$ branching ratio becomes close to the present experimental upper bound. We also show that the $`B_s`$$`\overline{B}_s`$ mixing can be significantly different from the presently allowed range in the standard model.
preprint: KEK Preprint 99-176 KEK-TH-677
Although the standard model (SM) of the elementary particle theory describes current experimental results very well, particles and interactions outside of the SM may appear beyond the energy scale available at current collider experiments. One of indications is already given by the atmospheric and the solar neutrino anomalies which have been interpreted as evidences of neutrino oscillation . A natural way to introduce small neutrino masses for the neutrino oscillation is the see-saw mechanism where the right-handed neutrino is introduced with a very heavy mass. This scenario suggests the existence of a new source of flavor mixings in the lepton sector at much higher energy scale than the electroweak scale.
In this letter we consider flavor changing neutral current (FCNC) processes and lepton flavor violation (LFV) of charged lepton decays in the model of a SU(5) supersymmetric (SUSY) grand unified theory (GUT) which incorporates the see-saw mechanism for the neutrino mass generation. In the SUSY theory, the superpartners of quarks and leptons, namely squarks and sleptons respectively, have new flavor mixings in their mass matrices. In the model based on the minimal supergravity these mass matrices are assumed to be flavor-blind at the Planck scale. However renormalization effects due to Yukawa coupling constants can induce flavor mixing in the squark/slepton mass matrices. In the present model sources of the flavor mixing are Yukawa coupling constant matrices for quarks and leptons as well as that for the right-handed neutrinos. Because the quark and lepton sectors are related by GUT interactions, the flavor mixing relevant to the Cabibbo-Kobayashi-Maskawa (CKM) matrix can generate the LFV such as $`\mu e\gamma `$ and $`\tau \mu \gamma `$ processes in addition to FCNC in hadronic observables . In the SUSY model with right-handed neutrinos, it is possible that the branching ratios of the LFV processes become large enough to be measured in near-future experiments . When we consider the right-handed neutrinos in the context of GUT, the flavor mixing related to the neutrino oscillation can be a source of the flavor mixing in the squark sector. We show that due to the large mixing of the second and third generations suggested by the atmospheric neutrino anomaly, $`B_s`$$`\overline{B}_s`$ mixing, the time-dependent CP asymmetry of the $`BM_s\gamma `$ process, where $`M_s`$ is a CP eigenstate including the strange quark, can have a large deviation from the SM prediction.
The Yukawa coupling part and the Majorana mass term of the superpotential for the SU(5) SUSY GUT with right-handed neutrino supermultiplets is given by
$$W=\frac{1}{8}f_U^{ij}\mathrm{\Psi }_i\mathrm{\Psi }_jH_5+f_D^{ij}\mathrm{\Psi }_i\mathrm{\Phi }_jH_{\overline{5}}+f_N^{ij}N_i\mathrm{\Phi }_jH_5+\frac{1}{2}M_\nu ^{ij}N_iN_j,$$
(1)
where $`\mathrm{\Psi }_i`$, $`\mathrm{\Phi }_i`$ and $`N_i`$ are $`\mathrm{𝟏𝟎}`$, $`\overline{\mathrm{𝟓}}`$ and $`\mathrm{𝟏}`$ representations of SU(5) gauge group. $`i,j=1,2,3`$ are the generation indices. $`H_5`$ and $`H_{\overline{5}}`$ are Higgs superfields with $`\mathrm{𝟓}`$ and $`\overline{\mathrm{𝟓}}`$ representations. In terms of the SU(3)$`\times `$SU(2)$`{}_{L}{}^{}\times `$U(1)<sub>Y</sub>, $`\mathrm{\Psi }_i`$ contains $`Q_i(\mathrm{𝟑},\mathbf{\hspace{0.17em}2},\frac{1}{6})`$, $`U_i(\overline{\mathrm{𝟑}},\mathbf{\hspace{0.17em}1},\frac{2}{3})`$ and $`E_i(\mathrm{𝟏},\mathbf{\hspace{0.17em}1},\mathrm{\hspace{0.17em}1})`$ superfields. Here the representations for SU(3) and SU(2) groups and the U(1)<sub>Y</sub> charge are indicated in the parentheses. $`\mathrm{\Phi }_i`$ includes $`D_i(\overline{\mathrm{𝟑}},\mathbf{\hspace{0.17em}1},\frac{1}{3})`$ and $`L_i(\mathrm{𝟏},\mathbf{\hspace{0.17em}2},\frac{1}{2})`$, and $`N_i`$ is a singlet of SU(3)$`\times `$SU(2)$`{}_{L}{}^{}\times `$U(1)<sub>Y</sub>. $`M_\nu `$ is the Majorana mass matrix. Below the GUT scale ($`2\times 10^{16}`$ GeV) and the Majorana mass scale ($`M_R`$) the superpotential for the minimal supersymmetric standard model (MSSM) fields is given by
$$W_{\text{MSSM}}=\stackrel{~}{f}_U^{ij}Q_iU_jH_2+\stackrel{~}{f}_D^{ij}Q_iD_jH_1+\stackrel{~}{f}_L^{ij}E_iL_jH_1+\mu H_1H_2\frac{1}{2}\kappa _\nu ^{ij}(L_iH_2)(L_jH_2),$$
(2)
where $`\kappa _\nu `$ is obtained by integrating out the heavy right-handed neutrino fields. At the right-handed neutrino mass scale $`\kappa _\nu `$ is given as $`\kappa _\nu ^{ij}=(f_N^𝐓M_\nu ^1f_N)^{ij}`$. The Yukawa coupling constants $`\stackrel{~}{f}_U^{ij}`$, $`\stackrel{~}{f}_D^{ij}`$ and $`\stackrel{~}{f}_L^{ij}`$ are related to the coupling constants $`f_U^{ij}`$ and $`f_D^{ij}`$ at the GUT scale. The quark, charged lepton and neutrino masses and mixings are determined from the superpotential Eq. (2) at the low energy scale.
As discussed above, the renormalization effects due to the Yukawa coupling constants induce various FCNC and LFV effects from the mismatch between the quark/lepton and squark/slepton diagonalization matrices. In particular the large top Yukawa coupling constant is responsible for the renormalization of the $`\stackrel{~}{q}_L`$ and $`\stackrel{~}{u}_R`$ mass matrices. At the same time the $`\stackrel{~}{e}_R`$ mass matrix receives sizable corrections between the Planck and the GUT scales and various LFV processes are induced. In a similar way, if the neutrino Yukawa coupling constant $`f_N^{ij}`$ is large enough, the $`\stackrel{~}{l}_L`$ mass matrix and the $`\stackrel{~}{d}_R`$ mass matrix receive sizable flavor changing effects due to renormalization between the Planck and the $`M_R`$ scales and the Planck and the GUT scales, respectively. These are sources of extra contributions to LFV processes and various FCNC processes such as $`bs\gamma `$, the $`B^0`$$`\overline{B}^0`$ mixing and the $`K^0`$$`\overline{K}^0`$ mixing.
It is particularly interesting that the chiral structure of the FCNC amplitudes due to the $`\stackrel{~}{d}_R`$ mixing is different from that expected in the SM. For example, the flavor mixing in the $`\stackrel{~}{d}_R`$ sector generates a sizable contribution to the $`bs\gamma _R`$ amplitude through gluino–$`\stackrel{~}{d}_R`$ loop diagrams, whereas this amplitude is suppressed by a factor $`m_s/m_b`$ over the dominant $`bs\gamma _L`$ amplitude in the SM. When the amplitudes with both chiralities exist, the mixing-induced time-dependent CP asymmetry in the $`BM_s\gamma `$ process can be induced. Using the Wilson coefficients $`c_7`$ and $`c_7^{}`$ in the effective Lagrangian for the $`bs\gamma `$ decay $`=c_7\overline{s}\sigma ^{\mu \nu }b_RF_{\mu \nu }+c_7^{}\overline{s}\sigma ^{\mu \nu }b_LF_{\mu \nu }+\text{H.c.}`$, the asymmetry is written as
$$\frac{\mathrm{\Gamma }(t)\overline{\mathrm{\Gamma }}(t)}{\mathrm{\Gamma }(t)+\overline{\mathrm{\Gamma }}(t)}=\xi A_t\mathrm{sin}\mathrm{\Delta }m_dt,A_t=\frac{2\text{Im}(\mathrm{e}^{i\theta _B}c_7c_7^{})}{|c_7|^2+|c_7^{}|^2},$$
where $`\mathrm{\Gamma }(t)`$ ($`\overline{\mathrm{\Gamma }}(t)`$) is the decay width of $`B^0(t)M_s\gamma `$ ($`\overline{B}^0(t)M_s\gamma `$) and $`M_s`$ is some CP eigenstate ($`\xi =+1(1)`$ for a CP even (odd) state) such as $`K_S\pi ^0`$ . $`\mathrm{\Delta }m_d=2|M_{12}(B_d)|`$ and $`\theta _B=\mathrm{arg}M_{12}(B_d)`$ where $`M_{12}(B_d)`$ is the $`B_d`$$`\overline{B}_d`$ mixing amplitude. Because the asymmetry can be only a few percent in the SM, a sizable asymmetry is a clear signal of new physics beyond the SM.
We calculated the following observables in the FCNC and LFV processes: the CP violation parameter in the $`K^0`$$`\overline{K}^0`$ mixing $`\epsilon _K`$, $`B_d`$$`\overline{B}_d`$ and $`B_s`$$`\overline{B}_s`$ mass splittings $`\mathrm{\Delta }m_d`$ and $`\mathrm{\Delta }m_s`$, respectively, $`A_t`$ and the branching ratios $`\text{B}(bs\gamma )`$, $`\text{B}(\mu e\gamma )`$, $`\text{B}(\tau \mu \gamma )`$ and $`\text{B}(\tau e\gamma )`$. We solved renormalization group equations (RGEs) for Yukawa coupling constants and the SUSY breaking parameters numerically keeping all flavor matrices. After demanding the condition of radiative electroweak symmetry breaking, the free parameters in the minimal supergravity model are the universal scalar mass $`m_0`$, the universal gaugino mass $`M_0`$, the scalar trilinear parameter $`A_0`$, the ratio of two vacuum expectation values $`\mathrm{tan}\beta `$ and the sign of the Higgsino mass parameter $`\mu `$. In addition we need to specify neutrino parameters. The phenomenological inputs from neutrino oscillation are two mass-squared differences and the Maki-Nakagawa-Sakata (MNS) matrix. In order to relate these parameters to $`f_N`$ and $`M_\nu `$, we work in the basis for $`N_i`$, $`L_i`$ and $`E_i`$ where $`\stackrel{~}{f}_L^{ij}=\widehat{f}_L^{ij}`$ and $`f_N^{ij}=(\widehat{f}_NV_L)^{ij}`$ ($`\widehat{f}_L`$ and $`\widehat{f}_N`$ are diagonal matrix) at the matching scale $`M_R`$. In this basis $`\kappa _\nu =V_L^𝐓\widehat{f}_NM_\nu ^1\widehat{f}_NV_L=V_{\text{MNS}}^0\widehat{\kappa }_\nu V_{\text{MNS}}^0`$ where $`V_{\text{MNS}}^0`$ is the MNS matrix at $`M_R`$ and $`\widehat{\kappa }_\nu `$ is a diagonal matrix. Note that although $`V_L=V_{\text{MNS}}^0`$ when $`M_\nu `$ is diagonal in this basis, two are independent in a general case. Once we specify three neutrino masses, $`V_{\text{MNS}}`$, $`V_L`$ and $`\widehat{f}_N`$ we can determine the $`M_\nu `$ matrix. Then using the GUT relation for Yukawa coupling constants, we can calculate all squark and slepton mass matrices through RGEs. Note that $`V_L`$ essentially determines the flavor mixing in the $`\stackrel{~}{d}_R`$ and $`\stackrel{~}{l}_L`$ sectors in this basis.
As typical examples of the neutrino parameters, we consider the following parameter sets corresponding to (i) the Mikheyev-Smirnov-Wolfenstein (MSW) small mixing angle and (ii) the MSW large mixing angle solutions for the solar neutrino problem .
(i) small mixing:
$`m_\nu `$ $`=`$ $`(2.236\times 10^3,\mathrm{\hspace{0.17em}3.16}\times 10^3,\mathrm{\hspace{0.17em}5.92}\times 10^2)\mathrm{eV},`$ (3)
$`V_{\text{MNS}}`$ $`=`$ $`\left(\begin{array}{ccc}0.999& 0.0371& 0\\ 0.0262& 0.707& 0.707\\ 0.0262& 0.707& 0.707\end{array}\right),`$ (7)
(ii) large mixing:
$`m_\nu `$ $`=`$ $`(4.0\times 10^3,\mathrm{\hspace{0.17em}5.831}\times 10^3,\mathrm{\hspace{0.17em}5.945}\times 10^2)\mathrm{eV},`$ (8)
$`V_{\text{MNS}}`$ $`=`$ $`\left(\begin{array}{ccc}1/\sqrt{2}& 1/\sqrt{2}& 0\\ 1/2& 1/2& 1/\sqrt{2}\\ 1/2& 1/2& 1/\sqrt{2}\end{array}\right).`$ (12)
In each example we also take $`M_\nu `$ to be proportional to a unit matrix with a diagonal element of $`M_R=4\times 10^{14}`$ GeV so that $`V_L=V_{\text{MNS}}^0`$ and $`\widehat{f}_N^{ii}=\sqrt{M_R\widehat{\kappa }_\nu ^{ii}}`$. We fix $`m_t^{\text{pole}}=175`$ GeV, $`m_b^{\text{pole}}=4.8`$ GeV and the CKM parameters as $`V_{cb}=0.04`$, $`|V_{ub}/V_{cb}|=0.08`$ and take several values of the phase parameter in the CKM matrix $`\delta _{13}`$ . We take $`\mathrm{tan}\beta =5`$ and vary other SUSY parameters $`m_0`$, $`M_0`$, $`A_0`$ and the sign of $`\mu `$. Various phenomenological constraints from SUSY particles search are included (for detail see ). We also impose $`2\times 10^4<\text{B}(bs\gamma )<4.5\times 10^4`$ in the following analysis.
Let us first discuss $`\text{B}(\mu e\gamma )`$ and $`\epsilon _K`$ which turn out to be strong constraints on the parameter space in this model. Fig. 1 shows the correlation between $`\text{B}(\mu e\gamma )`$ and $`\text{B}(\tau \mu \gamma )`$ for the neutrino parameter set (i) and (ii) for $`\delta _{13}=60^{}`$. We can see that the $`\text{B}(\mu e\gamma )`$ becomes a very strong constraint for the case (ii), which is a reflection of the large 1–2 mixing in the $`V_{\text{MNS}}`$ matrix. By requiring $`\text{B}(\mu e\gamma )<1.2\times 10^{11}`$ , $`\text{B}(\tau \mu \gamma )`$ becomes less than $`10^8`$ for the case (ii) whereas it can be close to the present experimental bound ($`1.1\times 10^6`$ ) for the case (i). We also calculated $`\text{B}(\tau e\gamma )`$ which turns out to be smaller than $`3\times 10^{12}`$ in both cases. The constraint from $`\epsilon _K`$ depends on the parameter $`\delta _{13}`$. After imposing the $`\text{B}(\mu e\gamma )`$ constraint, $`\epsilon _K`$ can be enhanced by 50% for the case (i) and by a factor of 2 for the case (ii). This means that compared to favorable values in the SM ($`50^{}<\delta _{13}<90^{}`$), a smaller value of $`\delta _{13}`$ is allowed due to the extra contributions.
The upper part of Fig. 2 shows a correlation between $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ and $`\text{B}(\tau \mu \gamma )`$ for case (i) and $`\delta _{13}=60^{}`$. Here we imposed the constraints from $`\text{B}(\mu e\gamma )`$ and $`\epsilon _K`$. We also imposed the constraint from $`\mathrm{\Delta }m_d`$ itself though the deviation of this quantity from the SM value is within 5%. For the theoretical uncertainties we allow $`\pm 15`$% difference for $`\epsilon _K`$ and $`\pm 40`$% for $`\mathrm{\Delta }m_d`$. For $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ we fix the hadronic parameters as $`f_{B_s}/f_{B_d}=1.17`$ and $`B_{B_s}/B_{B_d}=1`$. We can see that $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ can be enhanced up to 30% compared to the SM prediction. This feature is quite different from the minimal supergravity model without the GUT and right-handed neutrino interactions where $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ is almost the same as the SM value. $`A_t`$ for the same parameter set is shown as a function of $`\text{B}(\tau \mu \gamma )`$ in the lower part of Fig. 2. We can see that $`|A_t|`$ can be close to 30% when $`\text{B}(\tau \mu \gamma )`$ is larger than $`10^8`$. The large asymmetry arises because the renormalization effect due to $`f_N`$ induces sizable contribution to $`c_7^{}`$ through gluino–$`\stackrel{~}{d}_R`$ loop diagrams. The corresponding figure to Fig. 2 for the case (ii) shows that $`\text{B}(\tau \mu \gamma )`$ is cut off below $`10^8`$ and the maximal deviation of $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ from SM is within 6% and $`|A_t|`$ becomes at most 6%.
In Fig. 3 we show $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ for several values of $`\delta _{13}`$ for the case (i) and (ii). In these figures we impose $`\text{B}(\mu e\gamma )`$ constraint. Thin vertical lines correspond to the case without the experimental constraints from $`\epsilon _K`$, $`\mathrm{\Delta }m_d`$ and the lower bound for $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ , and the thick lines are allowed range with these constraints. The allowed range of the SM is also shown in these figures. Because the new contributions to $`B_d`$$`\overline{B}_d`$ amplitude is small, the time-dependent CP asymmetry of $`BJ/\psi K_S`$ in this model is essentially the same as the SM value, therefore we can obtain information on $`\delta _{13}`$ once this asymmetry is measured experimentally. For example the asymmetry of $`BJ/\psi K_S`$ mode is 0.4 for $`\delta _{13}=25^{}`$ and 0.65 for $`\delta _{13}=75^{}`$. This figure means the possible deviation from the SM may be seen in both cases once the CP asymmetry of $`BJ/\psi K_S`$ mode and $`\mathrm{\Delta }m_s/\mathrm{\Delta }m_d`$ are measured.
In our example we took $`M_R`$ which corresponds to the upper bound of $`f_N`$ because a larger $`M_R`$ would lead to the blow-up of $`f_N`$ below the Planck scale. If we take a lower value of $`M_R`$ the flavor changing amplitudes essentially scale as $`M_R`$.
Finally we would like to comment on genarizations of our result. Firstly for a large $`\mathrm{tan}\beta `$ case, the constraint from $`\text{B}(\mu e\gamma )`$ becomes stronger. As a result we cannot see any deviation from the SM for the large mixing case in the figures corresponding to Fig. 2 and 3 for $`\mathrm{tan}\beta =30`$, whereas the result is similar for the small mixing case. Secondly we considered the vacuum oscillation case. We see that the pattern of the deviation from the SM is similar to the small mixing case because the effect of the flavor mixing in 1–2 generations turns out to be suppressed by the degeneracy of the two light neutrino masses. In this letter we took the case where $`M_\nu `$ is proportional to a unit matrix. When we consider a more general case, $`V_L`$ is not necessarily equal to $`V_{\text{MNS}}^0`$. In addition, the superpotential (1) does not lead to the realistic fermion mass relation especially for the first and the second generations. In order to solve this problem we may have to introduce more free parameters. Although the precise values of the predictions depend on the detail of the model in question, the possible new physics signals such as $`\text{B}(\tau \mu \gamma )`$, $`B_s`$$`\overline{B}_s`$ mixing and $`A_t`$ may be expected as long as some of the neutrino Yukawa coupling constants are large. Because these signals provide quite different signatures compared to the SM and the minimal supergravity model without GUT and right-handed neutrino interactions, future experiments in $`B`$ physics and LFV can provide us important clues on the interactions at very high energy scale.
S. B. would like to thank KOSEF for financial support. The work of Y. O. was supported in part by the Grant-in-Aid of the Ministry of Education, Science, Sports and Culture, Government of Japan (No. 09640381), Priority area “Supersymmetry and Unified Theory of Elementary Particles” (No. 707), and “Physics of CP Violation” (No. 09246105).
|
warning/0002/astro-ph0002067.html
|
ar5iv
|
text
|
# Fast Reconnection of Magnetic Fields in Turbulent Fluids
## 1 Flux Freezing & Reconnection
Plasma conductivity is high for most astrophysical circumstances. This suggests that “flux freezing”, where magnetic field lines move with the local fluid elements, is usually a good approximation within astrophysical magnetohydrodynamics (MHD). The coefficient of magnetic field diffusivity in a fully ionized plasma is $`\eta =c^2/(4\pi \sigma )=10^{13}T^{3/2}`$ s<sup>-1</sup> cm<sup>2</sup> s<sup>-1</sup>, where $`\sigma =10^7T^{3/2}`$ s<sup>-1</sup> is the plasma conductivity and $`T`$ is the electron temperature. The characteristic time for field diffusion through a plasma slab of size $`y`$ is $`y^2/\eta `$, which is large for any “astrophysical” $`y`$.
What happens when magnetic field lines intersect? Do they deform each other and bounce back or they do change their topology? This is the central question of the theory of magnetic reconnection. In fact, the whole dynamics of magnetized fluids and the back-reaction of the magnetic field depends on the answer.
Magnetic reconnection is a long standing problem in theoretical MHD. This problem is closely related to the hotly debated issue of the magnetic dynamo (see Parker 1979; Moffatt 1978; Krause & Radler 1980). Indeed, it is impossible to understand the amplification of large scale magnetic fields without a knowledge of the mobility and reconnection of magnetic fields. Dynamo action invokes a constantly changing magnetic field topology<sup>1</sup><sup>1</sup>1Merely winding up a magnetic field can increase the magnetic field energy, but cannot increase the magnetic field flux. We understand dynamos in the latter sense. The Zel’dovich “fast” dynamo (Vainshtein & Zel’dovich 1972) also invokes reconnection for continuous dynamo action (Vainshtein 1970). and this requires efficient reconnection despite very slow Ohmic diffusion rates.
## 2 The Sweet-Parker Scheme and its Modifications
The literature on magnetic reconnection is rich and vast (see e.g., Biskamp 1993 and references therein). We start by discussing a robust scheme proposed by Sweet and Parker (Parker 1957; Sweet 1958). In this scheme oppositely directed magnetic fields are brought into contact over a region of $`L_x`$ size (see Fig. 1). The diffusion of magnetic field takes place over the vertical scale $`\mathrm{\Delta }`$ which is related to the Ohmic diffusivity by $`\eta V_r\mathrm{\Delta }`$, where $`V_r`$ is the velocity at which magnetic field lines can get into contact with each other and reconnect. Given some fixed $`\eta `$ one may naively hope to obtain fast reconnection by decreasing $`\mathrm{\Delta }`$. However, this is not possible. An additional constraint posed by mass conservation must be satisfied. The plasma initially entrained on the magnetic field lines must be removed from the reconnection zone. In the Sweet-Parker scheme this means a bulk outflow through a layer with a thickness of $`\mathrm{\Delta }`$. In other words, the entrained mass must be ejected, i.e., $`\rho V_rL_x=\rho ^{}V_A\mathrm{\Delta }`$, where it is assumed that the outflow occurs at the Alfvén velocity. Ignoring the effects of compressibility, then $`\rho =\rho ^{}`$ and the resulting reconnection velocity allowed by Ohmic diffusivity and the mass constraint is $`V_rV_A_L^{1/2}`$, where $`_L^{1/2}=(\eta /V_AL_x)^{1/2}`$ is the Lundquist number. This is a very large number in astrophysical contexts (as large as $`10^{20}`$ for the Galaxy) so that the Sweet-Parker reconnection rate is negligible.
It is well known that using the Sweet-Parker reconnection rate it is impossible to explain solar flares and it is impossible to reconcile dynamo predictions with observations. Are there any ways to increase the reconnection rate? In general, we can divide schemes for fast reconnection into those which alter the microscopic resistivity, broadening the current sheet, and those which change the global geometry, thereby reducing $`L_x`$. An example of the latter is the suggestion by Petschek (1964) that reconnecting magnetic fields would tend to form structures whose typical size in all directions is determined by the resistivity (‘X-point’ reconnection). This results in a reconnection speed of order $`V_A/\mathrm{ln}_L`$. However, attempts to produce such structures in simulations of reconnection have been disappointing (Biskamp 1984, 1986). In numerical simulations the X-point region tends to collapse towards the Sweet-Parker geometry as the Lundquist number becomes large (Biskamp 1996; Wang, Ma, & Bhattacharjee 1996). One way to understand this collapse is to consider perturbations of the original X-point geometry. In order to maintain this geometry reconnection has to be fast, which requires shocks in the original (Petschek) version of this model. These shocks are, in turn, supported by the flows driven by fast reconnection, and fade if $`L_x`$ increases. Naturally, the dynamical range for which the existence of such shocks is possible depends on the Lundquist number and shrinks when fluid conductivity increases. The apparent conclusion is that at least in the collisional regime reconnection occurs through narrow current sheets.
In the collisionless regime the width of the current sheets may be determined by the ion cyclotron (or Larmor) radius $`r_c`$ (Parker 1979) or by the ion skin depth (Ma & Bhattacharjee 1996; Biskamp, Schwarz, & Drake 1997; Shay et al. 1998) which differs from the former by the ratio of $`V_A`$ to ion thermal velocity. In laboratory conditions this often leads to a current sheet thickness which is much larger than expected (‘anomalous resistivity’). However, this effect is not likely to be important in the interstellar medium. The thickness of the current sheet $`\mathrm{\Delta }`$ scales in the Sweet-Parker scheme as $`L_x^{1/2}`$. Therefore, for a sufficiently large $`L_x`$ the natural Sweet-Parker sheet thickness becomes larger than the thickness entailed by anomalous effects. Note that the ion Larmor radius $`r_c`$ in an interstellar magnetic field is about a hundred kilometers. One cannot really hope to squeeze quickly the matter from many parsecs through a slot of this size!
One may invoke anomalous resistivity to stabilize the X-point reconnection for collisionless plasma. For instance, Shay et al. (1998) found that the reconnection speed in their simulations was independent of $`L_x`$, which would suggest that something like Petschek reconnection emerges in the collisionless regime. However, their dynamic range was small and the ion ejection velocity increased with $`L_x`$, with maximum speeds approaching $`V_A`$ for their largest values of $`L_x`$. Assuming that $`V_A`$ is an upper limit on ion ejection speeds we may expect a qualitative change in the scaling behavior of their simulations at slightly larger values of $`L_x`$. One may expect the generic problems intrinsic to X-point reconnection to persist for large $`_L`$.
If neither anomalous resistivity or/and X-point reconnection work, are there any other ways to account for fast reconnection? Can reconnection speeds be substantially enhanced if the plasma coupling with magnetic field is imperfect? This is the case in the presence of Bohm diffusion, which is a process that is observed in laboratory plasma but lacks a good theoretical explanation. Its characteristic feature is that ions appear to scatter about once per Larmor precession period. The resulting particle diffusion destroys the ‘frozen-in’ condition and allows significant larger magnetic field line diffusion. The effective diffusivity of magnetic field lines is $`\eta _{\mathrm{Bohm}}V_Ar_c`$ (see Lazarian & Vishniac 1999, henceforth LV99) which is a large increase over Ohmic resistivity. The major shortcoming of this idea is that it is unclear at all whether the concept of Bohm diffusion is applicable to astrophysical circumstances. Moreover, we note that even if we make this substitution, it can produce fast reconnection, of order $`V_A`$, only if $`r_cL_x`$. It therefore fails as an explanation for fast reconnection for the same reason that anomalous resistivity does.
Matter may also diffuse perpendicular to magnetic field lines if the plasma is partially ionized. Since neutrals are not directly affected by magnetic field lines the neutral outflow layer may be much broader than the $`\mathrm{\Delta }`$ determined by Ohmic diffusivity. The trouble with ambipolar diffusion is that ions and electrons are left in the reconnection zone. As a result, the reconnection speed is determined by a slow recombination process. Calculations in Vishniac & Lazarian (1999) show that the ambipolar reconnection rates are slow unless the ionization ratio is extremely low.
Can plasma instabilities increase the reconnection rate? The narrow current sheet formed during Sweet-Parker reconnection is unstable to tearing modes. A study of tearing modes in LV99 showed that an increase over the Sweet-Parker rates is possible and the resulting reconnection rates may be as high as $`V_A_L^{3/10}`$. However, these speeds are still exceedingly small in view of the enormous values of $`_L`$ encountered in astrophysical plasmas. Below we discuss a different approach to the problem of rapid reconnection i.e., we consider magnetic reconnection<sup>2</sup><sup>2</sup>2The mode of reconnection discussed here is sometimes is called free reconnection as opposed to forced reconnection. Wang et al. (1992) define free reconnection as a process caused by a nonideal instability driven by the free energy stored in an equilibrium. If the equilibrium is stable, reconnection can be forced if a perturbation is applied externally. in the presence of a weak random field component.
## 3 Turbulent Reconnection
### 3.1 Reconnection in Two and Three Dimensions
Two idealizations were used in the preceding discussion. First, we considered the process in only two dimensions. Second, we assumed that the magnetized plasma is laminar. The Sweet-Parker scheme can easily be extended into three dimensions. Indeed, one can always take a cross-section of the reconnection region such that the shared component of the two magnetic fields is perpendicular to the cross-section. In terms of the mathematics nothing changes, but the outflow velocity becomes a fraction of the total $`V_A`$ and the shared component of the magnetic field will have to be ejected together with the plasma. This result has motivated researchers to do most of their calculations in 2D, which has obvious advantages for both analytical and numerical investigations.
However, physics in two and three dimensions is very different. For instance, in two dimensions the properties of turbulence are very different. In LV99 we considered three dimensional reconnection in a turbulent magnetized fluid and showed that reconnection is fast. This result cannot be obtained by considering two dimensional turbulent reconnection (cf. Matthaeus & Lamkin 1986). Below we briefly discuss the idea of turbulent reconnection, while the full treatment of the problem is given in LV99.
### 3.2 A Model of Turbulent Reconnection
MHD turbulence guarantees the presence of a stochastic field component, although its amplitude and structure clearly depends on the model we adopt for MHD turbulence, as well as the specific environment of the field. We consider the case in which there exists a large scale, well-ordered magnetic field, of the kind that is normally used as a starting point for discussions of reconnection. This field may, or may not, be ordered on the largest conceivable scales. However, we will consider scales smaller than the typical radius of curvature of the magnetic field lines, or alternatively, scales below the peak in the power spectrum of the magnetic field, so that the direction of the unperturbed magnetic field is a reasonably well defined concept. In addition, we expect that the field has some small scale ‘wandering’ of the field lines. On any given scale the typical angle by which field lines differ from their neighbors is $`\varphi 1`$, and this angle persists for a distance along the field lines $`\lambda _{}`$ with a correlation distance $`\lambda _{}`$ across field lines (see Fig. 1).
The modification of the mass conservation constraint in the presence of a stochastic magnetic field component is self-evident. Instead of being squeezed from a layer whose width is determined by Ohmic diffusion, the plasma may diffuse through a much broader layer, $`L_yy^2^{1/2}`$ (see Fig. 1), determined by the diffusion of magnetic field lines. This suggests an upper limit on the reconnection speed of $`V_A(y^2^{1/2}/L_x)`$. This will be the actual speed of reconnection; the progress of reconnection in the current sheet itself does not impose a smaller limit. The value of $`y^2^{1/2}`$ can be determined once a particular model of turbulence is adopted, but it is obvious from the very beginning that this value is determined by field wandering rather than Ohmic diffusion as in the Sweet-Parker case.
What about limits on the speed of reconnection that arise from considering the structure of the current sheet? In the presence of a stochastic field component, magnetic reconnection dissipates field lines not over their entire length $`L_x`$ but only over a scale $`\lambda _{}L_x`$ (see Fig. 1), which is the scale over which a magnetic field line deviates from its original direction by the thickness of the Ohmic diffusion layer $`\lambda _{}^1\eta /V_{rec,local}`$. If the angle $`\varphi `$ of field deviation does not depend on the scale, the local reconnection velocity would be $`V_A\varphi `$ and would not depend on resistivity. In LV99 we claimed that $`\varphi `$ does depend on scale. Therefore, the local reconnection rate $`V_{rec,local}`$ is given by the usual Sweet-Parker formula but with $`\lambda _{}`$ instead of $`L_x`$, i.e. $`V_{rec,local}V_A(V_A\lambda _{}/\eta )^{1/2}`$. It is obvious from Figure 1 that $`L_x/\lambda _{}`$ magnetic field lines will undergo reconnection simultaneously (compared to a one by one line reconnection process for the Sweet-Parker scheme). Thus, the overall reconnection rate may be as large as $`V_{rec,global}V_A(L_x/\lambda _{})(V_A\lambda _{}/\eta )^{1/2}`$. Whether or not this limit is important depends on the value of $`\lambda _{}`$.
The relevant values of $`\lambda _{}`$ and $`y^2^{1/2}`$ depend on the magnetic field statistics. This calculation was performed in LV99 using the Goldreich-Sridhar (1995) model of MHD turbulence, the Kraichnan model (Iroshnikov 1963; Kraichnan 1965) and for MHD turbulence with an arbitrary spectrum. In all the cases the upper limit on $`V_{rec,global}`$ was greater than $`V_A`$, so that the diffusive wandering of field lines imposed the relevant limit on reconnection speeds. For instance, for the Goldreich-Sridhar (1995) spectrum the upper limit on the reconnection speed was
$$V_{r,up}=V_A\mathrm{min}[\left(\frac{L_x}{l}\right)^{\frac{1}{2}},\left(\frac{l}{L_x}\right)^{\frac{1}{2}}]\left(\frac{v_l}{V_A}\right)^2,$$
(1)
where $`l`$ and $`v_l`$ are the energy injection scale and turbulent velocity at this scale respectively. In LV99 we also considered other processes that can impede reconnection and find that they are less restrictive. For instance, the tangle of reconnection field lines crossing the current sheet will need to reconnect repeatedly before individual flux elements can leave the current sheet behind. The rate at which this occurs can be estimated by assuming that it constitutes the real bottleneck in reconnection events, and then analyzing each flux element reconnection as part of a self-similar system of such events. This turns out to limit reconnection to speeds less than $`V_A`$, which is obviously true regardless. As a result, we showed in LV99 that equation (1) is not only an upper limit, but is the best estimate of the speed of reconnection.
Naturally, when turbulence is negligible, i.e. $`v_l0`$, the field line wandering is limited to the Sweet-Parker current sheet and the Sweet-Parker reconnection scheme takes over. However, in practical terms this means an artificially low level of turbulence that should not be expected in realistic astrophysical environments. Moreover, the release of energy due to reconnection, at any speed, will contribute to the turbulent cascade of energy and help drive the reconnection speed upward.
We stress that the enhanced reconnection efficiency in turbulent fluids is only present if 3D reconnection is considered. In this case ohmic diffusivity fails to constrain the reconnection process as many field lines simultaneously enter the reconnection region. The number of lines that can do this increases with the decrease of resistivity and this increase overcomes the slow rates of reconnection of individual field lines. It is impossible to achieve a similar enhancement in 2D (see Zweibel 1998) since field lines can not cross each other.
### 3.3 Energy Dissipation and its Consequences
It is usually believed that rapid reconnection in the limit of vanishing resistivity implies a current singularity (Park, Monticello, & White 1984). Our model does not require such singularities. Indeed, they show that while the amount of Ohmic dissipation tends to 0 as $`\eta 0`$, the smallest scale of the magnetic field’s stochastic component decreases so that the rate of the flux reconnection does not decrease.
The turbulent reconnection process assumes that only small segments of magnetic field lines enter the reconnection zone and are subjected to ohmic annihilation. Thus, only a small fraction of the magnetic energy, proportional to $`_L^{2/5}`$ (LV99), is released in the form of ohmic heat. The rest of the energy is released in the form of non-linear Alfvén waves that are generated as reconnected magnetic field lines straighten up.
Naturally, the low efficiency of electron heating is of little interest when ion and electron temperatures are tightly coupled. When this is not the case the LV99 model for reconnection has some interesting consequences. As an example, we may consider advective accretion flows (ADAFs), following the general description given in Narayan and Yi (1995) in which advective flows can be geometrically thick and optically thin with a small fraction of the dissipation going into electron heating. If, as expected, the magnetic pressure is comparable to the gas pressure in these systems, then a large fraction of the orbital energy dissipation occurs through reconnection events. If a large fraction of this energy goes into electron heating (cf. Bisnovatyi-Kogan & Lovelace 1997) then the observational arguments in favor of ADAFs are largely invalidated. The results in LV99 suggest that reconnection, by itself, will not result in channeling more than a small fraction of the energy into electron heating. Of course, the fate of energy dumped into a turbulent cascade in a collisionless magnetized plasma then becomes a critical issue.
We also note that observations of solar flaring seem to show that reconnection events start from some limited volume and spread as though a chain reaction from the initial reconnection region initiated a dramatic change in the magnetic field properties. Indeed, solar flaring happens as if the resistivity of plasma were increasing dramatically as plasma turbulence grows (see Dere 1996 and references therein). In our picture this is a consequence of the increased stochasticity of the field lines rather than any change in the local resistivity. The change in magnetic field topology that follows localized reconnection provides the energy necessary to feed a turbulent cascade in neighboring regions. This kind of nonlinear feedback is also seen in the geomagnetic tail, where it has prompted the suggestion that reconnection is mediated by some kind of nonlinear instability built around modes with very small $`k_{}`$ (cf. Chang 1998 and references therein). The most detailed exploration of nonlinear feedback can be found in the work of Matthaeus and Lamkin (1986), who showed that instabilities in narrow current sheets can sustain broadband turbulence in two dimensional simulations. Although our model is quite different, and relies on the three dimensional wandering of field lines to sustain fast reconnection, we note that the concept of a self-excited disturbance does carry over and may describe the evolution of reconnection between volumes with initially smooth magnetic fields.
## 4 Implications
### 4.1 Turbulent Reconnection and Turbulent Diffusivity
We would like to stress that in introducing turbulent reconnection we do not intend to revive the concept of “turbulent diffusivity” as used in dynamo theories (Parker 1979). In order to explain why astrophysical magnetic fields do not reverse on very small scales, researchers have usually appealed to an ad hoc diffusivity which is many orders of magnitude greater than the ohmic diffusivity. This diffusivity is assumed to be roughly equal to the local turbulent diffusion coefficient. While superficially reasonable, this choice implies that a dynamically significant magnetic field diffuses through a highly conducting plasma in much the same way as a passive tracer. This is referred to as turbulent diffusivity and denoted $`\eta _t`$, as opposed to the Ohmic diffusivity $`\eta `$. Its name suggests that turbulent motions subject the field to kinematic swirling and mixing. As the field becomes intermittent and intermixed it can be assumed to undergo dissipation at arbitrarily high speeds.
Parker (1992) showed convincingly that the concept of turbulent diffusion is ill-founded. He pointed out that turbulent motions are strongly constrained by magnetic tension and large scale magnetic fields prevent hydrodynamic motions from mixing magnetic field regions of opposing polarity unless they are precisely anti-parallel. However, results in LV99 show that the mobility of a magnetic field in a turbulent fluid is indeed enhanced. For instance, due to fast reconnection the magnetic field will not form long lasting knots. Moreover, the magnetic field can be expected to straighten itself and remove small scale reversals as required, in a qualitative sense, by dynamo theory. Nevertheless, the underlying physics of this process is very different from what is usually meant by “turbulent diffusivity”. Within the turbulent diffusivity paradigm, magnetic fields of different polarity were believed to filament and intermix on very small scales while reconnection proceeded slowly. On the contrary, we have shown in LV99 that the global speed of reconnection is fast if a moderate degree of magnetic field line wandering is allowed. The latter, unlike the former, corresponds to a realistic picture of MHD turbulence and does not entail prohibitively high magnetic field energies at small scales.
On the other hand, the diffusion of particles through a magnetized plasma is greatly enhanced when the field is mildly stochastic. There is an analogy between the reconnection problem and the diffusion of cosmic rays (Barghouty & Jokipii 1996). In both cases charged particles follow magnetic field lines and in both cases the wandering of the magnetic field lines leads to efficient diffusion.
### 4.2 Dynamos
There is a general belief that magnetic dynamos operate in stars, galaxies (Parker 1979) and accretion disks (Balbus & Hawley 1998). In stars, and in many accretion disks, the plasma has a high $`\beta `$, that is the average plasma pressure is higher than the average magnetic pressure. In such situations the high diffusivity of the magnetic field can be explained by concentrating flux in tubes<sup>3</sup><sup>3</sup>3Note that flux tube formation requires initially high reconnection rates. Therefore, the flux tubes by themselves provide only a partial solution to the problem. (Vishniac 1995a,b). This trick does not work in the disks of galaxies, where the magnetic field is mostly diffuse (compare Subramanian 1998) and ambipolar diffusion impedes the formation of flux tubes (Lazarian & Vishniac 1996). This is the situation where our current treatment of magnetic reconnection is most relevant. However, our results suggest that magnetic reconnection proceeds regardless and that the concentration of magnetic flux in flux tubes via turbulent pumping is not a necessary requirement for successful dynamos in stars and accretion discs.
To enable sustainable dynamo action and, for example, generate a galactic magnetic field, it is necessary to reconnect and rearrange magnetic flux on a scale similar to a galactic disc thickness within roughly a galactic turnover time ($`10^8`$ years). This implies that reconnection must occur at a substantial fraction of the Alfvén velocity. The preceding arguments indicate that such reconnection velocities should be attainable if we allow for a realistic magnetic field structure, one that includes both random and regular fields.
One of the arguments against traditional mean-field dynamo theory is that the rapid generation of small scale magnetic fields suppresses further dynamo action (e.g., Kulsrud & Anderson 1992). Our results thus far show that a random magnetic field enhances reconnection by enabling more efficient diffusion of matter from the reconnection layer. This suggests that the existence of small scale magnetic turbulence is a prerequisite for a successful large scale dynamo. In other words, we are arguing for the existence of a kind of negative feed-back. If the magnetic field is too smooth, reconnection speeds decrease and the field becomes more tangled. If the field is extremely chaotic, reconnection speeds increase, making the field smoother. We note that it is common knowledge that magnetic reconnection can sometimes be quick and sometimes be slow. For instance, the existence of bundles of flux tubes of opposite polarity in the solar convection zone indicates that reconnection can be very slow. At the same time, solar flaring suggests very rapid reconnection rates.
Our results show that in the presence of MHD turbulence magnetic reconnection is fast, and this in turn allows the possibility of ‘fast’ dynamos in astrophysics (see the discussion of the fast dynamo in Parker 1992).
Finally, we have assumed that we are dealing with a strong magnetic field, where motions that tend to mix field lines of different orientations are largely suppressed. The galactic magnetic field is usually taken to have grown via dynamo action from some extremely weak seed field (cf. Zel’dovich, Ruzmaikin, & Sokoloff 1983; Lazarian 1992 and references contained therein). When the field is weak it can be moved as a passive scalar and its spectrum will mimic that of Kolmogorov turbulence. The difference between $`\lambda _{}`$ and $`\lambda _{}`$ vanishes, the field becomes tangled on small scales, and $`V_{rec,local}`$ becomes of the order of $`V_A`$. Of course, in this stage of evolution $`V_A`$ may be very small. However, on such small scales $`V_A`$ will grow to equipartition with the turbulent velocities on the turn over time of the small eddies. The enhancement of reconnection as $`V_A`$ increases accelerates the inverse cascade as small magnetic loops merge to form larger ones.
## 5 Discussion
It is not possible to understand the dynamics of magnetized astrophysical plasmas without understanding how magnetic fields reconnect. Here we have compared traditional approaches to the problem of magnetic reconnection and a new approach that includes the presence of turbulence in the magnetized plasma.
One of the more striking aspects of our result is that the global reconnection speed is relatively insensitive to the actual physics of reconnection. Equation (1) only depends on the nature of the turbulent cascade. Although this conclusion was reached by invoking a particular model for the strong turbulent cascade, we showed in LV99 that any sensible model gives qualitatively similar results. One may say that the conclusion that reconnection is fast, even when the local reconnection speed is slow, represents a triumph of global geometry over the slow pace of ohmic diffusion. In the end, reconnection can be fast because if we consider any particular flux element inside the contact volume, assumed to be of order $`L_x^3`$, the fraction of the flux element that actually undergoes microscopic reconnection vanishes as the resistivity goes to zero.
The new model of fast turbulent reconnection changes our understanding of many astrophysical processes. Firstly, it explains why dynamos do not suppress themselves through the excessive generation of magnetic noise, as some authors suggest (Kulsrud & Anderson 1992). The model also explains why reconnection may be sometimes fast and sometimes slow, as solar activity demonstrates. ADAFs and the acceleration of cosmic rays at reconnection sites are other examples of processes where a new model of reconnection should be applied.
Our results on turbulent reconnection assume that the turbulent cascade is limited by plasma resistivity. If gas is partially ionized collisions with neutrals may play an important role in damping turbulence. A study in Lazarian & Vishniac (2000) shows that for gas with low levels of ionization turbulent reconnection may be impeded as magnetic field wandering is suppressed on small scales. However, the level of suppression depends on the details of the energy injection into the turbulent cascade (see a discussion in Lazarian & Pogosyan 2000), which are far from being clear. Moreover, for very low ionization levels there will be an enhancement of the reconnection process as neutrals diffuse perpendicular to magnetic field lines. Thus, reconnection may still be an important process in the evolution of molecular clouds and in star formation.
###### Acknowledgements.
AL acknowledges valuable discussions with Chris McKee.
|
warning/0002/cond-mat0002339.html
|
ar5iv
|
text
|
# Tsallis Entropy and the transition to scaling in fragmentation
## Abstract
By using the maximum entropy principle with Tsallis entropy we obtain a fragment size distribution function which undergoes a transition to scaling. This distribution function reduces to those obtained by other authors using Shannon entropy. The treatment is easily generalisable to any process of fractioning with suitable constraints.
1.- Department of Theoretical Physics, Havana University, Habana 10400, Cuba.
2.- Department of Mathematical Sciences, Brunel University,
Uxbridge, Middlesex UB8 3PH, UK.
05.40.Fb, 24.60.-k
As a result of developments in materials science, combustion technology, geology and many other fields of research, there has been an increase of interest in the problem of fragmentation of objects. Within this general field there is a collection of papers where a transition occurs from a “classical” distribution of fragments (e.g. log-normal or Rossin-Ramler- like) to a power law distribution. This transition has not been adequately explained in terms of any general principles, although in the representation of the fragmentation process in terms of percolation on a Bethe lattice leads to a transition to a power law in the distribution of fragment sizes.
Some attempts have been made to derive the fragment size distribution function from the maximum entropy principle , subject to some constraints which mainly came from physical considerations about the fragmentation phenomena. The resulting fragment distribution function describes the distribution of sizes of the fragments in a regime in which scaling is not present.
As scaling invariably occurs when the energy of the fragmentation process is high, this suggests that the analysis is only applicable to low energies. However, the maximum entropy principle is completely universal and has an almost unlimited range of application. Consequently we would expect to be able to use it to describe the transition to scaling as the energy of the fracture grows.
The expression for the Boltzmann-Gibbs entropy $`S`$ (e.g. Shannon’s form) is given by
$$S=k\underset{i=1}{\overset{W}{}}p_ilnp_i,$$
(1)
where $`p_i`$ is the probability of finding the system in the microscopic state i, $`k`$ is Boltzmann’s constant, and $`W`$ is the total number of microstates.
This has been shown to be restricted to the domain of validity of Boltzmann-Gibbs (BG) statistics. These statistics seem to describe nature when the effective microscopic interactions and the microscopic memory are short ranged . The process of violent fractioning, like that of droplet microexplosions in combustion chambers, blasting and shock fragmentation with high energies, etc, leads to the existence of long-range correlations between all parts of the object being fragmented.
Fractioning is a paradigm of non extensivity, since the fractioning object can be considered as a collection of parts which, after division, have an entropy larger than that of their union i.e if we denote by $`A_i`$ the parts or fragments in which the object has been divided, its entropy $`S`$ obeys $`S(A_i)<_iS(A_i)`$, defining a “superextensivity” in this system . This suggests that it may be necessary to use non-extensive statistics, instead of the BG statistics. This kind of theory has already been proposed by Tsallis , who postulated a generalized form of entropy, given by
$$S_q=k\frac{1_0^{\mathrm{}}p^q(x)𝑑x}{q1}.$$
(2)
The integral runs over all admissible values of the magnitude $`x`$ and $`p(x)dx`$ is the probability of the system being in a state between $`x`$ and $`x+dx`$. This entropy can also be expressed as
$$S_q=p^q(x)ln_qp(x)𝑑x.$$
The generalized logarithm $`ln_q(p)`$ is defined in as
$$ln_q(p)=\frac{p^{1q}1}{1q},$$
(3)
where q is a real number. It is straighforward to see that $`S_qS`$ when $`q1`$, recovering BG statistics.
In this paper we use the entropy in eq.2 to consider the problem of atomization of liquid fuels.
The atomized drops in a low pressure regime follow a Nukiyama-Tanasawa-like distribution of sizes, a particular case of the Rossin-Ramler distribution . As we already pointed out in , this distribution tends to a power-law, revealing scaling as the atomization pressure grows. Incidentally, this is essentially the same behavior as that reported in experiments on falling glass rods , mercury drops, and on blasting oil drops . So, let us maximise $`\frac{S_q}{k}`$ given by eq.2. If we denote the volume of a drop by $`V`$ and some typical volume characteristic of the distribution by $`V_m`$, we can define a dimensionless volume $`v=\frac{V}{V_m}`$. Then, the constraints to impose are
$$_0^{\mathrm{}}p(v)𝑑v=1,$$
(4)
i.e., the normalization condition. The other condition to be imposed is mass conservation. But as the system is finite, this condition will lead to a very sharp decay in the asymptotic behavior of the droplet size distribution function (DSDF) for large sizes of the fragments. Consequently, we will impose a more general condition, like the “q-conservation” of the mass, in the form:
$$_0^{\mathrm{}}vp^q(v)𝑑v=1,$$
(5)
which reduces to the “classical” mass conservation when $`q=1`$.
Equations 4 and 5 are the constraints to impose in order to derive the DSDF using the method of Lagrange multipliers by means of the construction of the function:
$$L(p_i;\alpha _1;\alpha _2)=S_q\alpha _1_0^{\mathrm{}}p(v)𝑑v+\alpha _2_0^{\mathrm{}}p^q(v)v𝑑v$$
(6)
The Lagrange multipliers $`\alpha _1`$ and $`\alpha _2`$ are determined from eqs.4 and 5. The extremization of $`L(p_i;\alpha _1;\alpha _2)`$ leads to:
$$p(v)dv=C(1+(q1)\alpha _2v)^{\frac{1}{q1}}dv$$
(7)
where the constant $`C`$ is given by
$$C=\frac{q1}{q}\alpha _1^{\frac{1}{q1}}.$$
This is a DSDF expressed as a function of the volume of the droplets. It is convenient to formulate the problem in terms of a DSDF as a function of the dimensionless radius of the droplets $`r=v^{1/3}`$. Then the probability density is:
$$f_q(r)=3Cr^2[1+(q1)\alpha _2r^3)]^{\frac{1}{q1}}.$$
(8)
To obtain the DSDF the range of admissible values of $`q`$ is $`1<q<2`$. This range of values of $`q`$ also guarantees the adequate power law behavior of eq.8, since its asymptotic behavior for large $`r`$ is
$$f_q(r)\frac{1}{r^{\alpha +1}},$$
(9)
where $`\alpha `$ is the generic power law exponent, $`\alpha =3\frac{2q}{q1}`$. Also, if $`q1`$, eq.8 leads to:
$$f(r)=3r^2exp(r^3),$$
(10)
which is the a Nukiyama-Tanasawa DSDF, a particular case of the Rossin- Ramler distribution. This distribution has been previously obtained in .
Then, the DSDF that we have obtained reproduces the actual behavior of fragments in the process of breaking. It is easy to realise that the above viewpoint is applicable not only to atomization, but to any process of fractioning.
For a given regime of breakage, generally identified as that of the lowest energy of breakage, the fragment distribution function can be deduced through the maximum entropy principle using BG statistics. This low energy regime of breakage is such that the correlations between the different parts of the object are short-ranged. As the energy of breakage increases, long-range correlations become more and more important, which makes it necessary to introduce the Tsallis entropy as a generalization of the Shannon entropy.
Thus, we have confirmed that BG statistics cannot be applied to all fragmentation regimes and Tsallis entropy can be used to describe the transition into scaling. In this respect, the parameter q, which determines the “degree of nonextensivity” of the statistics, can be related to an effective temperature of breakage. As far as we know, this is the first formulation in terms of general principles that leads to a DSDF which exhibits a transition to scaling.
This work was partially supported by the “Alma Mater” contest, Havana University. One of us (O.S) is grateful to the Department of Mathematical Sciences of Brunel University for kind hospitality and the Royal Society, London for financial support.
|
warning/0002/quant-ph0002059.html
|
ar5iv
|
text
|
# Phase measurements at the theoretical limit
## I Introduction
The phase of an electromagnetic field cannot be measured directly using linear optics and photodetectors. Rather than measuring phase directly, phase measurement schemes rely on measuring quadratures of the field and inferring the phase from these measurements. In a typical experimental implementation, the mode to be measured is passed through a 50/50 beam splitter, in order to combine it with a much stronger local oscillator field. The difference photocurrent from the two output ports of the beam splitter yields a measurement of a particular quadrature.
The standard technique for measuring a completely unknown phase is heterodyne detection, where all quadratures are sampled with equal probability. This is achieved by using a local oscillator field with a frequency slightly different from the signal’s frequency, so its phase changes linearly with respect to the phase of the signal. More accurate phase measurements can be made using the homodyne technique, where the local oscillator phase is $`\mathrm{\Phi }=\phi +\pi /2`$, with $`\phi `$ the phase of the signal. The problem with this is that it requires initial knowledge of the phase of the signal, and so is not a phase measurement in the strict sense.
To maintain the unbiased nature of heterodyne phase measurements but obtain the increased sensitivity of homodyne measurements, an adaptive dyne technique can be used Wis95c ; semiclass ; fullquan ; BerWisZha99 . Here “dyne” detection is used to mean photodetection using a strong local oscillator at a beam splitter. The idea behind adaptive phase measurement schemes is to use the information gained so far during the measurement to estimate the system phase $`\phi `$. This is then used to adjust the local oscillator phase $`\mathrm{\Phi }`$ to approximate a homodyne measurement as above.
The apparatus for performing these measurements is shown schematically in Fig. 1. The signal and a local oscillator with amplitude $`\beta `$ are combined at the beam splitter and the outputs are measured with photodetectors. The outputs from the photodetectors, $`\delta N_+`$ and $`\delta N_{}`$, are subtracted and then fed into a digital signal processor that uses these measurements to estimate the phase of the system, and adjusts the phase of the local oscillator via an electro-optic phase modulator. The signal is shown here as from a cavity with a half-silvered mirror, as this is what is considered in the theory in Sec. IV.
The signal of interest is the difference between the photocurrents at the two ports. We therefore define the signal as
$$I(t)\delta t=\frac{\delta N_+\delta N_{}}{\beta e^{t/2}}.$$
(1)
Here we have used units of time such that the decay constant of the cavity is unity. We divide by a factor of $`e^{t/2}`$ because this is the square root of the mode function for the signal. We can take account of signals with more general mode functions $`u(t)`$ in a similar way semiclass .
When we take the limit of very large local oscillator amplitude and small time intervals $`\delta t`$, we find that
$$I(v)dv=2\text{Re}(\alpha _v^Se^{i\mathrm{\Phi }(v)})dv+dW(v),$$
(2)
where $`v`$ is time scaled to the unit interval and $`\alpha _v^S`$ is the scaled mean amplitude of the system (for which the $`S`$ superscript stands). The systematic variation in the coherent amplitude with time due to the mode shape is scaled out. This scaling is explained in more detail in Sec. IV. The final term $`dW(v)`$ is an infinitesimal Wiener increment such that $`dW(v)^2=dv`$ text .
It can be shown wise96 ; fullquan that just two complex numbers are necessary to encapsulate all of the relevant information in the photocurrent record up to a given time. These are
$`A_v`$ $`={\displaystyle _0^v}I(u)e^{i\mathrm{\Phi }(u)}𝑑u,`$ (3)
$`B_v`$ $`={\displaystyle _0^v}e^{2i\mathrm{\Phi }(u)}𝑑u.`$ (4)
For convenience, we often replace $`B_v`$ by a third complex number defined in terms of $`A_v`$ and $`B_v`$,
$$C_v=A_vv+B_vA_v^{}.$$
(5)
Generally the best estimate of the phase at time $`v`$ is $`\mathrm{arg}(C_v)`$ semiclass . The subscripts are omitted for the final values ($`v=1`$).
In adaptive measurement schemes the phase of the local oscillator is generally taken to be
$$\mathrm{\Phi }(v)=\widehat{\phi }(v)+\frac{\pi }{2},$$
(6)
where $`\widehat{\phi }(v)`$ is the estimated phase of the system at time $`v`$ using the measurement results $`A_v`$ and $`C_v`$. There are a number of possible phase estimates, giving different adaptive schemes. For the mark I scheme semiclass ; fullquan , both the running phase estimate $`\widehat{\phi }(v)`$ and the final phase estimate are taken to be $`\mathrm{arg}(A_v)`$. This is better than heterodyne measurements only if the field is very weak Wis95c ; fullquan . For the mark II adaptive phase measurements semiclass ; fullquan the best phase estimate $`\mathrm{arg}(C)`$ is used at the end of the measurement, but for the intermediate phase estimate $`\mathrm{arg}(A_v)`$ is used. This is better than heterodyne measurements for all field strengths. If $`\mathrm{arg}(C_v)`$ is generally the best phase estimate, it is apparent from Eq. (5) that $`\mathrm{arg}(A_v)`$ will only be the best phase estimate if $`B_v`$ is negligible (as it is in the case of heterodyne measurements). For adaptive phase measurements $`B_v`$ does not vanish and $`\mathrm{arg}(A_v)`$ is generally a much worse phase estimate than $`\mathrm{arg}(C_v)`$.
This raises the question of why this relatively poor intermediate phase estimate is used. There are two main reasons for this: (i) it is possible to obtain direct analytic results for this case, whereas using a better intermediate phase estimate requires numerical evaluation; (ii) the apparatus required to implement this method is much simpler than that required for a better intermediate phase estimate.
Even with the relatively poor intermediate phase estimate, the mark II adaptive scheme introduces a phase variance of just $`\frac{1}{8}(\overline{n}^S)^{1.5}`$, a good improvement over the heterodyne result of $`\frac{1}{4}(\overline{n}^S)^1`$. Here $`\overline{n}^S`$ is the mean photon number of the field being measured, and the actual measured phase variance is the introduced phase variance plus the intrinsic phase variance. The intrinsic phase variance for a state of mean photon number $`\overline{n}^S`$ can be as small as of order $`(\overline{n}^S)^2`$ SumPeg90 ; BerWisZha99 . This is far smaller than the introduced phase variance, so the latter is what limits the accuracy of phase measurements. Although the mark II results are far superior to the standard result of heterodyne detection, it is still possible to improve on the mark II result, and it is shown in Ref. fullquan that a theoretical lower limit to the phase variance that is introduced by an arbitrary phase measurement scheme (based on linear optics and photodetection) is $`\frac{1}{4}\mathrm{ln}(\overline{n}^S)\times (\overline{n}^S)^2`$.
In improving on the mark II result, the obvious thing to do is to use a better intermediate phase estimate. It turns out that using the best phase estimate $`\mathrm{arg}(C_v)`$ actually gives a worse result than the mark II case, for reasons that we will explain later. The phase estimates that we consider in this paper are therefore intermediate between $`\mathrm{arg}(A_v)`$ and the best phase estimate:
$$\widehat{\phi }(v)=\mathrm{arg}(C_v^{1ϵ(v)}A_v^{ϵ(v)}).$$
(7)
It is possible to obtain a marked improvement over the mark II case by using constant values of $`ϵ`$. We show in Sec. V that a scaling of roughly $`(\overline{n}^S)^{1.68}`$ is possible. One drawback is that the value of $`ϵ`$ required depends on the photon number.
We can obtain an even better result if we allow $`ϵ`$ to have a variation in time, and we show in Sec. V that we can obtain phase estimates very close to the theoretical limit if we use
$$ϵ(v)=\frac{v^2|B_v|^2}{C_v}\sqrt{\frac{v}{1v}}.$$
(8)
This expression does not explicitly depend on the photon number. This method works best if the phase estimates are updated in discrete time steps, and the magnitude of the steps depends weakly on the photon number. A more serious problem with this method is that it tends to produce values of $`|B|`$ that are too close to 1. This means that final phase estimates with an error close to $`\pi `$ occur sufficiently frequently to make a significant contribution to the phase uncertainty. We will show how this problem can be corrected.
The paper is structured as follows. In Sec. II we rederive the ultimate theoretical limit to phase measurements of Ref. fullquan . This is necessary to understand how the improved feedback algorithm of Eq. (7) can approach the theoretical limit, as explained in Sec. III. In Sec. IV we derive the results necessary for a numerical simulation of this algorithm, and in Sec. V present the results of those simulations. The problem of infrequent results with large errors is identified in Sec. VI and a solution proposed and evaluated in Sec. VII. We conclude with a summary and discussion in Sec. VIII.
## II The theoretical limit
In order to understand how to attain the theoretical limit, we must first understand the reason for the theoretical limit. It can be shown wise96 that the probability of obtaining the results $`A`$, $`B`$ from an arbitrary (adaptive or nonadaptive) measurement is
$$P(A,B)d^2Ad^2B=\mathrm{Tr}[\rho G(A,B)]d^2Ad^2B,$$
(9)
where $`\rho `$ is the state of the mode being measured. Here $`G(A,B)`$ is the POM (probability operator measure) for the measurement, and is given by
$$G(A,B)=Q(A,B)|\stackrel{~}{\psi }(A,B)\stackrel{~}{\psi }(A,B)|,$$
(10)
where $`Q(A,B)`$ is what the probability distribution $`P(A,B)`$ would be if $`\rho `$ were the vacuum state $`|00|`$, and $`|\stackrel{~}{\psi }(A,B)`$ is an unnormalized ket defined by
$$|\stackrel{~}{\psi }(A,B)=\mathrm{exp}\left[\frac{1}{2}B(a^{})^2Aa^{}\right]|0.$$
(11)
This is proportional to a squeezed state WalMil94 :
$$\mathrm{exp}[\frac{1}{2}B(a^{})^2Aa^{}]|0=\left(1|B|^2\right)^{1/4}\mathrm{exp}(A\alpha ^{}/2)|\alpha ,\xi ,$$
(12)
where
$$|\alpha ,\xi =\mathrm{exp}(\alpha a^{}\alpha ^{}a)\mathrm{exp}\left[\frac{1}{2}\xi ^{}a^2\frac{1}{2}\xi (a^{})^2\right]|0,$$
(13)
and the squeezing parameters are
$`\alpha `$ $`={\displaystyle \frac{A+BA^{}}{1|B|^2}},`$ (14)
$`\xi `$ $`={\displaystyle \frac{B\text{atanh}|B|}{|B|}}.`$ (15)
where atanh is the inverse hyperbolic tan function. In terms of these the POM is given by
$$G(A,B)=Q^{}(A,B)|\alpha ,\xi \alpha ,\xi |,$$
(16)
where
$$Q^{}(A,B)=Q(A,B)\left(1|B|^2\right)^{1/2}\mathrm{exp}\left[\text{Re}\left(A\alpha ^{}\right)\right].$$
(17)
If the system state is pure, $`\rho =|\psi \psi |`$ and the probability distribution is given by
$$P(A,B)=Q^{}(A,B)|\psi |\alpha ,\xi |^2.$$
(18)
For an unbiased measurement scheme the probability distribution for the phase resulting from this equation depends entirely on the inner product between the two states, and not on $`Q^{}(A,B)`$. To see this, note first that if the measurement is unbiased the vacuum probability distribution $`Q(A,B)`$ will be independent of the phase. Second, for the squeezed state $`|\alpha ,\xi `$, $`\xi \alpha ^{}/\alpha `$ is independent of the phase $`\mathrm{arg}(\alpha )`$. This in turn means that $`BA^{}/A`$ is independent of the phase. Since
$$A\alpha ^{}=(1+BA^{}/A)^{}\frac{|A|^2}{1|B|^2},$$
(19)
$`A\alpha ^{}`$ and therefore $`Q^{}(A,B)`$ are independent of the phase.
Since the probability distribution for the phase depends on the inner product between the two states, the variance in the measured phase will approximately be the sum of the intrinsic phase variance and the phase variance of the squeezed state $`|\alpha ,\xi `$. The maximum overlap between the states will be when the squeezed state has about the same photon number as the input state. This means that the theoretical limit to the phase variance that is introduced by the measurement is the phase variance of the squeezed state that has the same photon number as the input state and has been optimized for minimum intrinsic phase variance. Since the phase variance of a squeezed state optimized for minimum intrinsic phase variance is $`\mathrm{ln}\overline{n}/(4\overline{n}^2)`$ in the limit of large $`\overline{n}`$ collett , this is also the limit to the introduced phase variance.
The photon number of the squeezed state at maximum overlap will be mainly determined by the photon number of the input, but the degree and direction of squeezing (parametrized by $`\xi `$) will be determined by the multiplying factor $`Q^{}(A,B)`$. The multiplying factor can be expressed as a function of $`\overline{n}`$ and $`\zeta `$, for which we will use the same symbol $`Q^{}`$, even though it is a new function $`Q^{}(\overline{n},\zeta )`$. Here $`\overline{n}`$ is the mean photon number for the state $`|\alpha ,\xi `$ (and will be close to the photon number $`\overline{n}^S`$ of the input state), and $`\zeta =\xi \alpha ^{}/\alpha `$ is $`\xi `$ with the phase of $`\alpha `$ scaled out. The multiplying factor will tend to be concentrated along a particular line, effectively giving $`\zeta `$ as a function of $`\overline{n}`$. In order to obtain the theoretical limit, the measurement scheme must give a multiplying factor $`Q^{}(\overline{n},\zeta )`$ that tends to give values of $`\zeta `$ for each $`\overline{n}`$ that are the same as for optimized squeezed states.
We can determine the approximate variation of $`\zeta `$ with $`\overline{n}`$ in the multiplying factor if we can estimate how it varies for measurements on a coherent state. If we consider measurements on a coherent state with real amplitude $`\alpha ^S`$, then the maximum overlap with the state $`|\alpha ,\xi `$ will be for $`\alpha ^S\alpha `$. We use $`\alpha ^S`$ without a subscript to indicate the initial coherent amplitude before the measurement.
If we are using an adaptive scheme with intermediate phase estimates that are unbiased, it is easy to see that the maximum probability will be for $`B`$ real and therefore also $`A`$ real. These results imply that
$$\alpha \frac{A(1+B)}{1B^2}=\frac{A}{1B}.$$
(20)
In turn this gives $`\zeta `$ as
$`\zeta `$ $`\text{atanh}(1A/\alpha )`$ (21)
$`{\displaystyle \frac{1}{2}}\mathrm{ln}{\displaystyle \frac{A}{2\alpha }}{\displaystyle \frac{1}{2}}\mathrm{ln}{\displaystyle \frac{A}{2\sqrt{\overline{n}}}}.`$ (22)
Since the value of $`\zeta `$ is governed by the multiplying factor $`Q^{}(\overline{n},\zeta )`$, this result for $`\zeta `$ should hold for more general input states.
From Ref. collett the phase variance of a squeezed state is
$$\mathrm{\Delta }\varphi ^2\frac{n_0+1}{4\overline{n}^2}+2\text{erfc}(\sqrt{2n_0}).$$
(23)
where $`n_0=\overline{n}e^{2\zeta }`$ for real $`\zeta `$. This is minimized asymptotically as
$$\frac{\mathrm{ln}\overline{n}+\mathrm{\Delta }}{4\overline{n}^2},$$
(24)
where $`\mathrm{\Delta }2.43`$, for
$$n_0\mathrm{ln}(4\overline{n})\frac{1}{4}\mathrm{ln}(2\pi ).$$
(25)
If we use the result obtained for $`\zeta `$ in Eq. (22) we find that
$$n_0\frac{1}{2}|A|\sqrt{\overline{n}}.$$
(26)
This result means that in order for the measurement to be optimal, $`|A|`$ should scale with $`\overline{n}`$ as
$$|A|\frac{\mathrm{ln}\overline{n}}{\sqrt{\overline{n}}}.$$
(27)
For the case of mark II measurements we have the result that $`|A|=1`$ semiclass , which is why these measurements are not optimal. Note that if we substitute $`|A|=1`$ into the expression (26) to find $`n_0`$, and substitute that into Eq. (23), we obtain the correct result for the mark II introduced phase variance,
$$\mathrm{\Delta }\varphi ^2\frac{1}{8}\overline{n}^{1.5}.$$
(28)
## III Improved feedback
Now we have the result that for optimal feedback $`|A|`$ should decrease with photon number. Therefore in order to improve the phase measurement scheme we want one that gives $`|A|<1`$. To see in general how this can be achieved, consider a coherent state with amplitude $`\alpha ^S`$ and determine the Ito SDE (stochastic differential equation) for $`|A|^2`$:
$`d|A_v|^2`$ $`=A_v^{}(dA_v)+(dA_v^{})A_v+(dA_v^{})(dA_v)`$ (29)
$`=A_v^{}e^{i\mathrm{\Phi }(v)}I(v)dv+e^{i\mathrm{\Phi }(v)}I(v)dvA_v+dv`$ (30)
$`=[|A_v|I(v)2\mathrm{R}\mathrm{e}(e^{i\mathrm{\Phi }(v)}e^{i\phi _v^A})+1]dv,`$ (31)
where $`\phi _v^A=\mathrm{arg}A_v`$. In terms of the phase estimate $`\widehat{\phi }_v=\mathrm{\Phi }(v)\pi /2`$ this becomes
$$d|A_v|^2=[1+2|A_v|I(v)\mathrm{sin}(\phi _v^A\widehat{\phi }_v)]dv.$$
(32)
If we take the expectation value of $`I(v)`$ and simplify we get
$$I(v)=2|\alpha ^S|\mathrm{sin}(\widehat{\phi }_v\phi ),$$
(33)
where $`\phi =\mathrm{arg}\alpha ^S`$. If we use this result the expectation value for the increment in $`|A_v|^2`$ is
$$d|A_v|^2=\left[14|A_v||\alpha ^S|\mathrm{sin}(\widehat{\phi }_v\phi )\mathrm{sin}(\phi _v^A\widehat{\phi }_v)\right]dv.$$
(34)
The first term on its own will give $`|A|=1`$, and in order to get $`|A|<1`$ the two sines must have the same sign. This will be the case if the phase estimate is between the actual phase and the phase of $`A_v`$. It is for this reason that we consider phase estimates that are intermediate between the best phase estimate and the phase of $`A_v`$, i.e., of the form
$$\widehat{\phi }(v)=\mathrm{arg}[(A_vv+B_vA_v^{})^{1ϵ(v)}A_v^{ϵ(v)}].$$
(35)
In general, smaller values of $`|A|`$ can be obtained by using smaller values of $`ϵ`$. This is because $`\phi _v^A`$ tends to be a worse phase estimate, thus making it possible for the sines in Eq. (34) to be larger. Note that it is far too simplistic to use the best phase estimate (i.e., with $`ϵ=0`$), as we need to adjust $`ϵ`$ in order to make $`n_0`$ closer to optimal.
## IV Simulation method
The easiest input states to use for numerical simulations are coherent states, as they remain coherent with a deterministically decaying amplitude. However, in order to estimate the phase variance that is introduced by the measurement this would be very inefficient, as the phase variance would be dominated by the intrinsic phase variance. It is almost as easy (and much more efficient) to perform calculations on squeezed states, as squeezed states remain squeezed states under the stochastic evolution, and only the two squeezing parameters need be kept track of. The best squeezed states to use are those optimized for minimum intrinsic phase variance. For these states the total phase variance will be approximately twice the intrinsic phase variance when the measurements are close to optimal.
To determine the SDE’s for the squeezing parameters, we must first consider the SDE for the state. For dyne detection the stochastic evolution of the conditioned state vector is wise96
$`d|\psi (t)`$ $`=[dt({\displaystyle \frac{a^{}a}{2}}{\displaystyle \frac{a^{}a}{2}}+{\displaystyle \frac{a^{}\gamma +\gamma ^{}a}{2}}\gamma ^{}a)`$
$`+dN(t)({\displaystyle \frac{ae^{i\mathrm{\Phi }}+|\gamma |}{\sqrt{(a^{}+\gamma ^{})(a+\gamma )}}}1)]|\psi (t),`$
where $`a`$ is the annihilation operator for the mode, $`|\gamma |1`$ is the amplitude of the local oscillator, and $`\mathrm{\Phi }=\mathrm{arg}\gamma `$ is its phase. Here the mode being measured is assumed to come from a cavity with an intensity decay rate equal to unity. The point process $`dN(t)`$ has a mean $`\kappa dt`$, where
$$\kappa =(a^{}+\gamma ^{})(a+\gamma ).$$
(37)
The equation given in wise96 differs from Eq. (LABEL:thiseq) by a trivial phase factor. The form above is given because it is not possible to directly take the limit of large local oscillator amplitude using the form given in wise96 . To take the limit of large local oscillator amplitude we approximate the Poisson process $`\delta N(t)`$ by a Gaussian process
$$\delta N(t)\kappa \delta t+\sqrt{\kappa }\delta W(t),$$
(38)
where $`\delta W(t)`$ is a Gaussian random variable of zero mean and variance $`\delta t`$. Then we find that in the limit of large $`|\gamma |`$ we have
$`d|\psi (t)`$ $`=[(a^{}a/2+a\chi e^{i\mathrm{\Phi }}\chi ^2/2)dt`$
$`+(ae^{i\mathrm{\Phi }}\chi )dW]|\psi (t),`$ (39)
where
$$\chi =\frac{1}{2}(ae^{i\mathrm{\Phi }}+a^{}e^{i\mathrm{\Phi }}).$$
(40)
In order to determine the SDE’s for the squeezing parameters, we use the method of Rigo et al. rigo . Squeezed states obey the relation
$$(aB_t^Sa^{}A_t^S)|A_t^S,B_t^S.$$
(41)
The squeezing parameters $`A_t^S`$ and $`B_t^S`$ are related to the usual squeezing parameters in the same way as $`A`$ and $`B`$ are in Eq. (14) and Eq. (15). In the Stratonovich formalism
$$(aB_t^Sa^{}A_t^S)d|\psi (t)=(dB_t^Sa^{}+dA_t^S)|\psi (t).$$
(42)
Converting the SDE for the state to the Stratonovich form in the usual way text , we find
$`d|\psi (t)`$ $`=[({\displaystyle \frac{a^{}a}{2}}{\displaystyle \frac{a^2e^{2i\mathrm{\Phi }}}{2}}+2a\chi e^{i\mathrm{\Phi }}\chi ^2)dt`$
$`+(ae^{i\mathrm{\Phi }}\chi )dW`$
$`{\displaystyle \frac{1}{2}}[ad(e^{i\mathrm{\Phi }})d\chi ]dW]|\psi (t).`$ (43)
Here we have included the increments $`d(e^{i\mathrm{\Phi }})`$ and $`d\chi `$ because the phase of the local oscillator can vary stochastically. Using this form of the equation, the left hand side of Eq. (42) evaluates to
$`\{dt[(a^{}B_t^S+A_t^S/2)B_t^S(B_t^Sa^{}+A_t^S)e^{2i\mathrm{\Phi }}+2B_t^S\chi e^{i\mathrm{\Phi }}]`$
$`+dW[B_t^Se^{i\mathrm{\Phi }}{\displaystyle \frac{1}{2}}A_t^Sd(e^{i\mathrm{\Phi }})]\}|\psi (t).`$
This gives us the SDE’s for the squeezing parameters,
$`dB_t^S`$ $`=B_t^S(1+e^{2i\mathrm{\Phi }}B_t^S)dt,`$ (44)
$`dA_t^S`$ $`={\displaystyle \frac{1}{2}}A_t^Sdt+B_t^Sa^{}(1+e^{2i\mathrm{\Phi }}B_t^S)dt+B_t^Se^{i\mathrm{\Phi }}dW`$
$`{\displaystyle \frac{1}{2}}B_t^Sd(e^{i\mathrm{\Phi }})dW.`$ (45)
From these we find that the Stratonovich SDE for the standard (nonscaled) amplitude $`\alpha _t^S`$ is
$`d\alpha _t^S`$ $`={\displaystyle \frac{1}{2}}\alpha _t^Sdt+{\displaystyle \frac{B_t^SdW}{1|B_t^S|^2}}[(B_t^S)^{}e^{i\mathrm{\Phi }}+e^{i\mathrm{\Phi }}]`$
$`{\displaystyle \frac{1}{2}}{\displaystyle \frac{B_t^SdW}{1|B_t^S|^2}}[(B_t^S)^{}d(e^{i\mathrm{\Phi }})+d(e^{i\mathrm{\Phi }})].`$
Converting back to the Ito SDE, we get
$`d\alpha _t^S={\displaystyle \frac{1}{2}}\alpha _t^Sdt+{\displaystyle \frac{B_t^SdW}{1|B_t^S|^2}}[(B_t^S)^{}e^{i\mathrm{\Phi }}+e^{i\mathrm{\Phi }}].`$
The SDE for $`B_t^S`$ is unchanged under the change to Ito form. If we take the signal to be $`I(t)\delta t=(\delta N_+\delta N_{})/\beta `$ (for consistency with Ref. wise96 ), then take the limit of large oscillator amplitude and small time intervals $`\delta t`$, we obtain
$$I(t)dt=2\text{Re}(\alpha _t^Se^{i\mathrm{\Phi }(t)})dt+dW(t).$$
(48)
The parameters $`A_t`$ and $`B_t`$ are then defined as in wise96 by
$`A_t`$ $`={\displaystyle _0^t}e^{i\mathrm{\Phi }}e^{s/2}I(s)𝑑s,`$ (49)
$`B_t`$ $`={\displaystyle _0^t}e^{2i\mathrm{\Phi }}e^s𝑑s.`$ (50)
In order to get rid of the exponential factors, we change the time variable to
$$v=1e^t,$$
(51)
and we redefine the amplitude to remove the systematic variation:
$$\alpha _v^S=\alpha _t^Se^{t/2}.$$
(52)
Here we use the $`v`$ subscript to indicate the scaled amplitude, and the $`t`$ subscript to indicate the original, unscaled amplitude. Since these are equal to each other at zero time, there is no ambiguity in the initial amplitude $`\alpha ^S`$. Reverting to our original definition of the signal (1), we find
$$I(v)dv=2\text{Re}(\alpha _v^Se^{i\mathrm{\Phi }(v)})dv+dW(v).$$
(53)
With these changes of variables, the definitions for $`A_v`$ and $`B_v`$ become
$`A_v`$ $`={\displaystyle _0^v}e^{i\mathrm{\Phi }}I(u)𝑑u,`$ (54)
$`B_v`$ $`={\displaystyle _0^v}e^{2i\mathrm{\Phi }}𝑑u.`$ (55)
The differential equations for the squeezing parameters become
$`dB_v^S`$ $`={\displaystyle \frac{dv}{1v}}B_v^S(1+e^{2i\mathrm{\Phi }}B_v^S),`$ (56)
$`d\alpha _v^S`$ $`={\displaystyle \frac{1}{1v}}{\displaystyle \frac{B_v^SdW(v)}{1|B_v^S|^2}}[(B_v^S)^{}e^{i\mathrm{\Phi }}+e^{i\mathrm{\Phi }}].`$ (57)
Initial calculations were performed using these equations, but there is a further simplification that can be made. The solution for $`B_v^S`$ is
$$B_v^S=\frac{1v}{(B_0^S)^1B_v^{}}.$$
(58)
For calculations with time-dependent $`ϵ`$ this solution for $`B_v^S`$ was used rather than solving a separate differential equation for $`B_v^S`$.
## V Results
First we will describe the results for constant $`ϵ`$. For each mean photon number, $`ϵ`$ was varied to find the value that gave the minimum phase variance. This method does not give results close to the theoretical limit for photon numbers above about 5000, but the phase variances continue to get smaller as compared to the phase variances for mark II measurements. This indicates that the results are following a different scaling law, and fitting techniques give the power for the introduced phase variance as $`1.685\pm 0.007`$. The data and the fitted line along with the heterodyne and mark II cases and the theoretical limit are shown in Fig 2. These results are a significant improvement over the mark II case, but are still significantly above the theoretical limit.
In order to improve on this result we must vary $`ϵ`$ during the measurement. The value of $`ϵ`$ that we found to give the best result was
$$ϵ(v)=\frac{v^2|B_v|^2}{|C_v|}\sqrt{\frac{v}{1v}}.$$
(59)
The reason for the multiplying factor of $`(v^2|B_v|^2)/|C_v|`$ is that it is an estimator for $`1/|\alpha ^S|`$. This means that the value of $`ϵ`$ tends to be smaller for larger photon numbers, resulting in smaller values of $`|A|`$. The reason for the factor of $`\sqrt{v/(1v)}`$ is that it makes the value of $`ϵ`$ close to zero initially, and very large near the end of the measurement.
This second factor was found essentially by trial and error, and is thought to be related to the fact that the phase of $`\alpha _v^S`$ varies stochastically during the measurement. Recall that during the measurement we want the phase estimate to be between the phase of $`\alpha _v^S`$ and the phase of $`A_v`$. We only have an estimate of the phase of $`\alpha ^S`$ (the initial phase), so if we use a phase estimate that is too close to the actual phase when the phase variance of $`\alpha _v^S`$ is large, the phase estimate is likely to be outside the interval between the phase of $`\alpha _v^S`$ and the phase of $`A_v`$. Since the phase variance of $`\alpha _v^S`$ increases with time, the value of $`ϵ`$ is increased as well, to prevent this happening.
The results for this method are shown in Fig. 3 as a ratio to the theoretical limit. As this shows, the results are very close to the theoretical limit, and even for the largest photon number for which calculations have been performed the phase uncertainty is only about 4% above the theoretical limit. For these calculations the time steps used were approximately
$$\mathrm{\Delta }v=\frac{\overline{n}^S\mathrm{\Delta }\varphi ^2_{\mathrm{th}}}{25},$$
(60)
where $`\mathrm{\Delta }\varphi ^2_{\mathrm{th}}`$ is the theoretical limit to the phase uncertainty. With these time steps the uncertainty due to the finite step size is approximately 1%.
If the integration time step is reduced, while keeping the time interval at which the phase estimates are updated constant, the phase variance converges. If, however, the phase estimates are updated at smaller and smaller time intervals then the phase variance does not converge. For example, the phase uncertainty for measurements on an optimized squeezed state with a photon number of 1577 is $`1.54\times 10^6`$ if we use the time steps given above. If, however, we use time steps that are 100 times smaller, then the phase variance is $`1.93\times 10^6`$, and if the time steps are 1000 times smaller the phase variance is $`2.13\times 10^6`$. These results indicate that the phase estimates must be incremented in finite time intervals for this method to give good results, and the size of the time steps that should be used depends on the photon number. The phase variance is not strongly dependent on these time steps, however, and only an order of magnitude estimate of the photon number is required.
## VI Evaluation of method
A problem with determing the phase variance by the method above is that, for highly squeezed states (that are close to optimized for minimum phase variance), a significant contribution to the phase variance is from low probability results around $`\pi `$. In obtaining numerical results the actual phase variance for the measurement will tend to be underestimated because the results from around $`\pi `$ are obtained too rarely for good statistics. It would require an extremely large number of samples to estimate this contribution. However, we can estimate it nonstatistically as follows.
Recall that in order to have a measurement that is close to optimum the multiplying factor $`Q^{}(\overline{n},\zeta )`$ should give values of $`\zeta `$ for each $`\overline{n}`$ that are close to optimized for minimum phase uncertainty. To test this for the phase measurement scheme described above, the $`\overline{n}`$ and $`\zeta `$ were determined from the values of $`A`$ and $`B`$ from the samples. The resulting data along with the line for optimized $`\zeta `$ are plotted in Fig. 4. The imaginary part of $`\zeta `$ should be zero for optimum measurements, and is small for these results. Therefore in Fig. 4 we have plotted the real part $`\zeta _R`$. As can be seen, the vast majority of the data points are below the line, indicating greater squeezing than optimum. This means that if the low probability results around $`\pi `$ are taken into account the phase variance for these measurements will be above the theoretical limit.
First we consider the effect of variations in the modulus of $`\zeta `$, leaving consideration of error in the phase till later. In order to estimate how far above the theoretical limit the actual phase variance is, we make a quadratic approximation to the expression for the phase variance. From collett the expression for the phase variance of a squeezed state is, for real $`\zeta `$,
$$\delta \varphi ^2\frac{e^{2\zeta }}{4\overline{n}}+\frac{1}{4\overline{n}^2}2\text{erfc}(\sqrt{2\overline{n}}e^\zeta ).$$
(61)
Taking the derivative with respect to $`\zeta `$ gives
$$\frac{d}{d\zeta }\delta \varphi ^2\frac{e^{2\zeta }}{2\overline{n}}4e^\zeta \sqrt{\frac{2\overline{n}}{\pi }}e^{2\overline{n}e^{2\zeta }}.$$
(62)
Taking the second derivative and using the fact that the expression above is zero for minimum phase variance gives
$$\frac{d^2}{d\zeta ^2}\delta \varphi ^2\frac{n_0}{2\overline{n}^2}(1+4n_0).$$
(63)
This means that for values of $`\zeta `$ close to optimum the increase in the phase variance over the optimum value is
$$\mathrm{\Delta }\delta \varphi ^2(\mathrm{\Delta }|\zeta |)^2\frac{n_0}{4\overline{n}^2}(1+4n_0).$$
(64)
The main contribution to the phase uncertainty is $`n_0/(4\overline{n}^2)`$, so the increase in the phase uncertainty as a ratio to the minimum phase uncertainty is
$$\frac{\mathrm{\Delta }\delta \varphi ^2}{\delta \varphi ^2_{\mathrm{min}}}(\mathrm{\Delta }|\zeta |)^2(1+4n_0).$$
(65)
This estimate indicates that the actual phase variance for the measurement scheme described above can be significantly larger than the intrinsic phase variance. For example, for a mean photon number of about 332 000 the rms
deviation of $`|\zeta |`$ from the optimum value is only about 0.16, but a squeezed state with $`|\zeta |`$ differing this much from optimum will have a phase variance more than twice the optimum value. This indicates that if the low probability results around $`\pi `$ are taken into account the introduced phase variance is actually more than twice the theoretical limit.
Next, we estimate the contribution from error in the phase (rather than the modulus) of $`\zeta `$. For a squeezed state with real $`\alpha `$ the intrinsic uncertainty in the zero quadrature is
$$X_0^2=e^{2|\zeta |}\mathrm{cos}^2\frac{\mu }{2}+e^{2|\zeta |}\mathrm{sin}^2\frac{\mu }{2},$$
(66)
where $`\mu =\mathrm{arg}\zeta `$. Since $`X_0=2\alpha \mathrm{sin}(\varphi )2\alpha \varphi `$, the intrinsic uncertainty in the phase is
$$\delta \varphi ^2\frac{e^{2|\zeta |}\mathrm{cos}^2(\mu /2)+e^{2|\zeta |}\mathrm{sin}^2(\mu /2)}{4\overline{n}}.$$
(67)
If the phase of $`\zeta `$ is small, we can make the approximation
$$\delta \varphi ^2\frac{e^{2|\zeta |}+e^{2|\zeta |}\mu ^2/4}{4\overline{n}}.$$
(68)
Clearly the first term in the numerator is just the original phase variance, and the second term is the excess phase variance due to the error in the phase of $`\zeta `$. Therefore the extra phase variance due to error in the phase of $`\zeta `$ is given by
$$\mathrm{\Delta }\delta \varphi ^2\frac{(\mathrm{\Delta }\mathrm{arg}\zeta )^2}{16n_0}.$$
(69)
Using this estimate on the previous example it can be seen that this is not so much of a problem, with the introduced phase uncertainty being increased by less than 3% by this factor.
## VII Improved method
The problem of the large contribution of the low probability results around $`\pi `$ can be effectively eliminated in the following way. At each time step the photon number is estimated from the values of $`A_v`$ and $`B_v`$, and the optimum value of $`\zeta `$ is estimated using the asymptotic formula in collett . Then if $`\zeta _R`$ (the real part of $`\zeta `$) is too far below the optimum value, rather than using the feedback phase above, we use
$$\mathrm{\Phi }(v)=\frac{1}{2}\mathrm{arg}\left[B_vv\frac{C_v}{C_v^{}}\mathrm{tanh}\left|\zeta _{\mathrm{opt}}\right|\right].$$
(70)
Using this feedback phase takes $`B_v`$ directly towards the optimum value. To see this, note that the optimum value of $`B_v`$ is
$$B_v^{\mathrm{opt}}=v\frac{C_v}{C_v^{}}\mathrm{tanh}\left|\zeta _{\mathrm{opt}}\right|.$$
(71)
Taking the exponential of the feedback phase given by Eq. (70) gives $`e^{2i\mathrm{\Phi }}B_vB_v^{\mathrm{opt}}`$, so $`dB_vB_v^{\mathrm{opt}}B_v`$.
The details of exactly when $`\zeta _R`$ is considered too far below optimum can be varied endlessly, but for the results that will be presented here we use this alternate phase estimate after time $`v=0.9`$ and when
$$|\zeta |>|\zeta _{\mathrm{opt}}|e^{\lambda |\alpha _v|^2(1v)},$$
(72)
where $`\zeta _{\mathrm{opt}}`$ is the estimated optimum value of $`\zeta `$ and $`\alpha _v`$ is $`C_v/(v^2|B_v|^2)`$. Using the exponential multiplying factor means that the alternative feedback is used only toward the end of the measurement. Only considering the alternative feedback in the last 10% of the measurement is necessary for the smaller photon numbers, where Eq. (72) is too weak a restriction.
Another variation from the previous scheme is that, for the larger photon numbers, the values of $`ϵ`$ given by the original expression were reduced. The above correction corrects only for values of $`\zeta _R`$ that are below optimum, and for the larger photon numbers many of the uncorrected values of $`\zeta _R`$ tend to be above optimum (see Fig. 4). The corrections will still work well, however, if we use a dividing factor to bring the uncorrected values below the line. For the second largest photon number tested of around $`5\times 10^5`$, the best results were obtained when the values of $`ϵ`$ as given by Eq. (59) were divided by 1.1. For the largest mean photon number tested, $`2\times 10^7`$, the best results were obtained for a dividing factor of 1.2. The value of $`\lambda `$ that gave the best results with these dividing factors was $`5\times 10^4`$. For all other mean photon numbers the value of $`\lambda `$ used was $`10^3`$.
The estimated contributions to the phase variance due to error in the magnitude and phase of $`\zeta `$ are plotted in Fig. 5. As can be seen, the contribution due to error in the magnitude of $`\zeta `$ is very small, around 1.5% for the larger photon numbers tested. The contribution due to the error in the phase of $`\zeta `$ is a bit larger, but it still does not rise above 3%. Thus we can see that the introduced phase variance can be made very close to the theoretical limit, within 5% for the largest photon number tested.
With this modified technique the phase variance again does not converge as the feedback phase is updated in smaller and smaller time intervals. The phase variance is less dependent on the time step with this technique, however. For example, for a mean photon number of 1577 the total phase variance for measurements on an optimized squeezed state increases by only about 9% as the time steps are reduced by a factor of 1000. In contrast, the phase variance increases by a factor of 38% for the previous technique.
## VIII Conclusions
Any estimate of an initially unknown optical phase made using standard devices (linear optical and opto-electronic devices, a local oscillator, and photodetectors) must have an uncertainty above the intrinsic quantum uncertainty in the phase of the input state. The minimum magnitude of the added phase variance was determined in Ref. fullquan to scale asymptotically as
$$\frac{\mathrm{ln}\overline{n}^S}{4(\overline{n}^S)^2},$$
(73)
where $`\overline{n}^S`$ is the mean photon number of the input state. Previous phase measurement schemes do not approach this theoretical limit. In this paper we have shown that an adaptive phase measurement scheme not previously considered can attain this theoretical limit. In other words, we have determined what is essentially the best possible phase measurement technique.
In practice, phase measurements are currently limited by detector inefficiency. For detector efficiency $`\eta `$ the introduced phase variance cannot be reduced below semiclass
$$\frac{1\eta }{4\eta \overline{n}^S}.$$
(74)
When the mark II phase variance is less than this there is not likely to be any significant advantage to using a more advanced feedback scheme. For the best photodetectors available today, with around 98% efficiency polzik , the mark II phase variance falls below this limit for photon numbers above 1000. Below this photon number the mark II phase variance is never more than about 27% above the limits determined using Eqs. (73) and (74), so only relatively small improvements can be obtained by using a more advanced feedback scheme.
Nevertheless, the technology is always improving, and there is no fundamental reason why photodetectors cannot be built with efficiencies extremely close to 1 hidpc . When very efficient photodetectors are developed, the feedback techniques described here have the potential to give great improvements in the accuracy of phase measurements for applications where there is a limitation on the photon number that can be used. The other detrimental factors are relatively minor, although the time delay in the feedback loop will become significant for very short pulses.
The primary significance of the result obtained in this paper is theoretical, however, as it represents the culmination of the search for the best optical phase measurement schemes using standard devices. To do any better would require using nonlinear optical devices. For example, it is conceivable that down-converting some portion of the signal field, and then measuring the phase of the down-converted light, could enable the above theoretical limit to be surpassed. This is a question for future work.
|
warning/0002/math0002135.html
|
ar5iv
|
text
|
# 𝑆𝐿(2) and 𝑧-measures
## 1 Introduction
This paper is about the $`z`$-measures which are a remarkable two-parametric family of measures on partitions introduced in in the context of harmonic analysis on the infinite symmetric group. In a series of papers, A. Borodin and G. Olshanski obtained several fundamental results on these $`z`$-measures, see their survey which appears in this volume and also . The culmination of this development is an exact determinantal formula for the correlation functions of the $`z`$-measures in terms of the hypergeometric kernel . We mention as one of the applications of this formula. The main result of this paper is a representation-theoretic derivation of the formula of Borodin and Olshanski.
### 1.1
In the early days of $`z`$-measures, it was already noticed that $`z`$-measures have some mysterious connection to representation theory of $`SL(2)`$. For example, the $`z`$-measure is actually positive if its two parameters $`z`$ and $`z^{}`$ are either complex conjugate $`z^{}=\overline{z}`$ or $`z,z^{}(n,n+1)`$ for some $`n`$. In these cases $`zz^{}`$ is either imaginary or lies in $`(1,1)`$, which was certainly reminiscent of the principal and complementary series of representations of $`SL(2)`$.
Later, S. Kerov constructed an $`SL(2)`$-action on partitions for which the $`z`$-measures are certain matrix elements . Finally, Borodin and Olshanski computed the correlation functions of the $`z`$-measures is in terms of the Gauss hypergeometric function which is well known to arise as matrix elements of representations of $`SL(2)`$. The aim of this paper is to put these pieces together.
### 1.2
I want to thank A. Borodin, S. Kerov, G. Olshanski, and A. Vershik for numerous discussions of the $`z`$-measures. I also want to thank the organizers of the Random Matrices program at MSRI, especially P. Bleher, P. Deift, and A. Its. My research was supported by NSF under grant DMS-9801466.
The constructions of this paper were subsequently generalized beyond $`SL(2)`$ and $`z`$-measures in .
## 2 The $`z`$-measures, Kerov’s operators, and correlation functions
### 2.1 Definition of the $`z`$-measures
Let $`z,z^{}`$ be two parameters and consider the following measure on the set of all partitions $`\lambda `$ of $`n`$
$$_n(\lambda )=\frac{n!}{(zz^{})_n}\underset{\mathrm{}\lambda }{}\frac{(z+c(\mathrm{}))(z^{}+c(\mathrm{}))}{h(\mathrm{})^2},$$
(2.1)
where
$$(x)_n=x(x+1)\mathrm{}(x+n1),$$
the product is over all squares $`\mathrm{}`$ in the diagram of $`\lambda `$, $`h(\mathrm{})`$ is the length of the corresponding hook, and $`c(\mathrm{})`$ stands for the content of the square $`\mathrm{}`$. Recall that, by definition, the content of $`\mathrm{}`$ is
$$c(\mathrm{})=\text{column}(\mathrm{})\text{row}(\mathrm{}),$$
where $`\text{column}(\mathrm{})`$ denotes the column number of the square $`\mathrm{}`$. The reader is referred to for general facts about partitions.
It is not immediately obvious from the definition (2.1) that
$$\underset{|\lambda |=n}{}_n(\lambda )=1.$$
(2.2)
One possible proof of (2.2) uses the following operators on partitions introduced by S. Kerov.
### 2.2 Kerov’s operators
Consider the vector space with an orthonormal basis $`\{\delta _\lambda \}`$ indexed by all partitions of $`\lambda `$ of any size. Introduce the following operators
$`U\delta _\lambda `$ $`={\displaystyle \underset{\mu =\lambda +\mathrm{}}{}}`$ $`(z+c(\mathrm{}))`$ $`\delta _\mu `$
$`L\delta _\lambda `$ $`=`$ $`(zz^{}+2|\lambda |)`$ $`\delta _\lambda `$
$`D\delta _\lambda `$ $`={\displaystyle \underset{\mu =\lambda \mathrm{}}{}}`$ $`(z^{}+c(\mathrm{}))`$ $`\delta _\mu ,`$
where $`\mu =\lambda +\mathrm{}`$ means that $`\mu `$ is obtained from $`\lambda `$ by adding a square $`\mathrm{}`$ and $`c(\mathrm{})`$ is the content of this square. The letters $`U`$ and $`D`$ here stand for “up” and “down”.
These operators satisfy the commutation relations
$$[D,U]=L,[L,U]=2U,[L,D]=2D,$$
(2.3)
same as for the following basis of $`𝔰𝔩(2)`$
$$U=\left(\begin{array}{cc}0& 1\\ 0& 0\end{array}\right),L=\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right),D=\left(\begin{array}{cc}0& 0\\ 1& 0\end{array}\right).$$
In particular, it is clear that if $`|\lambda |=n`$ then
$$(U^n\delta _{\mathrm{}},\delta _\lambda )=dim\lambda \underset{\mathrm{}\lambda }{}(z+c(\mathrm{}))$$
where
$$dim\lambda =n!\underset{\mathrm{}\lambda }{}h(\mathrm{})^1$$
is the number of standard tableaux on $`\lambda `$. It follows that
$$_n(\lambda )=\frac{1}{n!(zz^{})_n}(U^n\delta _{\mathrm{}},\delta _\lambda )(L^n\delta _\lambda ,\delta _{\mathrm{}}).$$
Using this presentation and the commutation relations (2.3) one proves (2.2) by induction on $`n`$.
### 2.3 The measure $``$ and its normalization
In a slightly different language, with induction on $`n`$ replaced by the use of generating functions, this computation goes as follows.
The sequence of the measures $`_n`$ can be conveniently assembled as in into one measure $``$ on the set of all partitions of all numbers as follows
$$=(1\xi )^{zz^{}}\underset{n=0}{\overset{\mathrm{}}{}}\xi ^n\frac{(zz^{})_n}{n!}_n,\xi [0,1),$$
where $`\xi `$ is a new parameter. In other words, $``$ is the mixture of the measures $`_n`$ by means of a negative binomial distribution on $`n`$ with parameter $`\xi `$.
It is clear that (2.2) is now equivalent to $``$ being a probability measure. It is also clear that
$$(\lambda )=(1\xi )^{zz^{}}(e^{\sqrt{\xi }U}\delta _{\mathrm{}},\delta _\lambda )(e^{\sqrt{\xi }D}\delta _\lambda ,\delta _{\mathrm{}})$$
(2.4)
Therefore
$`{\displaystyle \underset{\lambda }{}}(\lambda )=(1\xi )^{zz^{}}(e^{\sqrt{\xi }D}e^{\sqrt{\xi }U}\delta _{\mathrm{}},\delta _{\mathrm{}})`$ (2.5)
It follows from the definitions that
$$D\delta _{\mathrm{}}=0,L\delta _{\mathrm{}}=zz^{}\delta _{\mathrm{}},U^{}\delta _{\mathrm{}}=0,$$
(2.6)
where $`U^{}`$ is the operator adjoint to $`U`$. Therefore, in order to evaluate (2.5), it suffices to commute $`e^{\sqrt{\xi }L}`$ through $`e^{\sqrt{\xi }U}`$.
The following computation in the group $`SL(2)`$
$$\left(\begin{array}{cc}1& 0\\ \beta & 1\end{array}\right)\left(\begin{array}{cc}1& \alpha \\ 0& 1\end{array}\right)=\left(\begin{array}{cc}1& \frac{\alpha }{1\alpha \beta }\\ 0& 1\end{array}\right)\left(\begin{array}{cc}\frac{1}{1\alpha \beta }& 0\\ 0& 1\alpha \beta \end{array}\right)\left(\begin{array}{cc}1& 0\\ \frac{\beta }{1\alpha \beta }& 1\end{array}\right)$$
implies that
$$\begin{array}{c}\mathrm{exp}\left(\beta D\right)\mathrm{exp}\left(\alpha U\right)=\hfill \\ \hfill \mathrm{exp}\left(\frac{\alpha }{1\alpha \beta }U\right)(1\alpha \beta )^L\mathrm{exp}\left(\frac{\beta }{1\alpha \beta }D\right),\end{array}$$
(2.7)
provided $`|\alpha \beta |<1`$. Therefore,
$`{\displaystyle \underset{\lambda }{}}(\lambda )`$ $`=(1\xi )^{zz^{}}(\mathrm{exp}\left({\displaystyle \frac{\sqrt{\xi }}{1\xi }}U\right)(1\xi )^L\mathrm{exp}\left({\displaystyle \frac{\sqrt{\xi }}{1\xi }}D\right)\delta _{\mathrm{}},\delta _{\mathrm{}})`$
$`=(1\xi )^{zz^{}}((1\xi )^L\delta _{\mathrm{}},\delta _{\mathrm{}})=1,`$
as was to be shown.
### 2.4 Correlation functions
Introduce the following coordinates on the set of partitions. To a partition $`\lambda `$ we associate a subset
$$𝔖(\lambda )=\{\lambda _ii+1/2\}+\frac{1}{2}.$$
For example,
$$𝔖(\mathrm{})=\{\frac{1}{2},\frac{3}{2},\frac{5}{2},\mathrm{}\}$$
This set $`𝔖(\lambda )`$ has the following geometric interpretation. Take the diagram of $`\lambda `$ and rotate it $`135^{}`$ as in the following picture:
The positive direction of the axis points to the left in the above figure. The boundary of $`\lambda `$ forms a zigzag path and the elements of $`𝔖(\lambda )`$, which are marked by $``$, correspond to moments when this zigzag goes up.
Subsets $`S+\frac{1}{2}`$ of the form $`S=𝔖(\lambda )`$ can be characterized by
$$|S_+|=|S_{}|<\mathrm{}$$
where
$$S_+=S\left(_0\frac{1}{2}\right),S_{}=\left(_0\frac{1}{2}\right)S.$$
The number $`|𝔖_+(\lambda )|=|𝔖_{}(\lambda )|`$ is the number of squares in the diagonal of the diagram of $`\lambda `$ and the finite set $`𝔖_+(\lambda )𝔖_{}(\lambda )+\frac{1}{2}`$ is known as the modified Frobenius coordinates of $`\lambda `$.
Given a finite subset $`X+\frac{1}{2}`$, define the *correlation function* by
$$\rho (X)=\left(\{\lambda ,X𝔖(\lambda )\}\right).$$
In , A. Borodin and G. Olshanski proved that
$$\rho (X)=det\left[K(x_i,x_j)\right]_{x_i,x_jX}$$
where $`K`$ the *hypergeometric kernel* introduced in . This kernel involves the Gauss hypergeometric function and the explicit formula for $`K`$ will be reproduced below.
It is our goal in the present paper to give a representation-theoretic derivation of the formula for correlation functions and, in particular, show how the kernel $`K`$ arises from matrix elements of irreducible $`SL(2)`$-modules.
## 3 SL(2) and correlation functions
### 3.1 Matrix elements of $`𝔰𝔩(2)`$-modules and Gauss hypergeometric function
The fact that the hypergeometric function arises as matrix coefficients of $`SL(2)`$ modules is well known. A standard way to see this is to use a functional realization of these modules; the computation of matrix elements leads then to an integral representation of the hypergeometric function, see for example how matrix elements of $`SL(2)`$-modules are treated in . An alternative approach is to use explicit formulas for the action of the Lie algebra $`𝔰𝔩(2)`$ and it goes as follows.
Consider the $`𝔰𝔩(2)`$-module $`V`$ with the basis $`v_k`$ indexed by all half-integers $`k+\frac{1}{2}`$ and the following action of $`𝔰𝔩(2)`$
$`Uv_k=`$ $`(z+k+\frac{1}{2})`$ $`v_{k+1},`$ (3.1)
$`Lv_k=`$ $`(2k+z+z^{})`$ $`v_k,`$ (3.2)
$`Dv_k=`$ $`(z^{}+k\frac{1}{2})`$ $`v_{k1}.`$ (3.3)
It is clear that
$$e^{\alpha U}v_k=\underset{s=0}{\overset{\mathrm{}}{}}\frac{\alpha ^s}{s!}(z+k+\frac{1}{2})_sv_{k+s}.$$
Introduce the following notation
$$(a)_s=a(a1)(a2)\mathrm{}(as+1).$$
With this notation we have
$$e^{\beta D}v_k=\underset{s=0}{\overset{\mathrm{}}{}}\frac{\beta ^s}{s!}(z^{}+k\frac{1}{2})_sv_{ks}.$$
Denote by $`\left[ij\right]_{\alpha ,\beta ,z,z^{}}`$ the coefficient of $`v_j`$ in the expansion of $`e^{\alpha U}e^{\beta D}v_i`$
$$e^{\alpha U}e^{\beta D}v_i=\underset{j}{}\left[ij\right]_{\alpha ,\beta ,z,z^{}}v_j.$$
A direct computation yields
$$\begin{array}{c}\left[ij\right]_{\alpha ,\beta ,z,z^{}}=\hfill \\ \hfill \{\begin{array}{cc}\frac{\alpha ^{ji}}{(ji)!}(z+i+\frac{1}{2})_{ji}F(\begin{array}{c}zi+\frac{1}{2},z^{}i+\frac{1}{2}\\ ji+1\end{array};\alpha \beta ),\hfill & ij,\hfill \\ \frac{\beta ^{ij}}{(ij)!}(z^{}+j+\frac{1}{2})_{ij}F(\begin{array}{c}zj+\frac{1}{2},z^{}j+\frac{1}{2}\\ ij+1\end{array};\alpha \beta ),\hfill & ij,\hfill \end{array}\end{array}$$
(3.4)
where
$$F(\begin{array}{c}a,b\\ c\end{array};z)=\underset{k=0}{\overset{\mathrm{}}{}}\frac{(a)_k(b)_k}{(c)_kk!}z^k,$$
is the Gauss hypergeometric function.
Consider now the dual module $`V^{}`$ spanned by functionals $`v_j^{}`$ such that
$$v_i^{},v_j=\delta _{ij}$$
and equipped with the dual action of $`𝔰𝔩(2)`$
$`Uv_k^{}`$ $`=(z+k\frac{1}{2})`$ $`v_{k1}^{},`$
$`Dv_k^{}`$ $`=(z^{}+k+\frac{1}{2})`$ $`v_{k+1}^{}.`$
Denote by $`\left[ij\right]_{\alpha ,\beta ,z,z^{}}^{}`$ the coefficient of $`v_j^{}`$ in the expansion of $`e^{\alpha U}e^{\beta D}v_i^{}`$
$$e^{\alpha U}e^{\beta D}v_i^{}=\underset{j}{}\left[ij\right]_{\alpha ,\beta ,z,z^{}}^{}v_j^{}.$$
We have
$$\begin{array}{c}\left[ij\right]_{\alpha ,\beta ,z,z^{}}^{}=\hfill \\ \hfill \{\begin{array}{cc}\frac{(\beta )^{ji}}{(ji)!}(z^{}+i+\frac{1}{2})_{ji}F(\begin{array}{c}z+j+\frac{1}{2},z^{}+j+\frac{1}{2}\\ ji+1\end{array};\alpha \beta ),\hfill & ij,\hfill \\ \frac{(\alpha )^{ij}}{(ij)!}(z+j+\frac{1}{2})_{ij}F(\begin{array}{c}z+i+\frac{1}{2},z^{}+i+\frac{1}{2}\\ ij+1\end{array};\alpha \beta ),\hfill & ij,\hfill \end{array}\end{array}$$
(3.5)
### 3.2 Remarks
#### 3.2.1 Periodicity
Observe that representations whose parameters $`z`$ and $`z^{}`$ are related by the tranformation
$$(z,z^{})(z+m,z^{}+m),m,$$
are equivalent. The above transformation amounts to just a renumeration of the vectors $`v_k`$. G. Olshanski pointed out that this periodicity in $`(z,z^{})`$ is reflected in a similar periodicity of various asymptotic properties of $`z`$-measures, see Sections 10 and 11 of .
#### 3.2.2 Unitarity
Recall that the $`z`$-measures are positive if either $`z^{}=\overline{z}`$ or $`z,z^{}(n,n+1)`$ for some $`n`$. By analogy with representation theory of $`SL(2)`$, these cases were called the principal and the complementary series.
Observe that in these case the above representations have a positive defined Hermitian form $`Q`$ which is invariant in the following sense
$$Q(Lu,v)=Q(u,Lv),Q(Uu,v)=Q(u,Dv).$$
The form $`Q`$ is given by
$$Q(v_k,v_k)=\{\begin{array}{cc}1\hfill & z^{}=\overline{z},\hfill \\ \frac{\mathrm{\Gamma }(z^{}+k+\frac{1}{2})}{\mathrm{\Gamma }(z+k+\frac{1}{2})}\hfill & z,z^{}(n,n+1),\hfill \end{array}$$
and $`Q(v_k,v_l)=0`$ if $`kl`$. It follows that the operators
$$\frac{i}{2}L,\frac{1}{2}(UD),\frac{i}{2}(U+D)𝔰𝔩(2),$$
which form a standard basis of $`𝔰𝔲(1,1)`$, are skew-Hermitian and hence this representation of $`𝔰𝔲(1,1)`$ can be integrated to a unitary representation of the universal covering group of $`SU(1,1)`$. This group $`SU(1,1)`$ is isomorphic to $`SL(2,)`$ and the above representations correspond to the principal and complementary series of unitary representations of the universal covering of $`SL(2,)`$, see .
### 3.3 The infinite wedge module
Consider the module $`\mathrm{\Lambda }^\frac{\mathrm{}}{2}V`$ which is, by definition, spanned by vectors
$$\delta _S=v_{s_1}v_{s_2}v_{s_3}\mathrm{},$$
where $`S=\{s_1>s_2>\mathrm{}\}+\frac{1}{2}`$ is a such subset that both sets
$$S_+=S\left(_0\frac{1}{2}\right),S_{}=\left(_0\frac{1}{2}\right)S$$
are finite. We equip this module with the inner product in which the basis $`\{\delta _S\}`$ is orthonormal. Introduce the following operators
$$\psi _k,\psi _k^{}:\mathrm{\Lambda }^\frac{\mathrm{}}{2}V\mathrm{\Lambda }^\frac{\mathrm{}}{2}V.$$
The operator $`\psi _k`$ is the exterior multiplication by $`v_k`$
$$\psi _k\left(f\right)=v_kf.$$
The operator $`\psi _k^{}`$ is the adjoint operator; it can be also given by the formula
$$\psi _k^{}\left(v_{s_1}v_{s_2}v_{s_3}\right)=\underset{i}{}(1)^{i+1}v_k^{},v_{s_i}v_{s_1}v_{s_2}\mathrm{}\widehat{v_{s_i}}\mathrm{}.$$
These operators satisfy the canonical anticommutation relations
$$\psi _k\psi _k^{}+\psi _k^{}\psi _k=1,$$
all other anticommutators being equal to $`0`$. It is clear that
$$\psi _k\psi _k^{}\delta _S=\{\begin{array}{cc}\delta _S,\hfill & kS,\hfill \\ 0,\hfill & kS.\hfill \end{array}$$
(3.6)
A general reference on the infinite wedge space is Chapter 14 of the book .
The Lie algebra $`𝔰𝔩(2)`$ acts on $`\mathrm{\Lambda }^\frac{\mathrm{}}{2}V`$. The action of $`U`$ and $`D`$ are the obvious extensions of the action on $`V`$. In terms of the fermionic operators $`\psi _k`$ and $`\psi _k^{}`$ they can be written as follows
$`U`$ $`={\displaystyle \underset{k+\frac{1}{2}}{}}(z+k+\frac{1}{2})\psi _{k+1}\psi _k^{},`$
$`D`$ $`={\displaystyle \underset{k+\frac{1}{2}}{}}(z^{}+k+\frac{1}{2})\psi _k\psi _{k+1}^{}.`$
The easiest way to define the action of $`L`$ is to set it equal to $`[D,U]`$ by definition. We obtain
$$L=2H+(z+z^{})C+zz^{},$$
where $`H`$ is the energy operator
$$H=\underset{k>0}{}k\psi _k\psi _k^{}\underset{k<0}{}k\psi _k^{}\psi _k,$$
and $`C`$ is the charge
$$C=\underset{k>0}{}\psi _k\psi _k^{}\underset{k<0}{}\psi _k^{}\psi _k.$$
It is clear that
$$C\delta _S=\left(|S_+||S_{}|\right)\delta _S$$
and, similarly,
$$H\delta _S=\left(\underset{kS_+}{}k\underset{kS_{}}{}k\right)\delta _S.$$
The charge is preserved by the $`𝔰𝔩(2)`$ action.
Consider the zero charge subspace, that is, the kernel of $`C`$
$$\mathrm{\Lambda }_0\mathrm{\Lambda }^\frac{\mathrm{}}{2}V.$$
It is spanned by vectors which, abusing notation, we shall denote by
$$\delta _\lambda =\delta _{S(\lambda )},S(\lambda )=\{\lambda _1\frac{1}{2},\lambda _2\frac{3}{2},\lambda _3\frac{5}{2},\mathrm{}\},$$
(3.7)
where $`\lambda `$ is a partition. One immediately sees that the action of $`𝔰𝔩(2)`$ on $`\{\delta _\lambda \}`$ is identical with Kerov’s operators.
### 3.4 Correlation functions
Recall that the correlation functions were defined by
$$\rho (X)=\left(\{\lambda ,X𝔖(\lambda )\}\right),$$
where the finite set
$$X=\{x_1,\mathrm{},x_s\}+\frac{1}{2}$$
is arbitrary.
The important observation is that (2.4) and (3.6) imply the following expression for the correlation functions
$$\rho (X)=(1\xi )^{zz^{}}(e^{\sqrt{\xi }D}\underset{xX}{}\psi _x\psi _x^{}e^{\sqrt{\xi }U}\delta _{\mathrm{}},\delta _{\mathrm{}}).$$
(3.8)
We apply to (3.8) the same strategy we applied to (2.5) which is to commute the operators $`e^{\sqrt{\xi }D}`$ and $`e^{\sqrt{\xi }U}`$ all the way to the right and left, respectively, and then use (2.6). From (2.7), we have for any operator $`A`$ the following identity
$$\begin{array}{c}e^{\beta D}Ae^{\alpha U}=\hfill \\ \hfill e^{\frac{\alpha }{1\alpha \beta }U}\left[e^{\frac{\alpha }{1\alpha \beta }U}e^{\beta D}Ae^{\beta D}e^{\frac{\alpha }{1\alpha \beta }U}\right](1\alpha \beta )^Le^{\frac{\beta }{1\alpha \beta }D}.\end{array}$$
(3.9)
We now apply this identity with $`\alpha =\beta =\sqrt{\xi }`$ and $`A=\psi _x\psi _x^{}`$ to obtain
$$\rho (X)=(G\underset{xX}{}\psi _x\psi _x^{}G^1\delta _{\mathrm{}},\delta _{\mathrm{}}),$$
(3.10)
where
$$G=\mathrm{exp}\left(\frac{\sqrt{\xi }}{\xi 1}U\right)\mathrm{exp}\left(\sqrt{\xi }D\right).$$
Consider the following operators
$`\mathrm{\Psi }_k`$ $`=G\psi _kG^1`$ $`={\displaystyle \underset{i}{}}\left[ki\right]\psi _i,`$ (3.11)
$`\mathrm{\Psi }_k^{}`$ $`=G\psi _k^{}G^1`$ $`={\displaystyle \underset{i}{}}\left[ki\right]^{}\psi _i^{},`$ (3.12)
with the understanding that matrix elements without parameters stand for the following choice of parameters
$$\left[ki\right]=\left[ki\right]_{\xi ^{1/2}(\xi 1)^1,\xi ^{1/2},z,z^{}},$$
(3.13)
and with same choice of parameters for $`\left[ki\right]^{}`$. The first equality in both (3.11) and (3.12) is a definition and the second equality follows from the definition of the operators $`\psi _i`$ and the definition of the matrix coefficients $`\left[ij\right]_{\alpha ,\beta ,z,z^{}}`$.
From (3.10) we obtain
$$\rho (X)=(\underset{xX}{}\mathrm{\Psi }_x\mathrm{\Psi }_x^{}\delta _{\mathrm{}},\delta _{\mathrm{}}).$$
(3.14)
Applying Wick’s theorem to (3.14), or simply unraveling the definitions in the right-hand side of (3.14), we obtain the following
###### Theorem 1.
We have
$$\rho (X)=det\left[K(x_i,x_j)\right]_{1i,js},$$
(3.15)
where the kernel $`K`$ is defined by
$$K(i,j)=(\mathrm{\Psi }_i\mathrm{\Psi }_j^{}\delta _{\mathrm{}},\delta _{\mathrm{}}).$$
Observe that
$$(\psi _l\psi _m^{}\delta _{\mathrm{}},\delta _{\mathrm{}})=\{\begin{array}{cc}1,\hfill & l=m<0,\hfill \\ 0,\hfill & \text{otherwise}.\hfill \end{array}$$
Therefore, applying the formulas (3.11) and (3.12) we obtain
###### Theorem 2.
We have
$$K(i,j)=\underset{m=1/2,3/2,\mathrm{}}{}\left[im\right]\left[jm\right]^{},$$
(3.16)
with the agreement (3.13) about matrix elements without parameters.
The formula (3.16) is the analog of the Proposition 2.9 in for the discrete Bessel kernel.
We conclude this section with the following formula which, after substituting the formulas (3.4) and (3.5) for matrix elements, becomes the formula of Borodin and Olshanski .
###### Theorem 3.
We have
$$K(i,j)=\frac{z^{}\sqrt{\xi }\left[i\frac{1}{2}\right]\left[j\frac{1}{2}\right]^{}z\frac{\sqrt{\xi }}{(\xi 1)^2}\left[i\frac{1}{2}\right]\left[j\frac{1}{2}\right]^{}}{ij},$$
(3.17)
where for $`i=j`$ the right-hand side is defined by continuity.
More generally, set
$$K(i,j)_{\alpha ,\beta }=(\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )\delta _{\mathrm{}},\delta _{\mathrm{}}).$$
where
$`\mathrm{\Psi }_k`$ $`=e^{\alpha U}e^{\beta D}\psi _ke^{\beta D}e^{\alpha U}`$ $`={\displaystyle \underset{i}{}}\left[ki\right]_{\alpha ,\beta ,z,z^{}}\psi _i`$
$`\mathrm{\Psi }_k^{}`$ $`=e^{\alpha U}e^{\beta D}\psi _k^{}e^{\beta D}e^{\alpha U}`$ $`={\displaystyle \underset{i}{}}\left[ki\right]_{\alpha ,\beta ,z,z^{}}^{}\psi _i^{}.`$
We will prove that
$$\begin{array}{c}K(i,j)_{\alpha ,\beta }=(\beta z^{}[i\frac{1}{2}]_{\alpha ,\beta ,z,z^{}}[j\frac{1}{2}]_{\alpha ,\beta ,z,z^{}}^{}\hfill \\ \hfill \alpha (\alpha \beta 1)z[i\frac{1}{2}]_{\alpha ,\beta ,z,z^{}}[j\frac{1}{2}]_{\alpha ,\beta ,z,z^{}}^{})/(ij).\end{array}$$
(3.18)
First, we treat the case $`ij`$ in which case we can clear the denominators in (3.18). From the following computation with $`2\times 2`$ matrices
$$\begin{array}{c}\left(\begin{array}{cc}1& \alpha \\ 0& 1\end{array}\right)\left(\begin{array}{cc}1& 0\\ \beta & 1\end{array}\right)\left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\left(\begin{array}{cc}1& 0\\ \beta & 1\end{array}\right)\left(\begin{array}{cc}1& \alpha \\ 0& 1\end{array}\right)=\hfill \\ \hfill \left(\begin{array}{cc}12\alpha \beta & 2\alpha (\alpha \beta 1)\\ 2\beta & 2\alpha \beta 1\end{array}\right)\end{array}$$
we conclude that
$$e^{\alpha U}e^{\beta D}Le^{\beta D}e^{\alpha U}=L+T,$$
where
$$T=2\alpha \beta L+2\beta D+2\alpha (\alpha \beta 1)U.$$
This can be rewritten as follows
$`[L,e^{\alpha U}e^{\beta D}]`$ $`=Te^{\alpha U}e^{\beta D},`$ (3.19)
$`[L,e^{\alpha U}e^{\beta D}]`$ $`=e^{\alpha U}e^{\beta D}T.`$ (3.20)
Note that
$$[L,\psi _i\psi _j^{}]=2(ij)\psi _i\psi _j^{}.$$
(3.21)
From (3.19), (3.20), and (3.21) we have
$$\begin{array}{c}[L,\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )]=\hfill \\ \hfill [T,\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )]+2(ij)\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta ).\end{array}$$
(3.22)
Since $`\delta _{\mathrm{}}`$ is an eigenvector of $`L`$ we have
$$([L,\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )]\delta _{\mathrm{}},\delta _{\mathrm{}})=0$$
Expand this equality using (3.22) and the relations
$`T\delta _{\mathrm{}}`$ $`=2\alpha \beta zz^{}\delta _{\mathrm{}}+2\alpha (\alpha \beta 1)z\delta _{\mathrm{}},`$
$`T^{}\delta _{\mathrm{}}`$ $`=2\alpha \beta zz^{}\delta _{\mathrm{}}+2\beta z^{}\delta _{\mathrm{}},`$
where $`T^{}`$ is the operator adjoint to $`T`$ and $`\delta _{\mathrm{}}`$ is the vector corresponding to the partition $`(1,0,0,\mathrm{})`$. We obtain
$$\begin{array}{c}(ij)K(i,j)_{\alpha ,\beta }=\beta z^{}(\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )\delta _{\mathrm{}},\delta _{\mathrm{}})\hfill \\ \hfill \alpha (\alpha \beta 1)z(\mathrm{\Psi }_i(\alpha ,\beta )\mathrm{\Psi }_j^{}(\alpha ,\beta )\delta _{\mathrm{}},\delta _{\mathrm{}})\end{array}$$
In order to obtain (3.18) for $`ij`$, it now remains to observe that
$`(\psi _l\psi _m^{}\delta _{\mathrm{}},\delta _{\mathrm{}})`$ $`=\{\begin{array}{cc}1,\hfill & l=\frac{1}{2},m=\frac{1}{2},\hfill \\ 0,\hfill & \text{otherwise},\hfill \end{array}`$
$`(\psi _l\psi _m^{}\delta _{\mathrm{}},\delta _{\mathrm{}})`$ $`=\{\begin{array}{cc}1,\hfill & l=\frac{1}{2},m=\frac{1}{2},\hfill \\ 0,\hfill & \text{otherwise}.\hfill \end{array}`$
In the case $`i=j`$ we argue by continuity. It is clear from (3.16) that $`K(i,j)`$ is an analytic function of $`i`$ and $`j`$ and so is the right-hand side of (3.17). The passage from (3.16) to (3.17) is based on the fact that the product $`i`$ times $`\left[im\right]_{\alpha ,\beta ,z,z^{}}`$ is a linear combination of $`\left[im\right]_{\alpha ,\beta ,z,z^{}}`$ and $`\left[im\pm 1\right]_{\alpha ,\beta ,z,z^{}}`$ with coefficients which are linear functions of $`m`$. Since the matrix coefficients are, essentially, the hypergeometric function, such a relation must hold for any $`i`$, not just half-integers. Hence, (3.16) and (3.17) are equal for any $`ij`$, not necessarily half-integers. Therefore, they are equal for $`i=j`$.
### 3.5 Rim-hook analogs
The same principles apply to rim-hook analogs of the $`z`$-measures which were also considered by S. Kerov .
Recall that a rim hook of a diagram $`\lambda `$ is, by definition, a skew diagram $`\lambda /\mu `$ which is connected and lies on the rim of $`\lambda `$. Here connected means that the squares have to be connected by common edges, not just common vertices. Rim hooks of a diagram $`\lambda `$ are in the following 1-1 correspondence with the squares of $`\lambda `$: given a square $`\mathrm{}\lambda `$, the corresponding rim hook consists of all squares on the rim of $`\lambda `$ which are (weakly) to the right of and below $`\mathrm{}`$. The length of this rim hook is equal to the hook-length of $`\mathrm{}`$.
The entire discussion of the previous section applies to the more general operators
$`U_rv_k=`$ $`\left(z+\frac{k}{r}+\frac{1}{2}\right)`$ $`v_{k+r},`$
$`L_rv_k=`$ $`\left(\frac{2k}{r}+z+z^{}\right)`$ $`v_k,`$
$`D_rv_k=`$ $`\left(z^{}+\frac{k}{r}\frac{1}{2}\right)`$ $`v_{kr},`$
which satisfy the same $`𝔰𝔩(2)`$ commutation relations. The easiest way to check the commutation relations is to consider $`\frac{k}{r}`$ rather than $`k`$ as the index of $`v_k`$; the above formulas then become precisely the formulas (3.1). The operator $`U_r`$ acts on the basis $`\{\delta _\lambda \}`$ as follows
$$U_r\delta _\lambda =\underset{\mu =\lambda +\text{ rim hook}}{}(1)^{\text{height}+1}\left(z+\frac{1}{r^2}\underset{\mathrm{}\text{rim hook}}{}c(\mathrm{})\right)\delta _\mu ,$$
(3.23)
where the summation is over all partitions $`\mu `$ which can be obtained from $`\lambda `$ by adding a rim hook of length $`r`$, height is the number of horizontal rows occupied by this rim hook and $`c(\mathrm{})`$ stands, as usual for the content of the square $`\mathrm{}`$. Similarly, the operator $`D_r`$ removes rim hooks of length $`r`$. These operators were considered by Kerov .
It is clear that the action of the operators $`e^{\alpha U_r}`$ and $`e^{\beta D_r}`$ on a half-infinite wedge product like
$$v_{s_1}v_{s_2}v_{s_3}\mathrm{},$$
essentially (up to a sign which disappears in formulas like (3.8)) factors into the tensor product of $`r`$ separate actions on
$$\underset{s_ik+\frac{1}{2}modr}{}v_{s_i},k=0,\mathrm{},r1.$$
Consequently, the analogs of the correlation functions (3.8) have again a determinantal form with a certain kernel $`K_r(i,j)`$ which has the following structure. If $`ijmodr`$ then $`K_r(i,j)`$ is essentially the kernel $`K(i,j)`$ with rescaled arguments. Otherwise, $`K_r(i,j)=0`$.
This factorization of the action on $`\mathrm{\Lambda }^\frac{\mathrm{}}{2}V`$ is just one more way to understand the following well-known phenomenon. Let $`𝕐_r`$ be the partial ordered set formed by partitions with respect to the following ordering: $`\mu _r\lambda `$ if $`\mu `$ can be obtained from $`\lambda `$ by removing a number of rim hooks with $`r`$ squares. The minimal elements of $`𝕐_r`$ are called the $`r`$-cores. The $`r`$-cores are precisely those partitions which do not have any hooks of length $`r`$. We have
$$𝕐_r\underset{r\text{-cores}}{}(𝕐_1)^r$$
(3.24)
as partially ordered sets. Here the Cartesian product $`(𝕐_1)^r`$ is ordered as follows:
$$(\mu _1,\mathrm{},\mu _r)(\lambda _1,\mathrm{},\lambda _r)\mu _i_1\lambda _i,i=1,\mathrm{},r,$$
and the partitions corresponding to different $`r`$-cores are incomparable in the $`_r`$-order. Combinatorial algorithms which materialize the isomorphism (3.24) are discussed in Section 2.7 of the book . The $`r`$-core and the $`r`$-tuple of partitions which the isomorphism (3.24) associates to a partition $`\lambda `$ are called the $`r`$-core of $`\lambda `$ and the $`r`$-quotient of $`\lambda `$. Among more recent papers dealing with $`r`$-quotients let us mention where an approach similar to the use of $`\mathrm{\Lambda }^\frac{\mathrm{}}{2}V`$ is employed, an analog of the Robinson-Schensted algorithm for $`𝕐_r`$ is discussed, and further references are given.
The factorization (3.24) and the corresponding analog of the Robinson-Schensted algorithm play the central role in the recent paper , see also .
|
warning/0002/astro-ph0002256.html
|
ar5iv
|
text
|
# Theory of pixel lensing towards M31 I: the density contribution and mass of MACHOs
## 1 Introduction
### 1.1 Conventional microlensing: landmarks and limitations
The detection of the gravitational microlensing effect due to compact objects in the Galaxy is undoubtedly one of the great success stories in astrophysics over the past decade. Surveys have discovered around 20 candidates towards the Magellanic clouds and several hundred towards the Galactic Bulge \[Udalski et al. 1994, Alcock et al. 1997, Alard & Guibert 1997, Lasserre et al. 1999, Alcock et al. 2000\]. Amongst these candidates a number of exotic lensing phenomena have been catalogued, such as parallax effects, binary lensing (including spectacular examples of caustic-crossing events), and finite source-size effects. These discoveries are facilitated by coordinated follow-up campaigns such as PLANET \[Albrow et al. 1998\] and MPS \[Rhie et al. 1999\] which act on microlensing alerts broadcast by the survey teams. The absence of certain microlensing signals has also yielded a clearer insight into the nature of halo dark matter. The null detection of short duration events towards the Large Magellanic Cloud (LMC) by the EROS and MACHO surveys indicates that, for a range of plausible halo models, massive compact halo objects (MACHOs) within the mass interval $`10^710^3\text{M}_{\mathrm{}}`$ provide less than a quarter of the dark matter \[Alcock et al. 1998\]. This is an important result when set against the current insensitivity of other techniques to this mass range.
Despite these successes a number of unsolved problems remain. The optical depth measured towards the Galactic Bulge is at least a factor two larger than can be accommodated by theoretical models (e.g. Bissantz et al. 1997; Sevenster et al. 1999). Towards the LMC the rate of detected events is consistent with the discovery of a significant fraction of the halo dark matter. However, the implied lens mass range ($`0.11\text{M}_{\mathrm{}}`$) is not easily reconciled with existing constraints on baryonic dark matter candidates \[Carr 1994\], though the MACHOs need not necessarily be baryonic. Furthermore, the discovery of two possible binary caustic-crossing events towards the LMC and the Small Magellanic Cloud (SMC) has thrown into question the very existence of MACHOs. Their caustic-crossing timescales, which provide an indicator of their line-of-sight position, seem to exclude either as being of halo origin, a statistically unlikely occurrence if the halo comprises a significant MACHO component \[Kerins & Evans 1999\]. As a result, there is a growing body of opinion that all events observed so far towards the LMC and SMC may reside in the clouds themselves. However, this explanation is itself problematic because it requires that the clouds must either have a higher MACHO fraction than the Galaxy or comprise substantial but diffuse stellar components not in hydrodynamical equilibrium (Evans & Kerins 2000, and references therein).
These problems highlight two principal constraints on the ability of conventional microlensing experiments to determine the nature and distribution of MACHOs in the halo. The first limitation is their inefficiency in differentiating between lensing by MACHOs and self-lensing by the source population, since for most events one observes only a duration and a position on the sky. These observables are only weakly correlated with the location of the events along the line of sight. The second constraint is the limited number of suitable lines of sight through the halo. Conventional microlensing surveys require rich yet resolved stellar fields and are thus limited to just two lines of sight, the LMC and SMC, with which to probe MACHOs. The line of sight to the Galactic Bulge is dominated by bulge and disc lensing. The paucity of halo lines of sight, together with the rather weak dynamical and kinematical constraints on Galactic halo structure, also diminishes the prospect of being able to decouple information on the Galactic distribution function and MACHO mass function.
### 1.2 Beyond the Galaxy: a new target, a new technique
The possibility of detecting MACHOs in an external galaxy, specifically M31, was initially explored by Crotts (1992) and by Baillon et al. (1993). Crotts (1992) pointed out that the high inclination of the disc of M31 would result in an asymmetry in the observed rate of microlensing if the disc is surrounded by a MACHO halo, as illustrated in Figure 1. The fact that the M31 MACHO microlensing rate should be lower towards the near side of the disc than the far side, which lies behind a larger halo column density, means that the presence of MACHOs in M31 can be established unambiguously. In particular, neither variable stars nor stellar self-lensing events in the disc of M31 should exhibit near-far asymmetry. Additionally, the external vantage point serves to reduce systematic model uncertainties in two ways. Firstly, it permits a more accurate determination of the rotation curve and surface brightness profile than is possible for the Galaxy, which reduces the prior parameter space of viable galactic models. Secondly, it provides many independent lines of sight through the halo of M31, allowing the MACHO distribution across the face of the disc to be mapped and thus the halo distribution function to be constrained more or less directly.
As pointed out by Baillon et al. (1993), another appeal of directing observations towards more distant large galaxies like M31 is the increase in the number of potential source stars, more than a factor of one thousand over the number available in the LMC and SMC, and all confined to within a few square degrees. However, this also presents a fundamental problem in that the source stars are resolved only whilst they are lensed (and even then only if the magnification is sufficiently large). The presence of many stars per detector pixel means it is often impossible to identify which is being lensed. Furthermore, the flux contribution of the unlensed stars dilutes the observed flux variation due to microlensing. Nonetheless, Baillon et al. (1993) determined from numerical simulations that the number of observable events, due to either the lensing of bright stars or high magnification events, is expected to be large. As a result of these studies, the Andromeda Galaxy Amplified Pixel Experiment (AGAPE) and another group, Columbia-VATT, commenced observing programs towards M31 \[Ansari et al. 1997, Crotts & Tomaney 1997\].
One of the biggest technical difficulties facing surveys which look for variable sources against unresolved stellar fields is how to distinguish between flux variations due to changing observing conditions and intrinsic variations due to microlensing or stellar variability. For example, changes in seeing induce variations in the detected flux within a pixel. One must also deal with the consequences of positional misalignment between exposures, spatial and temporal variations in the point spread function (PSF) and photometric variations due to atmospheric transparency and variable sky background.
AGAPE has employed the Pixel Method to cope with the changing observing conditions \[Ansari et al. 1997\]. AGAPE thoroughly tested this technique with a three-year campaign using the 2m Bernard Lyot telescope at Pic du Midi from 1994 to 1996 \[Ansari et al. 1997, Ansari et al. 1999, Le Du 2000\]. Six fields covering about 100 arcmin<sup>2</sup> centred on the bulge of M31 were monitored. Whilst the field of view was insufficient to conclude much about the nature of MACHOs, 19 candidate events were detected, though it is still premature to rule out many of them being intrinsically variable sources, such as Miras or novae. One event, AGAPE Z1, appears to be a convincing lensing candidate as its flux increase and colour are inconsistent with that of a Mira or nova \[Ansari et al. 1999\]. A longer baseline is needed to determine how many of the other candidates are due to microlensing.
A major observing programme began on the 2.5m Isaac Newton Telescope (INT) in La Palma in the Autumn of 1999, with a run of one hour per night for almost sixty nights over six months. The POINT-AGAPE collaboration is a joint venture between UK-based astronomers and AGAPE (where POINT is an acronym for “Pixel-lensing Observations with INT”). We are exploiting the 0.3 deg<sup>2</sup> field of view of the INT Wide-field Camera (WFC) to map the distribution of microlensing events across a large region of the M31 disc. Our initial observations of M31 with the INT employed a $`V`$ filter and the simulations reported here have been undertaken with parameters appropriate to V-band observations. The strategy employed for the actual M31 monitoring campaign involves observations in three bands, $`g`$, $`r`$, and $`i`$ \[very similar to the bands employed by SLOAN \[Fukugita et al. 1996\]\]. The multi-colour observations will improve our ability to discriminate against variable stars and the $`gri`$-filter plus CCD combination offers a significant improvement in sensitivity (the $`g`$-band zero-point is some 0.4 magnitudes fainter than that for $`V`$). The simulation parameters are thus somewhat conservative in this regard. The programme is being conducted in consort with the Microlensing Exploration of the Galaxy and Andromeda (MEGA) survey \[Crotts, Uglesich & Gyuk 1999\], the successor program to Columbia-VATT. Whilst POINT-AGAPE and MEGA are sharing the data, different techniques are being employed to search for microlensing events. Henceforth we use the term pixel lensing \[Gould 1996\] to describe microlensing against unresolved stellar fields, regardless of the detection technique.
Whilst the technical viability of pixel lensing is now clearly established, a number of important theoretical issues are still outstanding. The principal concern is that the main observable in classical microlensing, the Einstein crossing time, is generally not accessible in pixel lensing. The Einstein crossing time is directly related to the lens mass, its transverse velocity and the observer–lens–source geometry. In pixel lensing the observed timescale depends upon additional factors, such as the local surface brightness and the source luminosity and magnification, so the dependence on lens parameters is much weaker than for classical microlensing.
The first detailed study of pixel lensing was undertaken by Gould (1996). He defined two regimes: a semi-classical regime in which the source star dominates the pixel flux and the observable timescale provides a fair tracer of the Einstein crossing time; and the “spike” regime where only high-magnification events are identified, and the timescales are only weakly correlated with the underlying Einstein crossing duration. Remarkably, Gould showed that, despite the loss of timescale information, in the spike regime one can still measure the microlensing optical depth. Using Gould’s formalism, Han (1996) provided the first pixel event rate estimates for the M31 line of sight. However, Gould’s formalism assumes a fixed sampling rate and unchanging observing conditions. As such it is of limited applicability to a ground-based observing program. Gondolo (1999) has proposed an optical depth estimator based on the observed pixel event timescale. Whilst this estimator can be readily employed by a ground-based campaign, it is somewhat sensitive to the shape of the source luminosity function and is valid only to the extent that this can be taken to be the same for all source components. More recently, Baltz & Silk (1999) derived expressions for the pixel rate and timescale distribution in terms of the observable timescale, rather than the Einstein crossing time. Again, their study assumes constant sampling and observing conditions, as would be the case for space-borne programmes.
Whilst these studies provide a solid foundation for predictions of pixel-lensing quantities (i.e. timescales, rates and optical depth), none of them address to what extent one can constrain galactic and lens parameters, in particular the MACHO mass, from pixel lens observables. Gyuk & Crotts (2000) have shown that a reliable measure of the optical depth from pixel lensing can be used to probe the core radius and flattening of the M31 MACHO halo.
In this paper we quantitatively assess the degree to which the POINT-AGAPE campaign directed towards M31 will constrain the fractional contribution and mass of the MACHOs. Since the answer inevitably depends upon the assumed galactic distribution function, we focus attention here on the simple case of spherically-symmetric near-isothermal halo models. The line of sight towards M31 is sensitive to two MACHO populations, our own and that in M31 itself, so we investigate the extent to which they can be distinguished and probed independently. We also model the expected background due to variable stars and lenses residing in the disc and bulge of M31.
The plan of the paper is as follows. In Section 2 we summarize the basic principles of pixel lensing, with emphasis on the differences between pixel lensing and classical microlensing. We describe our Monte-Carlo pixel-lensing simulations in Section 3, including our event selection criteria and the incorporation of realistic sampling and observing conditions. In Section 4 we construct a reference model for the lens and source populations in the halo of the Galaxy and the halo, disc and bulge of M31, seeking consistency with the observed M31 rotation curve and surface brightness profiles. In Section 5 we present predictions for the POINT-AGAPE survey based on our simulations. In Section 6 we use the simulations to generate artificial data-sets and we investigate to what extent the MACHO mass and fractional contribution in the two galaxies can be recovered from the data. The results are summarized and discussed in Section 7.
## 2 Principles of pixel lensing
We review here some of the main aspects of pixel lensing and its differences with classical microlensing. A more detailed overview can be found in Gould (1996).
### 2.1 Detecting pixel events
Whilst in classical microlensing one monitors individual sources, in pixel lensing the sources are resolved only whilst they are lensed. We can therefore only monitor the flux in each detector element rather than the flux from individual sources. If a star is magnified sufficiently due to a lens passing close to its line of sight, then the total flux in the detector element containing the source star (due to the lensed star, other nearby unlensed stars and the sky background) will rise significantly above the noise level and be recorded as an event.
Before treating seeing variations the sequence of images must be geometrically and photometrically aligned with respect to some reference image, $``$, as described in Ansari et al. (1997). The variations remaining after alignment are primarily due to changes in seeing and source flux, including microlensing events. To minimize the effects of seeing we define our base detector element to be a superpixel: a square array of pixels. A superpixel is defined for each pixel, with that pixel lying at the centre, so that neighbouring superpixels overlap with an offset of one pixel. The optimal size for the superpixel array is set by the ratio of the size of the seeing disc on images obtained in poor seeing to the individual pixel size. The INT Wide-field Camera (WFC) has a pixel scale corresponding to $`0\stackrel{}{.}33`$ on the sky, whilst poor seeing at La Palma is $`2\mathrm{}`$. Adopting a very conservative value of $`2\stackrel{}{.}4`$ for the worst seeing leads to an optimized choice of $`7\times 7`$ pixels for the superpixel array. A larger array would overly dilute source variations, whilst a smaller array would be overly sensitive to changing observing conditions.
Whilst seeing variations are reduced by binning the photon count into superpixels, this by itself is not enough to make them negligible. Residual variations are minimized by the Pixel Method, in which a simple, empirically-derived statistical correction is applied to each image to match it to the characteristics of the reference image $``$. The Pixel Method is discussed in Ansari et al. (1997) and described in detail by Le Du (2000). The method strikes a good balance between computational efficiency and optimal signal-to-noise ratio, with the resulting noise level approaching the photon noise limit.
After alignment and seeing corrections the excess superpixel photon count $`\mathrm{\Delta }N_{\mathrm{pix}}`$ on an image $`i`$ obtained at epoch $`t_i`$ due to an ongoing microlensing event is
$$\mathrm{\Delta }N_{\mathrm{pix}}(t_i)N_{\mathrm{bl}}[A_{\mathrm{pix}}(t_i)1]=f_{\mathrm{see}}N_\mathrm{s}[A(t_i)1].$$
(1)
Here $`N_\mathrm{s}`$ and $`N_{\mathrm{bl}}`$ are the source and baseline photon counts in the absence of lensing, $`A`$ is the source magnification factor due to lensing and $`f_{\mathrm{see}}`$ is the fraction of the seeing disc contained within the superpixel. The baseline photon count, $`N_{\mathrm{bl}}=N_{\mathrm{gal}}()+N_{\mathrm{sky}}()`$, is the sum of the local M31 surface brightness (including $`N_\mathrm{s}`$) and sky background contributions on the reference image. Whilst the quantities $`N_{\mathrm{bl}}`$ and $`f_{\mathrm{see}}N_\mathrm{s}(A1)`$ can be determined independently, $`N_\mathrm{s}`$ and $`A`$ cannot in general be inferred separately. It is therefore convenient to define $`A_{\mathrm{pix}}`$ as the superpixel count variation factor, which acts as the observable analogue of $`A`$.
The superpixel noise on image $`i`$ is
$$\sigma _i=\mathrm{max}[\sigma _\mathrm{T}(t_i),\alpha _iN_{\mathrm{pix}}(t_i)^{1/2}],$$
(2)
where
$$N_{\mathrm{pix}}(t_i)=\mathrm{\Delta }N_{\mathrm{pix}}(t_i)+N_{\mathrm{sky}}(t_i)+N_{\mathrm{gal}}$$
(3)
refers to the superpixel photon count on image $`i`$ prior to correction and, similarly, $`N_{\mathrm{sky}}`$ and $`N_{\mathrm{gal}}`$ are the uncorrected sky background and galaxy surface brightness contributions. The threshold noise level $`\sigma _\mathrm{T}`$ is determined by the superpixel flux stability, and the scaling factor $`\alpha _i`$ takes account of the fact that the Pixel Method is not photon-noise limited. A preliminary analysis of a sequence of INT WFC images taken in 1998 demonstrated a flux stability level of $`0.10.3\%`$ \[Melchior 1999\]. We therefore adopt a conservative minimum noise level of $`\sigma _\mathrm{T}=2.5\times 10^3N_{\mathrm{bl}}`$ for our simulations. We also apply a constant scaling factor $`\alpha _i=1.2`$, which is a little larger than typical for the AGAPE Pic du Midi data \[Le Du 2000\]. In reality $`\alpha _i`$ varies slightly between images though we neglect this variation in our simulations.
Note that $`N_{\mathrm{gal}}`$ in equation (3) is constant, despite the changing observing conditions. Though some variable fraction of the local patch of surface brightness is dispersed over neighbouring superpixels, the same amount of surface brightness leaks into the superpixel from neighbouring patches, so there is no net variation. The variation in $`N_{\mathrm{sky}}`$ results from changing moonlight and atmospheric transparency.
We regard a signal as being statistically significant if it occurs at a level $`3\sigma _i`$ above the baseline count $`N_{\mathrm{bl}}`$. Our estimate of $`N_{\mathrm{bl}}`$ must be obtained from a sequence of images and operationally is defined to be the minimum of a sliding average of superpixel photon counts over ten consecutive epochs. In order for a signal to be detected on image $`i`$ we therefore require a superpixel count variation factor $`A_{\mathrm{pix}}(t_i)1+3\sigma _i/N_{\mathrm{bl}}`$. From equation (1), a microlensed source satisfies this inequality provided that it is magnified by a factor exceeding
$$A_{\mathrm{min}}(t_i)=1+\frac{3\sigma _i}{f_{\mathrm{see}}N_\mathrm{s}}.$$
(4)
A special case of equation (4) occurs when $`\sigma _i=\sigma _\mathrm{T}`$, giving a threshold magnification of
$$A_\mathrm{T}=1+0.0075\frac{N_{\mathrm{bl}}}{f_{\mathrm{see}}N_\mathrm{s}}.$$
(5)
The extent to which residual temporal variations in $`f_{\mathrm{see}}`$ and $`N_{\mathrm{bl}}`$ remain after image processing determines the factor by which $`\sigma _i`$ exceeds the photon noise limit, so this excess noise is explicitly accounted for in equation (4).
Equation (4) illustrates some important characteristics of pixel lensing. Firstly, pixel lensing does not depend directly on the local surface brightness or sky background, but it does depend on their contribution to the noise $`\sigma _i`$. Secondly, if the exposure time $`T_{\mathrm{exp}}`$ is short, or the source star constitutes only a small fraction of the superpixel flux, so that $`N_\mathrm{s}\sigma _i`$, only rare high-magnification events are detected. The relationship between lens magnification and lens–source impact distance (measured in the lens plane) is as for the classical case:
$$A=\frac{u^2+2}{u\sqrt{u^2+4}}$$
(6)
where $`u`$ is the impact distance in units of the Einstein radius. The maximum value for the impact distance can be obtained by inverting equation (6) for $`A=A_{\mathrm{min}}`$:
$$u_{\mathrm{max}}=2^{1/2}\left[\frac{A_{\mathrm{min}}}{\sqrt{A_{\mathrm{min}}^21}}1\right]^{1/2}A_{\mathrm{min}}^1(A_{\mathrm{min}}10).$$
(7)
For pixel lensing in M31 we are often in the regime where $`N_\mathrm{s}\sigma _i`$ because the source flux is much less than that of the galaxy and background, so it is not unusual to require $`A_{\mathrm{min}}10`$. In this case equations (4) and (7) imply
$$u_{\mathrm{max}}\frac{f_{\mathrm{see}}N_\mathrm{s}}{3\sigma _i}<\frac{f_{\mathrm{see}}N_\mathrm{s}}{3N_{\mathrm{pix}}^{1/2}}(A_{\mathrm{min}}10),$$
(8)
Since $`u_{\mathrm{max}}1`$ \[typically $`u_{\mathrm{max}}𝒪(10^210^3)`$\] only a small fraction of classical ($`u1`$) microlensing events are detectable.
The dependence of $`u_{\mathrm{max}}`$ on $`N_\mathrm{s}`$ means that the pixel event rate depends on the source luminosity function $`\varphi (M)`$, the number density of sources in the absolute magnitude interval $`(M,M+dM)`$. We can compute a theoretical upper limit, $`\mathrm{\Gamma }_\mathrm{p}`$, for the pixel-lensing rate at sky coordinate $`(x,y)`$ by taking $`A_{\mathrm{min}}=A_\mathrm{T}`$ so that $`u_{\mathrm{max}}=u(A_\mathrm{T})=u_\mathrm{T}`$. In this case
$$\mathrm{\Gamma }_\mathrm{p}(x,y)=u_\mathrm{T}(x,y)_\varphi \mathrm{\Gamma }_\mathrm{c}(x,y),$$
(9)
where $`x`$ and $`y`$ are Cartesian coordinates centred on M31 and aligned respectively along the major and minor axes of the projected light profile. We define $`y`$ to be positive towards the near side of the disc. The quantity $`\mathrm{\Gamma }_\mathrm{c}`$ is the classical ($`u1`$) event rate integrated over lens and source populations \[Griest 1991, Kiraga & Paczyński 1994\], and
$$u_\mathrm{T}(x,y)_\varphi \frac{u_\mathrm{T}(M,x,y)\varphi (M)𝑑M}{\varphi (M)𝑑M}$$
(10)
is the mean threshold impact parameter at $`(x,y)`$ averaged over $`\varphi `$.
Whilst useful in providing a rough order of magnitude estimate, $`\mathrm{\Gamma }_\mathrm{p}`$ cannot be compared directly with observations because it assumes perfect sensitivity to all event durations and it also assumes that observing conditions are unchanging. Since one usually has $`A_{\mathrm{min}}>A_\mathrm{T}`$, equation (10) also tends to overestimate the true mean pixel-lensing cross-section. One can regard $`\mathrm{\Gamma }_\mathrm{p}`$, evaluated under the best observing conditions, as providing a strict theoretical upper limit to the observed event rate, in much the same way as $`\mathrm{\Gamma }_\mathrm{c}`$ provides an upper limit to the observed rate in classical lensing. In Section 3 we set about obtaining a more realistic estimate of the observed pixel lensing rate.
### 2.2 Degenerate and non-degenerate regimes
In classical microlensing the most important observable is the Einstein radius crossing time, since this is directly related to the position, motion and mass of the lens. Can we obtain similar information from the duration of pixel events?
For a lens moving at constant velocity across the line of sight, $`u`$ evolves with time $`t`$ as in the classical case:
$$u(t)^2=u(t_0)^2+\left(\frac{tt_0}{t_\mathrm{e}}\right)^2,$$
(11)
where $`t_0`$ is the epoch of minimum impact distance and $`t_\mathrm{e}`$ is the Einstein radius crossing time. From equations (6) and (11), $`t_\mathrm{e}`$ gives the timescale over which the source magnification $`A`$ varies significantly. For large magnifications $`uA^1`$ from equation (7), and inserting equation (11) into equation (1) gives
$$\mathrm{\Delta }N_{\mathrm{pix}}(t)\frac{f_{\mathrm{see}}A_{\mathrm{max}}N_\mathrm{s}}{\sqrt{1+\left(\frac{tt_0}{t_\mathrm{e}A_{\mathrm{max}}^1}\right)^2}}[A(t)10],$$
(12)
where $`A_{\mathrm{max}}A(t_0)`$ is the maximum magnification. We infer that in pixel lensing the timescale over which the signal varies significantly is $`t_\mathrm{e}A_{\mathrm{max}}^1`$ rather than $`t_\mathrm{e}`$. This means that, in the high-magnification regime, the pixel-lensing timescale bears little relation to $`t_\mathrm{e}`$. We also see that the light-curve is degenerate under transformations $`A_{\mathrm{max}}\alpha A_{\mathrm{max}}`$, $`N_\mathrm{s}N_\mathrm{s}/\alpha `$ and $`t_\mathrm{e}\alpha t_\mathrm{e}`$ \[Woz̀niak & Paczyǹski 1997\]. So neither $`t_\mathrm{e}`$, $`A_{\mathrm{max}}`$ nor $`N_\mathrm{s}`$ can be determined independently. It may sometimes be possible to break this degeneracy by looking at the wings of the light-curve \[Baltz & Silk 1999\], where differences between the true magnification and its degenerate form can become apparent. From equation (6), the difference between the exact expression for $`A(u)1`$ appearing in equation (1) and its degenerate approximation, $`u^1`$, is
$$\mathrm{\Delta }(A1)=\frac{u^2+2}{u\sqrt{u^2+4}}1\frac{1}{u}\frac{3u}{8}1(u1).$$
(13)
To discriminate reliably (say at the $`3\sigma `$ level) between the degenerate and non-degenerate cases requires $`f_{\mathrm{see}}N_\mathrm{s}|\mathrm{\Delta }(A1)|>3\sigma _i`$, so for the high-magnification regime we can write the condition for non-degeneracy as
$$\sigma _i\frac{f_{\mathrm{see}}N_\mathrm{s}}{3}(u1).$$
(14)
Equation (14) demands that the superpixel noise be no greater than the contribution of the unlensed source to the superpixel flux. In general this will not be the case, so observations will not be able to break the light-curve degeneracy and thus will not directly probe the Einstein crossing time.
Since the underlying duration $`t_\mathrm{e}`$ is not generally measurable we use the observed full-width half-maximum (FWHM) event duration:
$$t_{\mathrm{FWHM}}=2\sqrt{2}t_\mathrm{e}\left[\frac{a+2}{\sqrt{a^2+4a}}\frac{a+1}{\sqrt{a^2+2a}}\right]^{1/2},$$
(15)
where $`a=A_{\mathrm{max}}1`$. Since $`A_{\mathrm{max}}`$ for detected events is typically larger in regions of higher surface brightness, and for fainter stars, $`t_{\mathrm{FWHM}}`$ is correlated both with the disc surface brightness and the source luminosity function. This means that it is less strongly correlated than $`t_\mathrm{e}`$ with the lens mass and velocity and the lens and source distances.
The observed duration, $`t_{\mathrm{FWHM}}`$, does not afford us with as direct a probe of lens parameters as $`t_\mathrm{e}`$. We are therefore forced to rely on other observables, such as spatial distribution, in order to probe the underlying MACHO properties. For M31 MACHOs one can test for near-far asymmetry in the event rate \[Crotts 1992\]. For Galaxy MACHOs there is no comparable signature. Looking from the centre of the Galaxy towards M31 the halo density distribution in the two galaxies is highly symmetric about the observer–source midpoint. Since the microlensing geometry is also symmetric about the midpoint the timescale distributions for Galaxy and M31 MACHOs are similar for the same mass function. Since our displacement from the Galactic centre is only 8 kpc (small compared to the scale of the haloes and the Galaxy–M31 separation) this geometrical symmetry is largely preserved at our location. However, the Galaxy MACHO distribution ought to be less concentrated than that of stellar lenses. One might hope to see this as an excess of events at faint isophotes which remains the same towards both the near and far disc. If MACHOs exist, the overall pixel-lens distribution will be superposition of several lens populations (Galaxy halo, M31 halo, disc and bulge) together with variable stars which, at least in the short term, appear indistinguishable from microlensing. The task of disentangling each is therefore potentially tricky.
## 3 Simulating pixel events
A straightforward method for probing the lens populations is to construct simulations of the expected distribution of events for a particular telescope configuration, set of observing conditions and selection criteria and then compare these predictions to observations. To this end we have constructed a detailed simulation of a realistic pixel-lensing experiment.
Our simulation works by first computing a theoretical upper limit to the pixel rate for assumed M31 and Galaxy models. This estimate provides the basis for generating trial pixel microlensing events for which light-curves are constructed and selection criteria applied. The precise details of our input galaxy models are discussed in Section 4; in this section we lay down the general framework for the simulation. For each generated trial event, a pixel light-curve is constructed using a realistic distribution of observing epochs interrupted by poor weather and scheduling constraints. The effects of the sky background and seeing are explicitly taken into account in computing flux realizations and errors for each “observation”. The observing sequence is then examined to see whether the event passes the detection criteria — if it does, then the trial counts as a detected event. The simulation is terminated once $`10^4`$ events are detected or $`10^6`$ trials generated, whichever is reached first. The fraction of trial events which are detected is used to compute the observed pixel rate. The statistical error on the rate determination is typically about $`3\%`$.
### 3.1 Generating trial events
As the starting point for our simulation we use the theoretical pixel event rate as a function of position, $`\mathrm{\Gamma }_\mathrm{p}(x,y)`$, defined by equation (9). This quantity, evaluated for the best seeing conditions, always provides an upper limit to the detection rate at a given location and is therefore convenient to use to generate trial events. We compute $`\mathrm{\Gamma }_{\mathrm{p},j}`$ over a grid of locations $`(x,y)`$ for each combination $`j`$ of lens and source population. Near the centre of M31, $`j=1\mathrm{}8`$ since there are two source populations (M31 disc and bulge) and four lens populations (Galaxy halo, M31 halo, M31 disc and M31 bulge). Beyond 8 kpc the M31 bulge is not in evidence, so $`j=1\mathrm{}3`$.
Given the grid of $`\mathrm{\Gamma }_{\mathrm{p},j}(x,y)`$, one can write the probability of observing an event at location $`(x,y)`$ as
$$P(x,y)\mathrm{\Delta }x\mathrm{\Delta }y\underset{j}{}S_j(x,y)\mathrm{\Gamma }_{\mathrm{p},j}(x,y),$$
(16)
where $`S_j`$ is the source surface density at $`(x,y)`$ for lens–source configuration $`j`$, and $`\mathrm{\Delta }x`$ and $`\mathrm{\Delta }y`$ are the local $`x`$ and $`y`$ grid spacings (required only for non-uniform grids). $`P(x,y)`$ therefore reflects the total event rate in a box of area $`\mathrm{\Delta }x\mathrm{\Delta }y`$ centred on $`(x,y)`$. The box should be sufficiently small that $`S_j(x,y)`$ and $`\mathrm{\Gamma }_{\mathrm{p},j}(x,y)`$ provide good estimates of the source density and theoretical rate anywhere within it. Having fixed the event location, $`\mathrm{\Gamma }_{\mathrm{p},j}`$ is then used to select the lens and source components from the probability distribution
$$P(j)=\frac{S_j(x,y)\mathrm{\Gamma }_{\mathrm{p},j}(x,y)}{_jS_j(x,y)\mathrm{\Gamma }_{\mathrm{p},j}(x,y)}.$$
(17)
Once the event location and lens and source populations have been decided, the next choice is the line-of-sight distances to the lens, $`D_\mathrm{l}`$, and source, $`D_\mathrm{s}`$:
$`P(D_\mathrm{s})`$ $``$ $`\rho _\mathrm{s}(D_\mathrm{s})D_\mathrm{s}^{3/2}{\displaystyle _0^{D_\mathrm{s}}}P(D_\mathrm{l})𝑑D_\mathrm{l}`$
$`P(D_\mathrm{l})`$ $``$ $`\rho _\mathrm{l}(D_\mathrm{l})\sqrt{D_\mathrm{l}(D_\mathrm{s}D_\mathrm{l})},`$ (18)
where $`\rho _\mathrm{l}`$ and $`\rho _\mathrm{s}`$ are respectively the lens and source mass densities. These distributions reflect the dependency of the microlensing rate $`\mathrm{\Gamma }_{\mathrm{p},j}`$ on $`D_\mathrm{s}`$, integrated over all possible $`D_\mathrm{l}`$, and on $`D_\mathrm{l}`$, for a given $`D_\mathrm{s}`$. Next we require the lens mass $`m`$ and relative transverse speed $`V_\mathrm{t}`$. The lens mass realization is generated from the distribution
$$P(m)m^{1/2}\psi (m),$$
(19)
since, in the absence of finite source-size effects, $`\mathrm{\Gamma }_\mathrm{p}R_\mathrm{e}\psi m^{1/2}\psi `$, where $`\psi `$ is the lens mass function (i.e. the number density of lenses per unit mass interval) and $`R_\mathrm{e}`$ is the Einstein radius. The transverse speed $`V_\mathrm{t}(𝑽_𝐥,𝑽_𝐬)`$ is drawn from the assumed velocity distributions $`P_\mathrm{l}(𝑽_𝐥)`$ and $`P_\mathrm{s}(𝑽_𝐬)`$ (see section 4), with $`𝑽_𝐥`$ and $`𝑽_𝐬`$ the lens and source three-dimensional velocity vectors. Since the microlensing rate $`\mathrm{\Gamma }_\mathrm{p}`$ is proportional to $`V_\mathrm{t}P_\mathrm{l}P_\mathrm{s}`$ rather than just $`P_\mathrm{l}P_\mathrm{s}`$, each of our realizations must be weighted by $`V_\mathrm{t}`$ in computing the final detection rate. Finally, we also need to generate the source absolute magnitude $`M`$ (defined for some photometric band). The dependency of $`\mathrm{\Gamma }_\mathrm{p}`$ on $`M`$ derives from the luminosity function $`\varphi `$ and the threshold impact parameter $`u_\mathrm{T}`$, so we have
$$P(M)u_\mathrm{T}(M,x,y)\varphi (M).$$
(20)
### 3.2 Generating light-curves
At this point we have only simulated events according to the underlying distributions which govern $`\mathrm{\Gamma }_\mathrm{p}`$; we have yet to take into account the distribution of observing epochs, variations in observing conditions, or candidate selection criteria.
The observing season runs from the beginning of August to the end of January, so we adopt the duration of an observing season to be $`\mathrm{\Delta }T=180`$ days. We assume 60 scheduled observing epochs per season — approximately the number of nights awarded for our 1999/2000 season. To construct a realistic sequence of observing epochs we assume that the WFC is mounted on the telescope and available for two-week periods every four weeks and that, on average, $`25\%`$ of scheduled observations are precluded by bad weather. Periods of poor weather are superposed on our initial observing sequence to obtain a final sequence which typically comprises 40–50 epochs per season. In practice we expect to obtain observations on more epochs than this, but for the purposes of these simulations we assume 40–50 as a conservative lower limit. For example during the 1999/2000 season we have had observations on 56 nights.
The epoch of maximum magnification $`t_0`$ and the minimum impact parameter $`u(t_0)`$ are both chosen at random. $`u(t_0)`$ is selected from the interval $`[0,u_\mathrm{T}]`$, where the threshold impact parameter $`u_\mathrm{T}`$ is computed from equations (5) and (7) taking $`A_{\mathrm{min}}=A_\mathrm{T}`$. This is all that is required to generate the underlying microlensing light-curve.
To compute the pixel light-curve, we must also model the galaxy surface brightness and sky background. The simulations presented here are performed in the $`V`$ band and we use the radially-averaged surface brightness profile in Table VI of Walterbos & Kennicutt (1987) to estimate the contribution to the pixel flux of the galaxy background at the event location. The assumed sky background corresponding to a dark sky is listed in Table 1, along with other INT detector and site characteristics. The sky background varies over lunar phase and we adopt a contribution to the sky background from the full moon equivalent to $`10^3`$ tenth magnitude stars per deg<sup>2</sup> (c.f. Krisciunas & Schaefer 1991). The contribution is modulated according to the lunar phase. The lunar contribution to the sky background also depends upon whether the moon is above the horizon and on its angular distance from M31. Our assumed value is taken to be an average over the positional dependence, so the true variation in the sky background will be somewhat larger than we consider. We also simplify the computation of the seeing fraction $`f_{\mathrm{see}}`$ by adopting a Gaussian PSF with a FWHM equal to the seeing of the reference image. The position of the PSF maximum for the reference image is selected at random within the central pixel of the superpixel array.
Using our computed values for $`f_{\mathrm{see}}`$, the INT detector and site characteristics summarized in Table 1, and the microlensing parameters generated for each event, we construct superpixel light-curves via equation (1). The error at each epoch $`i`$ is given by equation (2). Poisson realizations for the superpixel flux at each epoch are generated from $`N_{\mathrm{pix}}(t_i)`$ and $`\sigma _i`$.
### 3.3 Selection criteria and the observed rate
The adoption of selection criteria inevitably reduces the number of detected events, but they are necessary to minimize the number of contaminating non-microlensing signals. As in all microlensing experiments the selection criteria must be based upon the quality of the data and the characteristics of non-microlensing variations. Ultimately the criteria must be derived from the data themselves, so they are inevitably experiment-specific and evolve as the experiment progresses. For our simulations we impose criteria based loosely on the previous AGAPE pixel-lensing at Pic du Midi \[Ansari et al. 1997, Le Du 2000\].
The principal criterion for the selection of microlensing events in our simulation is that one and only one significant bump be identified on the light-curve. The bump must comprise at least three consecutive measurements lying at least $`3\sigma `$ above the baseline superpixel flux. Quantitatively, the significance of a bump is defined by its likelihood
$$L_{\mathrm{bump}}=\underset{i=j}{\overset{i=j+n,n3}{}}P(\mathrm{\Theta }>\mathrm{\Theta }_i|\mathrm{\Theta }_i3),$$
(21)
where $`\mathrm{\Theta }_i=[N_{\mathrm{pix}}(t_i)N_{\mathrm{bl}}]/\sigma _i`$ and $`P(\mathrm{\Theta })`$ is the probability of observing a deviation at least as large as $`\mathrm{\Theta }`$ by chance. For a Gaussian error distribution, $`P=\frac{1}{2}\text{erfc}(\mathrm{\Theta }/\sqrt{2})`$. Equation (21) indicates that we evaluate $`P(\mathrm{\Theta }_i)`$ only when $`\mathrm{\Theta }_i3`$. For our simulations we demand that a candidate have one bump with $`\mathrm{ln}L_{\mathrm{bump}}>100`$ and no other bump with $`\mathrm{ln}L_{\mathrm{bump}}>20`$. We further demand that the epoch of maximum magnification $`t_0`$ lies within an observing season; we reject candidates which attain their maximum brightness between seasons, even if they last long enough for the tails of the light-curve to be evident. This helps to ensure a reliable estimate of the peak flux, and in turn the FWHM timescale $`t_{\mathrm{FWHM}}`$.
The bump criterion is both a signal-to-noise ratio condition and a test for non-periodicity. It is crucial for distinguishing microlensing events from periodic variables, though long-period variables, such as Miras, may pass this test in the short term. In addition to the bump test, one can also test the goodness of fit of the light-curve to microlensing, which helps to distinguish microlensing from typical novae light-curves. Though the presence of the background means that pixel events will not in general be achromatic, the ratio of the flux increase to baseline flux in different colours should nonetheless be independent of time, so this provides another test for microlensing. Colour information may also help to exclude some long-period variables in the absence of a sufficient baseline of observations. In Section 6 we also exploit differences in spatial distribution to separate statistically lensing events from variable stars.
For real data-sets we would require more criteria in order to avoid excessive contamination from variable stars. For now we are simulating only microlensing events, so we are assured of no contamination in our selection. However, the cuts adopted above would be responsible for many of the rejected candidates in a real experiment, so the absence of further criteria should not lead to a gross overestimate of the rate. In any case, we have been deliberately conservative with our choices of sky background level, worst seeing scale, the number of epochs per season and the pixel stability level $`\sigma _\mathrm{T}`$. We therefore feel our predictions are more likely to be underestimates of the actual detection rate.
The observed rate can be now readily computed from $`\mathrm{\Gamma }_\mathrm{p}`$, the number of generated trials and the fraction of these which pass the detection criteria. As mentioned in Section 3.1, the way in which velocities are generated in the simulations means that the correct rate is obtained by weighting each event by its transverse speed $`V_\mathrm{t}`$. Thus, the observed rate for lens component $`j`$ is
$$\mathrm{\Gamma }_{\mathrm{p},j}^{\mathrm{obs}}=\mathrm{\Gamma }_{\mathrm{p},j}_{x,y}\frac{_{l=1}^{N_{\mathrm{det}}(j)}V_{\mathrm{t},l}}{_{k=1}^{N_{\mathrm{trial}}(j)}V_{\mathrm{t},k}},$$
(22)
where $`\mathrm{\Gamma }_{\mathrm{p},j}_{x,y}`$ is the spatial average of $`\mathrm{\Gamma }_{\mathrm{p},j}`$ (summed over source populations), the lower summation is over all $`N_{\mathrm{trial}}`$ trial events generated for lens component $`j`$ and the upper summation is over the $`N_{\mathrm{det}}`$ detected events which pass the selection criteria. The total number of events after $`n`$ observing seasons is
$$N=n\mathrm{\Delta }T\mathrm{\hspace{0.17em}10}^{0.4(MM_{\mathrm{gal}})}\underset{j}{}\mathrm{\Gamma }_{\mathrm{p},j}^{\mathrm{obs}},$$
(23)
where $`M`$ is the average absolute magnitude of the sources (integrated over the luminosity function) and $`M_{\mathrm{gal}}`$ is the absolute magnitude of M31 ($`M_V=21.2`$).
### 3.4 Simulated light-curves
Three light-curves generated for a first-season simulation involving $`0.1\text{M}_{\mathrm{}}`$ MACHOs are shown in Figure 2. The galactic models required for the simulation are discussed in Section 4. The light-curves illustrate the range in signal-to-noise ratio. The down-time for the WFC is evidenced by the way in which the epochs are clumped into two-week periods. The variation in the size of the error bars reflects the simulated variation in observing conditions.
Figure 2a shows an M31 halo lens magnifying a bulge star ($`M_V=0.4`$) and is a typical example. The underlying maximum magnification for this event is $`A_{\mathrm{max}}=18`$, whilst the maximum enhancement in superpixel flux is $`A_{\mathrm{pix}}(t_0)=1.06`$, indicating that the unlensed source is contributing less than $`0.4\%`$ of the superpixel flux. For this event $`t_{\mathrm{FWHM}}=5`$ days and $`t_\mathrm{e}=28`$ days. Figure 2b, which illustrates a poor candidate with a low signal-to-noise ratio, involves a Galaxy MACHO and $`M_V=1.8`$ bulge source contributing only $`0.1\%`$ of the superpixel flux ($`A_{\mathrm{max}}=42`$, $`A_{\mathrm{pix}}(t_0)=1.05`$). In this example $`t_{\mathrm{FWHM}}=5`$ days and $`t_\mathrm{e}=68`$ days. Though there appears to be evidence of a second bump after the main peak these points are all within $`3\sigma `$ of the baseline and so do not count as a bump. Figure 2c shows a high signal-to-noise ratio “gold-plated” event in which a very luminous ($`M_V=4`$) disc source is lensed by an M31 MACHO ($`A_{\mathrm{max}}=5`$, $`A_{\mathrm{pix}}(t_0)=2.1`$) with an observed duration $`t_{\mathrm{FWHM}}=19`$ days and underlying timescale $`t_\mathrm{e}=33`$ days. Here the bright unlensed source accounts for $`27\%`$ of the superpixel flux.
## 4 Lens and source models
In order to make quantitative estimates for pixel-lensing observables, we must specify models for the principal Galaxy and M31 lens and source components. For M31 the main populations are the bulge, the disc and the dark MACHO halo. For the Galaxy only the MACHO halo is important since the disc does not contribute significantly. Our complete model therefore consists of these four populations. Two populations, the M31 disc and bulge, also provide the sources, so in total we have eight different lens–source configurations. For each population we must specify distributions for the density and velocity. Additionally, we must specify the lens mass and a luminosity function for the source populations. Throughout we assume a disc inclination of $`77\mathrm{°}`$ and a distance to M31 of 770 kpc, consistent with recent determinations (e.g. Stanek & Garnavich 1998).
Whilst the present paper is concerned only with quantities relating to M31 and Galaxy MACHOs, we must nonetheless include other significant lens components in our modeling in order to properly characterize the complexity of extracting physical information from observations. For the observations, unlike the simulations, we do not know in which population a particular lens resides.
The haloes are modeled as simple near-isothermal spheres with cores, having density profiles
$$\rho _\mathrm{h}=\{\begin{array}{cc}\rho _\mathrm{h}(0)\frac{a^2}{a^2+r^2}\hfill & (rR_{\mathrm{max}})\hfill \\ 0\hfill & (r>R_{\mathrm{max}})\hfill \end{array},$$
(24)
where $`\rho _\mathrm{h}(0)`$ is the central density, $`a`$ is the core radius, $`R_{\mathrm{max}}`$ is the cutoff radius and $`r`$ is the radial distance measured from the centre of either M31 or the Galaxy. The assumed values for $`\rho _\mathrm{h}(0)`$, $`a`$ and $`R_{\mathrm{max}}`$ are given in Table 2. The halo fraction determinations in Section 6 are made with respect to these density normalizations. In our model the M31 halo has about twice the mass of the Galactic halo, though this mass ratio is controversial and has been challenged recently by Evans & Wilkinson (2000) who have studied the kinematics of several satellite galaxies around M31.
The M31 disc is modeled by the sech-square law:
$$\rho _\mathrm{d}=\rho _\mathrm{d}(0)\mathrm{exp}\left(\frac{\sigma }{h}\right)\mathrm{sech}^2\left(\frac{z}{H}\right),$$
(25)
where $`\sigma `$ is the radial distance measured in the disc plane and $`z`$ is the height above the plane. The normalization $`\rho _\mathrm{d}(0)`$, scale-height $`H`$ and scale-length $`h`$ are given in Table 2.
The bulge distribution is based on the work of Kent (1989). Kent models the bulge as a set of concentric oblate-spheroidal shells with axis ratios which vary as a function of semi-major axis. We use the tabulated spatial luminosity density values in Table 1 of Kent (1989) and normalize the bulge mass under the assumption that the light traces the mass (constant bulge mass-to-light ratio). The mass normalization $`M_\mathrm{b}`$ is listed in Table 2. The assumption of axisymmetry may be over-simplistic since the misalignment between the disc and bulge position angles probably implies a triaxial structure for the bulge. However, we are only indirectly concerned with bulge lensing in so much as it contaminates halo lensing statistics, so deviations from axisymmetry are not crucial.
The rotation curve and surface brightness profile for the adopted M31 components are shown in Figure 3. In constructing the surface brightness profile, we have assumed $`B`$-band mass-to-light ratios $`M/L_B=4`$ for the disc and $`M/L_B=9`$ for the bulge, consistent with that expected for typical disc and bulge populations. The overall surface brightness profile is shown by the solid line in Figure 3a, with the disc and bulge contributions indicated by the dashed and dot-dashed lines, respectively. The crosses are the radially averaged measurements from Table VI of Walterbos & Kennicutt (1987). In Figure 3b the solid, dashed and dot-dashed lines show the overall, disc and bulge contributions to the rotation curve, with the dotted line giving the halo contribution. The crosses are from Figure 2 of Kent (1989) and are based on the emission-line curves of Brinks & Shane (1984) and Roberts, Whitehurst & Cram (1978). The fit to both the surface brightness and rotation profiles is good, given the simplicity of the models.
The lens and source velocities are described by rotational and random components. The rotation velocity for each component is given in the 4th column of Table 2. The random motions are modeled by an isotropic Gaussian distribution with a one-dimensional velocity dispersion given by the 5th column. When calculating the relative transverse lens speed $`V_\mathrm{t}`$, we take account of both the motion of the source and the observer. The observer is assumed to move in a circular orbit about the centre of the Galaxy with a speed of 220 km s<sup>-1</sup>. We do not assume any relative transverse bulk motion between the Galaxy and M31. In practice, only the observer’s motion is of consequence for Galaxy lenses, and only the source motion for M31 lenses.
Since one of the questions we wish to address is how well pixel-lensing observables can characterize the MACHO mass, we shall simply model the Galaxy and M31 MACHO mass distributions by a Dirac $`\delta `$-function:
$$\psi (m_\mathrm{h})\frac{1}{m_\mathrm{h}}\delta (mm_\mathrm{h}),$$
(26)
The stellar lens mass distribution in the disc and bulge is described by a broken power law:
$$\psi (m_\mathrm{s})\{\begin{array}{cc}m_\mathrm{s}^{0.75}\hfill & (m_\mathrm{l}<m_\mathrm{s}<0.5\text{M}_{\mathrm{}})\hfill \\ m_\mathrm{s}^{2.2}\hfill & (0.5\text{M}_{\mathrm{}}<m_\mathrm{s}<m_\mathrm{u})\hfill \end{array}.$$
(27)
The mass function is normalized to yield the same value for $`\psi (0.5\text{M}_{\mathrm{}})`$ for either slope. We take a lower mass cut-off $`m_\mathrm{l}=0.08\text{M}_{\mathrm{}}`$ and an upper cut-off $`m_\mathrm{u}=10\text{M}_{\mathrm{}}`$, corresponding closely to the local Solar neighbourhood mass function \[Gould, Bahcall & Flynn 1997\]. Whilst this is a reasonable assumption for stars in the M31 disc, the mass function will overestimate the contribution of massive stars in the older bulge. The higher $`M/L_B`$ assumed for the bulge also requires that the disc and bulge mass functions be different. However, the slope at high masses is steep, so the contribution of high mass stars to the lensing rate is in any case small. Furthermore, as already mentioned, we are only interested in the bulge population as a contaminant of the halo lensing statistics. The choice of upper mass cut-off for the bulge is therefore not critical for the present study, so we simply adopt the same mass function for the disc and bulge.
The stellar components provide both lenses and sources. We assume that the lens and source populations are the same and so described by the same density, velocity and mass distributions. For the disc and bulge sources, we use the $`V`$-band luminosity function of Wielen, Jahreiss & Krüger (1983) for stars with $`M_V>5`$ and that of Bahcall & Soneira (1980) for $`M_V5`$. The two functions are normalized to the same value at $`M_V=5`$. A more detailed study of the M31 luminosity function is underway \[Lastennet et al. 2000\].
## 5 Predictions and trends for pixel lensing
The simulations for the POINT-AGAPE survey are performed over 1, 3 and 10 observing seasons for 9 MACHO masses spanning the range $`10^310\text{M}_{\mathrm{}}`$. Each simulation produces an estimate of the number of events across the whole M31 disc for each lens component, together with a library of typically $`10^4`$ candidates containing information such as the lens position, duration and transverse velocity. Since $`t_\mathrm{e}`$ cannot generally be measured from the light-curve, we output both $`t_\mathrm{e}`$ and $`t_{\mathrm{FWHM}}`$. The event libraries can be filtered to provide an estimate of the pixel-lensing rate for any field placement.
### 5.1 Number of events
Whilst the factor $`10^3`$ gain over LMC/SMC searches in the number of sources certainly boosts the rate of events, the fact that M31 pixel-lensing searches can typically detect only high-magnification events means that the gain in the rate is not of the same order. Nonetheless, as Figure 4 indicates, the expected pixel-lensing rate is almost an order of magnitude larger than for current LMC/SMC experiments for same lens mass and halo fraction. In the figure we have plotted the expected number of events for M31 MACHOs (solid line) and Galaxy MACHOs (dashed line) per season per deg<sup>2</sup>, assuming MACHOs comprise all the halo dark matter of both galaxies. The rates are averages over the whole M31 disc (rather than for a specific field placement) determined from simulations spanning ten seasons and 460 observing epochs. Within the first season the sensitivity to very massive MACHOs will be a little less than indicated in Figure 4.
The rate of events occurring within the two INT WFC fields for their first season (1999/2000) positions are displayed in Table 3. This excludes events occurring within 5 arcmin of the centre of M31 because this region is dominated by stellar self-lensing (see Section 5.3). Only a couple of self-lensing events per season are expected outside the exclusion zone. The Monte-Carlo error in the values in Table 3 is small, only about $`3\%`$, but one should expect a larger variation when comparing different seasons with different numbers of epochs (in addition to Poisson variations).
From Figure 4 and Table 3 we see that the sensitivity to MACHOs peaks at a mass around $`0.0030.01\text{M}_{\mathrm{}}`$, when around 140 MACHO events can be expected within the INT WFC fields for full haloes. Below $`10^3\text{M}_{\mathrm{}}`$ finite-source size effects become important, so the expected number of events will drop off rapidly. At the high mass end, even haloes comprising MACHOs as massive as $`10\text{M}_{\mathrm{}}`$ provide a rate of several events per season. The number of M31 MACHOs is about twice as large as the number of Galaxy MACHOs for the same mass and fractional contribution, which is a direct consequence of the mass ratio of the halo models we adopt.
### 5.2 Timescale distributions
In Figure 5 we plot the timescale distributions for the detected MACHOs for a range of masses in terms of $`t_{\mathrm{FWHM}}`$. The distributions for nine MACHO masses, spanning four orders of magnitude, are plotted. The masses are as listed in Table 3, with darker lines corresponding to more massive MACHOs. Since the timescale distributions for Galaxy and M31 MACHOs are practically indistinguishable for a given mass, in Figure 5 we have combined their timescale distributions, so the normalization of each curve is determined by the combined pixel-lensing rate shown in Figure 4 for each halo.
Whilst there is a clear trend of increasing $`t_{\mathrm{FWHM}}`$ with increasing MACHO mass, the correlation is much weaker than for $`t_\mathrm{e}`$. For example, a duration $`t_{\mathrm{FWHM}}=1020`$ days is typical of a $`0.1\text{M}_{\mathrm{}}`$ lens, but it is also not unusual for a lens as light as $`10^3\text{M}_{\mathrm{}}`$ or as heavy as $`10\text{M}_{\mathrm{}}`$. Figure 6 shows how the average duration $`t_{\mathrm{FWHM}}`$ varies with mass separately for M31 (solid line) and Galaxy (dashed line) MACHOs. Over four orders of magnitude in mass $`t_{\mathrm{FWHM}}`$ varies by about one order of magnitude, increasing from 4 days for $`10^3\text{M}_{\mathrm{}}`$ MACHOs to 35 days for $`10\text{M}_{\mathrm{}}`$ MACHOs (see also Table 4). For our sampling strategy we find empirically that $`t_{\mathrm{FWHM}}m_\mathrm{h}^{1/4}`$, whereas the average Einstein radius crossing timescale for the underlying population of microlensing events (with $`u1`$) scales as $`t_\mathrm{e}_{\mathrm{pop}}m_\mathrm{h}^{1/2}`$.
The mean ratio $`t_{\mathrm{FWHM}}/t_\mathrm{e}`$ is displayed in Figure 7 for detected events. It is clear that the ratio is not fixed but steadily decreases with MACHO mass. For low MACHO masses with short durations, sampling imposes a lower limit on $`t_{\mathrm{FWHM}}`$ and a loose lower limit on $`t_\mathrm{e}`$ as well. Whilst most events involving $`10^3\text{M}_{\mathrm{}}`$ lenses are too short to be detected, those that are either have an unusually long $`t_\mathrm{e}`$ or occur in regions of low surface brightness (which maximizes $`t_{\mathrm{FWHM}}`$ for a given magnification). Thus $`t_{\mathrm{FWHM}}/t_\mathrm{e}`$ is typically larger for the observed events. At the other end of the mass scale the converse is true. The total observation baseline imposes a maximum cutoff in $`t_{\mathrm{FWHM}}`$ and a loose upper limit in $`t_\mathrm{e}`$. Those events which are detected either have an unusually short $`t_\mathrm{e}`$ or else tend to occur in regions of high surface brightness where $`t_{\mathrm{FWHM}}`$ is minimized for a given magnification. So $`t_{\mathrm{FWHM}}/t_\mathrm{e}`$ tends to be smaller for observed events. From Table 4 we see that the average duration of detected events $`t_\mathrm{e}_{\mathrm{det}}`$ does not trace the population average $`t_\mathrm{e}_{\mathrm{pop}}`$. This is a consequence of sampling bias.
### 5.3 Spatial distributions
Since event timescales give only limited information in pixel lensing, the location of each event on the sky is a crucial observable. A robust measurement of near-far asymmetry in the event distribution would indicate the existence of an extended spheroidal population of lenses within which the visible M31 disc and bulge are embedded. Thus it would represent very firm evidence for the existence of MACHOs.
In Figure 8 we display the distribution of events across the face of the M31 disc after three observing seasons for the case where the haloes of both M31 and the Galaxy are full of $`0.3\text{M}_{\mathrm{}}`$ MACHOs. The axes are labeled in arcmins and are aligned along the major and minor axes of the disc light profile. The dashed-line templates indicate the positions of the two INT WFC fields for the 1999/2000 observing season.
In Figure 8a the positions of all detectable events are shown. MACHOs from the Galaxy halo are shown in green whilst M31 MACHOs are shown in blue. We find that within the central 5 arcmins (denoted by the circle) most events are produced by ordinary stellar lenses in the disc and bulge (shown in red). In Section 6, where we try to estimate MACHO parameters from simulated data-sets, we disregard events occurring within this region so as to minimize contamination from stellar lenses.
Figure 8b shows only the M31 MACHO distribution. The excess of events between $`y=10`$ and $`20`$ arcmins (along the minor axis towards the far side of the disc) compared to the number between $`y=+10`$ and $`+20`$ arcmins is a consequence of near-far asymmetry in the pixel-lensing rate. The strength of this asymmetry depends upon the number of M31 MACHOs which, in turn, depends upon their mass and density contribution, as well as the span of the observation baseline. The presence of Galaxy MACHOs makes the asymmetry harder to detect, so the ratio of M31 to Galaxy MACHOs is another factor which determines whether or not the asymmetry is measurable. It is evident from the figure that there are very few events at $`|y|25`$ arcmin. This is due to the decrease in both the number of sources and the signal-to-noise ratio (because the sky background provides a larger fraction of the total superpixel flux). The presence of the sky background effectively imposes a cut-off in the spatial distribution.
Figure 9 shows the spatial distribution for a range of MACHO masses expected after three seasons. We again assume that the MACHO mass is the same in both galaxies and that MACHOs provide all the dark matter in the two haloes. Figure 9a is for a MACHO mass of $`0.1\text{M}_{\mathrm{}}`$. In Figures 9b and 9c the MACHO mass is $`1\text{M}_{\mathrm{}}`$ and $`10\text{M}_{\mathrm{}}`$ respectively. The most obvious trend in the MACHO distributions is the decrease in the number of detectable events for models with more massive MACHOs. However, even for a mass as large as $`10\text{M}_{\mathrm{}}`$ we still expect to detect $`3040`$ MACHOs within the INT fields if they make up all the dark matter. After three seasons even these massive MACHOs out-number the disc and bulge lenses lying outside of our exclusion zone. This highlights one of the benefits of pixel lensing: the reduction in $`t_{\mathrm{FWHM}}`$ due to the presence of many neighbouring unresolved sources means that more events with relatively large $`t_\mathrm{e}`$ can be detected and characterized within a given observing period. In this respect, pixel lensing is relatively more sensitive to massive MACHOs than conventional microlensing experiments, which require resolved sources.
Another noticeable trend in Figure 9 is that more massive MACHOs are concentrated towards the central regions of the M31 disc. The main reason is that the MACHO and source surface densities are largest in this region, so the probability of an event occurring there is larger. However, another factor is that it is in the regions of highest surface brightness that the ratio $`t_{\mathrm{FWHM}}/t_\mathrm{e}`$ is minimized for a given magnification. For the $`10\text{M}_{\mathrm{}}`$ MACHO model, where many events may have a duration $`t_\mathrm{e}`$ exceeding the survey lifetime, this means more light-curves can be fully characterized, enabling these events to be flagged as microlensing candidates within the observing period. The converse is true for low-mass MACHOs with short $`t_\mathrm{e}`$. Their distribution is biased towards regions of lower surface brightness where $`t_{\mathrm{FWHM}}/t_\mathrm{e}`$ is maximized. This effect provides a further degree of discrimination for different lens masses and means that, for example, a halo with a modest contribution of low mass MACHOs may be distinguished from one with a substantial fraction of more massive lenses, even if the number of events for the two models is comparable. This in part makes up for the fact that $`t_{\mathrm{FWHM}}`$ is a less powerful discriminant than $`t_\mathrm{e}`$.
## 6 Estimating MACHO parameters
In the previous section we found that, whilst the timescale information in pixel-lensing studies is somewhat more restricted than in conventional microlensing we do, at least for M31, have important information from the spatial distribution of lenses. We now address to what extent pixel-lensing observables permit a reconstruction of the MACHO mass and halo fraction in the Galaxy and M31.
### 6.1 Maximum-likelihood estimation
Alcock et al. (1996) presented a Bayesian maximum likelihood technique to estimate the Galaxy MACHO mass and halo fraction from the observed event timescales towards the LMC. Evans & Kerins (2000) extended this to exploit the spatial distribution of observed events, and also to allow for more than one significant lens population. For pixel lensing towards M31 we must also consider the effect of contamination by variable stars. This is likely to be a significant problem in the short term. A baseline of more than three years should be sufficient to exclude periodic variables, such as Miras, but there still remains the possibility that, occasionally, the signal-to-noise ratio may be insufficient to distinguish between novae and microlensing events. By taking account of variable stars in our likelihood estimator we allow ourselves to make an estimate of the MACHO mass and lens fraction which, even in the short term, is robust and unbiased.
In order to allow for different MACHO parameters in the two galaxies we propose an estimator which is sensitive to five parameters: the MACHO mass and halo fraction in both the Galaxy and M31, and the degree of contamination by variable stars. We define our model likelihood $`L`$ by
$`\mathrm{ln}L(f_{\mathrm{var}},f_j,\psi _j)=`$ $``$ $`\left[f_{\mathrm{var}}N_{\mathrm{var}}+{\displaystyle \underset{j=1}{\overset{n_\mathrm{c}}{}}}f_jN(\psi _j)\right]`$ (28)
$`+`$ $`{\displaystyle \underset{i=1}{\overset{N_{\mathrm{obs}}}{}}}\mathrm{ln}[f_{\mathrm{var}}{\displaystyle \frac{d^3N_{\mathrm{var}}}{dt_{\mathrm{FWHM}}^{}{}_{i}{}^{}dx_idy_i}}`$
$`+`$ $`{\displaystyle \underset{j=1}{\overset{n_\mathrm{c}}{}}}f_j{\displaystyle \frac{d^3N(\psi _j)}{dt_{\mathrm{FWHM}}^{}{}_{i}{}^{}dx_idy_i}}],`$
where $`f_{\mathrm{var}}`$ is the fraction of variable stars relative to some fiducial model expectation number $`N_{\mathrm{var}}`$, $`f_j`$ and $`\psi _j`$ are the lens fraction and mass function for component $`j`$, $`n_\mathrm{c}`$ is the number of lens components and $`N_{\mathrm{obs}}`$ the number of observed events. For the disc and bulge components $`f_j`$ and $`\psi _j`$ are both fixed, with $`f_j=1`$ and $`\psi _j`$ given by equation (27), whilst for the Galaxy and M31 haloes $`\psi _jm_j^1\delta (mm_j)`$, as in equation (26), and $`f_j`$ and $`m_j`$ are free parameters. We define $`f_j`$ with respect to the halo density normalizations in Table 2.
The resolution of our simulation is insufficient to evaluate reliably the third derivatives in equation (28), so we decouple the timescale and spatial distributions by computing $`(dN/dt_{\mathrm{FWHM}})(d^2N/dxdy)`$ instead of $`d^3N/dt_{\mathrm{FWHM}}dxdy`$ within our fields. By averaging over spatial variations in the timescale distribution we are ignoring correlations which could provide us with further discriminatory information. However, in the limit of infinite data and perfect measurements we are still able to recover precisely the underlying parameters because the average event duration is known with infinite precision.
We assume that the distribution of variable stars traces the M31 surface brightness. In reality variable stars will be harder to detect in regions of higher surface brightness, so our idealized distribution is somewhat more concentrated than we should expect for a real experiment. We assume the timescale distribution of detectable variables is log-normal, with a mean and dispersion $`\mathrm{ln}t_{\mathrm{FWHM}}=2`$ and $`\sigma (\mathrm{ln}t_{\mathrm{FWHM}})=0.5`$ (where $`t_{\mathrm{FWHM}}`$ is expressed in days). Their timescales are therefore assumed to be typical of a wide range of lens masses (see Figure 5) and are thus least helpful as regards discrimination between lensing events and variable stars.
To test the likelihood estimator we generate data-set realizations and compute their likelihood over a five-dimensional grid of models spanning a range of MACHO masses and variable star and MACHO fractions. For the grid sampling we assume uniform priors in the variable star and MACHO fractions and logarithmic priors for the MACHO masses. Since the events in the inner 5 arcmin of the M31 disc are predominately due to stellar lenses (mostly bulge self-lensing) we count only events occurring outside of this region.
### 6.2 First-season expectations
Figure 10 shows the degree to which the MACHO parameters can be recovered after one season in the optimal case where the data-set contains no variable stars. For the realization we have adopted a MACHO fraction of 0.25 and mass of $`0.5\text{M}_{\mathrm{}}`$ for both the Galaxy and M31 haloes, and have set $`f_{\mathrm{var}}=0`$. The MACHO parameters correspond to those preferred by the most recent analyses of the EROS and MACHO teams \[Lasserre et al. 1999, Alcock et al. 2000\]. Each panel in Figure 10 represents a two-dimensional projection of the five-dimensional likelihood, in which each point on the two-dimensional plane is a summation of likelihoods over the remaining three dimensions. Contours are constructed about the two-dimensional maximum likelihood solution which enclose a given fraction of the total likelihood over the plane. The contours shown enclose $`34\%`$ (solid line), $`68\%`$ (dashed line), $`90\%`$ (dot–dashed line), $`95\%`$ (dotted line) and $`99\%`$ (triple dot–dashed line) of the total likelihood. The star in each plane shows the input values for the realization.
The four panels in Figure 10 depict the likelihood planes for M31 MACHO fraction and mass (top left), Galaxy MACHO fraction and mass (top right), M31 and Galaxy MACHO fractions (bottom left) and M31 MACHO and variable star fractions (bottom right). From the top-left panel we see that, after just one season, useful constraints are already possible for M31 parameters. In this realization the $`90\%`$ confidence level spans around two orders of magnitude in MACHO mass ($`0.0510\text{M}_{\mathrm{}}`$) and an order of magnitude in halo fraction ($`0.11.1`$). The brown-dwarf regime is mostly excluded. In the upper-right panel we see that the Galaxy MACHO parameters are ill-defined after one season. This is unsurprising since Galaxy MACHOs are out-numbered two to one by M31 MACHOs and they have no signature comparable to the near-far asymmetry of their M31 counterparts. The panel shows a suggestive spike in the likelihood contours occurring at about the right mass range, though the contours marginally prefer a Galaxy halo with no MACHO component. The one firm conclusion that can be drawn is that a substantial contribution of low-mass lenses is strongly disfavoured by the data. The strongest constraints occur at $`0.003\text{M}_{\mathrm{}}`$, where the expected number of events peaks for a given fractional contribution. The likelihood estimator indicates that $`0.003\text{M}_{\mathrm{}}`$ lenses contribute no more than $`5\%`$ of the Galactic dark matter with $`90\%`$ confidence. In the lower-left and lower-right panels of Figure 10 we see the trade-off between M31 and Galaxy MACHO fractions and between M31 MACHO and variable star fractions, respectively. The lower-left panel indicates that a scenario in which there are no MACHOs is excluded with very high confidence, despite the large uncertainty in the halo fraction determinations. In the lower-right panel we see that the likelihood estimator has correctly determined that there is little, if any, contamination due to variable stars, with a $`90\%`$ confidence upper limit of $`f_{\mathrm{var}}<0.03`$.
In Figure 11 we show the results for a simulation over one season in which there are no microlensing events, only variable stars. We adopt $`N_{\mathrm{var}}=100`$ and $`f_{\mathrm{var}}=1`$ within the INT WFC fields. It is important to establish whether, in the event of there being no MACHOs, our likelihood estimator is able to correctly determine a null result even if a significant number of variable stars pass the microlensing selection criteria. The four panels in Figure 11 indicate that our estimator has been very successful as regards the M31 MACHO contribution. The M31 MACHO fraction is constrained with $`90\%`$ confidence to be below 0.2 for lenses in the mass range $`0.0010.1\text{M}_{\mathrm{}}`$ and below $`0.4`$ for MACHOs up to a few Solar masses. This despite a rate in variable stars comparable to full haloes of MACHOs. In the upper-right panel we see that there is considerable uncertainty in the Galaxy MACHO parameters, though interesting upper limits on the halo fraction are obtained for lenses in the mass range $`0.030.1\text{M}_{\mathrm{}}`$. In the lower-left panel we see that a non-zero MACHO contribution is preferred though the contours are consistent with the input model at about the $`70\%`$ confidence level. In the lower-right panel we see that the estimator is able to constrain the number of variables to within $`\pm 30\%`$ of the input value. Thus our likelihood estimator has provided us with not just an estimate of the MACHO parameters but also an estimate of the level of contamination in the data-set. This estimate is completely independent of (and thus does not rely upon) additional information one might obtain from colour changes or asymmetry in the light-curves of individual events, or from follow-up observations.
### 6.3 Evolution of parameter estimation
Figure 12 shows another first-season simulation in which we adopt the same MACHO parameters as in Figure 10 but this time we also take $`N_{\mathrm{var}}=100`$ and $`f_{\mathrm{var}}=1`$. The contours in the plane of M31 MACHO mass and fraction appear largely unaffected by the presence of significant variable star contamination, and qualitatively resemble those in Figure 10. There is no evidence of estimator bias due to the presence of variables, which for our realization out-number the MACHOs from both haloes combined. However the Galaxy MACHO parameter estimation is clearly led astray by the presence of variables, with upper limits on halo fraction possible for only a narrow range of lens masses. The estimator nonetheless strongly excludes a no-MACHO hypothesis (lower-left panel) and provides a good estimate of variable star contamination levels.
Figure 13 shows the constraints after three seasons assuming the same parameters as for Figure 12, except that we have reduced the contamination level to $`f_{\mathrm{var}}=0.3`$. A significant decrease in contamination would be expected as the increase in observation baseline permits the exclusion of a larger number of periodic variables. The constraints for M31 MACHO parameters have tightened up considerably, with a $`90\%`$ confidence uncertainty of a factor four in halo fraction and an order of magnitude in MACHO mass. The constraints on Galaxy MACHO parameters have also sharpened considerably, allowing strong upper limits on the halo fraction to be made over a wide mass range, though the data in this case is consistent with a complete absence of Galaxy MACHOs. However, in the lower-left panel we see that the joint constraint on M31 and Galaxy MACHO fraction advocates a significant overall MACHO contribution. The lower-right panel also shows an accurate determination of contamination levels.
In Figure 14 we depict constraints for ten seasons of data, comparable to the lifetime of current LMC surveys, with the variable star contamination level reduced further to $`f_{\mathrm{var}}=0.1`$. The M31 MACHO fraction is now essentially specified to within about a factor of three, whilst the MACHO mass uncertainty is within an order of magnitude. We now also have a positive estimation of the Galaxy MACHO contribution and mass. The constraints on Galaxy parameters are only a little worse than those for M31 after three seasons. The variable star contamination level is once again robustly determined.
Figures 12 to 14 show that the likelihood estimator is able to distinguish clearly between microlensing events and our naive model for the variable star population. They also show that, given a lifetime comparable to the current LMC surveys, a sustained campaign on the INT should determine M31 MACHO parameters rather precisely and should also provide a useful estimate of Galaxy MACHO parameters. A more modest campaign lasting three seasons would provide a robust estimate of M31 MACHO parameters and useful constraints on the Galaxy MACHO fraction.
Since all the above simulations assume the same halo fraction and MACHO mass for both galaxies, we decided to test whether our likelihood estimator was sensitive to Galaxy MACHO parameters independently of M31 MACHO values. We therefore ran a simulation over three seasons in which $`30\%`$ of the M31 halo comprises $`0.5\text{M}_{\mathrm{}}`$ lenses and $`60\%`$ of the Galaxy halo comprises $`0.03\text{M}_{\mathrm{}}`$ lenses. The Galaxy MACHOs actually out-number the M31 MACHOs in this model. Whilst the model is somewhat contrived, and is already ruled out with high confidence \[Lasserre et al. 1999, Alcock et al. 2000\], it provides a useful test case for our estimator. We find that the estimator successfully resolves the mass scales of the two populations within $`90\%`$ confidence, though with a slight tendency to overestimate the Galaxy MACHO mass and underestimate the M31 MACHO mass. Whilst we find a large overlap in preferred halo fraction, this is consistent with the sensitivity typically achieved after three seasons when the MACHO masses in the two galaxies are the same. The Galaxy MACHO parameters are much better defined than in Figure 13, though for this case the variable star contamination level was set to zero.
There is one aspect, however, in which our simulation presents an over-optimistic picture. The success of the estimator in discriminating between variable stars and microlensing events is mostly due to the fact that our adopted variable star distribution is significantly more concentrated than the microlensing distribution of either M31 or Galaxy MACHOs. The assumption we have made, that their observed distribution traces the M31 surface brightness, is reasonable only for very bright variable phenomena which would be detected regardless of where it occurred in M31. For less prominent variables there will be a bias against their detection in the central regions of M31, where the surface brightness is high and so their contribution to the superpixel flux relatively small. We might well expect a realistic distribution of variable stars to resemble that of Galaxy MACHOs because the surface density of Galaxy MACHOs does not vary significantly over the M31 disc, so their distribution would also trace the M31 surface brightness if there was no detection bias away from regions of high surface brightness. However, in the absence of a conspiracy between the flux distribution of variable stars at peak luminosity and the flux distribution of microlensed sources at peak magnification, there should be some distinction between the spatial distributions of Galaxy MACHOs and M31 variable stars, though this may be only mild. In any case, even if the two distributions are indistinguishable this should not significantly affect the determination of M31 MACHO parameters because the likelihood relies heavily on evidence of near-far asymmetry (which is why the likelihood contours are much better defined for M31 MACHOs than for Galaxy MACHOs). This cannot be replicated by variable stars. Only if several hundred variable stars passed the selection criteria every season would the signature of asymmetry be washed out and the constraints on M31 MACHO parameters severely degraded. Such an occurrence would warrant critical re-examination of the selection criteria!
## 7 Conclusions
Pixel lensing is a relatively new and powerful method to allow microlensing searches to be extended to targets where the sources are unresolved. It heralds the possibility of detecting or constraining MACHO populations in external galaxies. Though pixel lensing is hampered by changes in observing conditions, which introduce spurious variations in detected pixel flux, techniques have been developed which minimize these variations to a level where genuine microlensing signals can be detected.
POINT-AGAPE and another team (MEGA) have embarked on a major joint observing programme using the Isaac Newton Telescope (INT) to monitor unresolved stars in M31 for evidence of pixel lensing due to MACHOs either in the Galaxy or M31 itself. Two techniques, the Pixel Method and difference imaging, are available to minimize flux variations induced by the changing observing conditions. In this paper we have assessed the extent to which the Pixel Method allows us to determine the mass and fractional contribution of MACHOs in both M31 and the Galaxy from pixel-lensing observables. Our assessment takes account of realistic variations in observing conditions, due to changes in seeing and sky background, together with irregular sampling.
Pixel lensing observables differ from those in classical microlensing, where one targets resolved stellar fields, in that one is generally unable to measure the Einstein radius crossing time, $`t_\mathrm{e}`$, of an event. The fact that the source stars are resolved only whilst they are lensed means that one is unable to determine their baseline luminosity, so neither the magnification nor the total duration of the event can be measured. As an alternative to $`t_\mathrm{e}`$ one may measure the full-width half-maximum timescale, $`t_{\mathrm{FWHM}}`$, directly from the light-curve. However, this provides only a lower limit to the underlying event duration. Fortunately, M31 provides a signature which permits an unambiguous determination of whether or not MACHOs reside in its halo: near-far asymmetry \[Crotts 1992\]. If M31 is embedded in a dark spheroidal halo of MACHOs the high disc inclination should provide a measurable gradient in the observed pixel lensing rate. The strength of this signature depends both on the mass and fractional contribution of MACHOs in M31, as well as the level of “contamination” by variable stars, M31 stellar lensing events and foreground Galaxy MACHOs.
We have employed detailed Monte-Carlo simulations to estimate the timescale and spatial distributions of MACHOs in both our Galaxy and M31 for spherically-symmetric near-isothermal halo models. We also model the lensing contribution due to disc and bulge self-lensing. The expected number of M31 MACHOs for our two INT fields peaks at about 100 events for $`0.01\text{M}_{\mathrm{}}`$ MACHOs, the Galaxy MACHO contribution being about half as large. For a given mass and halo fraction we expect to detect about an order of magnitude more events than current conventional surveys targeting the LMC.
The timescale distributions for Galaxy and M31 MACHOs are practically identical because of the symmetry of the microlensing geometry. Our simulations also confirm that $`t_{\mathrm{FWHM}}`$ is less strongly correlated with lens mass than $`t_\mathrm{e}`$. For our sampling we find that, empirically, $`t_{\mathrm{FWHM}}t_\mathrm{e}^{1/2}m^{1/4}`$ for lens mass $`m`$. Sampling introduces a significant bias in the duration of detected events with respect to the underlying average for very massive and very light MACHOs.
Our simulations clearly show the near-far asymmetry in the M31 MACHO spatial distribution. However, the presence of the foreground Galaxy MACHOs makes its measurement more difficult. We also find that the distribution of very massive MACHOs is noticeably more centrally concentrated than that of less massive lenses. Stellar self-lensing events are found to be mostly confined to within the inner 5 arcmin of the M31 disc, and are mostly due to bulge self-lensing. Their tight spatial concentration means that they do not pose a serious contamination problem for analysis of the Galaxy and M31 MACHO populations.
We have constructed a maximum likelihood estimator which uses timescale and position observables to simultaneously constrain the MACHO mass and halo fraction of both M31 and the Galaxy. The statistic is devised to be robust to data-set contamination by variable stars. We find that M31 MACHO parameters can be reliably constrained by pixel lensing. For simulated INT data-sets we find pixel-lensing constraints on the M31 halo to be comparable to those obtained for the Galaxy halo by the conventional microlensing surveys. Even with severe contamination from variable stars the M31 MACHO parameters are well determined within three years. In particular, if there are few MACHOs in M31 this should become apparent after just one season of data collection, even if as many as a hundred variable stars pass the microlensing selection criteria, because of the absence of near-far asymmetry. Pixel lensing is less sensitive to Galaxy MACHO parameters. Our simulations indicate that we require at least three times as much observing time in order to produce comparable constraints on Galaxy MACHO parameters. If the spatial distribution of variable stars closely follows that of Galaxy MACHOs, then it may become very difficult to reliably constrain Galaxy MACHO parameters.
The work presented here clearly demonstrates that a vigorous monitoring campaign on a 2m class telescope with a wide-field camera can identify and characterize MACHOs in M31. We now have the opportunity to unambiguously establish the existence or absence of MACHOs in an external galaxy. The advantage of targeting M31 over our own Galaxy is that we have many lines of sight through the halo of M31 and a clear signature with which to distinguish M31 MACHOs from stellar self-lensing, the primary source of systematic uncertainty for Galaxy halo microlensing surveys. M31 therefore represents one of the most promising lines of sight for MACHO studies.
## acknowledgments
EK, EL and SJS are supported by PPARC postdoctoral fellowships. NWE is supported by the Royal Society. EK would like to thank Yannick Giraud-Héraud and Jean Kaplan for many helpful discussions.
|
warning/0002/astro-ph0002071.html
|
ar5iv
|
text
|
# Alternative Solutions to Big Bang Nucleosynthesis
## 1. Introduction
The success of standard big bang nucleosynthesis (SBBN) in predicting the observed abundances of the light elements has led to the widespread view that SBBN must be correct. According to this view, any remaining disagreements must be due to systematic errors in observations or incorrect, or too crude, chemical evolution models. While this view may well be the right one, we should not be blind to other possibilities.
I will stay within the context of the Hot Big Bang (for an alternative, see Burbidge & Hoyle 1998), and discuss some models of nonstandard big bang nucleosynthesis (NSBBN). NSBBN scenarios range from small modifications to SBBN to a complete change in the decisive physical phenomena, like in the late-decaying massive particle scheme of Dimopoulos et al. (1988).
Motivations for studying NSBBN go in two directions. First, the remarkable success of SBBN allows one to severely constrain the physics of the early universe. If one tries to change the conditions from the standard assumptions the resulting abundances of the light elements differ from the observed ones. For many modifications, BBN provides the strongest constraints. BBN gives also the strongest constraint on the single parameter of SBBN, the baryon density, usually given as the baryon-to-photon ratio,
$$\eta \frac{n_b}{n_\gamma },\eta _{10}10^{10}\eta .$$
(1)
Second, one may try to improve on SBBN. From time to time it has seemed that there might be some discrepancy between observations and SBBN, which could then be explained by NSBBN. In particular, there has been tension between D/H and $`Y_p`$ (see, e.g., Hata et al. 1995). To relieve this tension, either a lower D or a lower <sup>4</sup>He yield has been looked for. Also one may want to relax the SBBN bounds to $`\eta `$. Other astronomical considerations have given motivation for trying to raise the upper limit to $`\eta `$. If one believes that the energy density of the universe is dominated by vacuum energy (the cosmological constant) and accepts the newer observations on D/H and $`Y_p`$ favoring somewhat larger $`\eta `$ within SBBN, this motivation largely disappears.
There is a very large body of work on NSBBN. Extensive reviews are given by Malaney & Mathews (1993) and Sarkar (1996), which contain, respectively, over 500 and over 700 references. Here I will be able to mention only a random few.
Most of the work on NSBBN can be divided into four broad classes:
1. Inhomogeneous BBN. Usually this means inhomogeneity in the baryon-to-photon ratio, $`\eta `$, but there are also other possibilities, like inhomogeneity in the neutrino chemical potentials.
2. Nonstandard neutrino physics, e.g., additional (“sterile”) neutrino flavors, neutrino degeneracy (asymmetry), massive $`\nu _\tau `$, or neutrino oscillations.
3. Late-decaying ($`\tau `$ = 1–$`10^8`$ s) massive particles, black holes, cosmic strings, etc.
4. Time-varying fundamental constants.
In the interest of time and space, I will discuss the first two classes only.
## 2. Inhomogeneous Big Bang Nucleosynthesis
The single parameter of SBBN is the baryon-to-photon ratio $`\eta `$, or the density of baryonic matter. In inhomogeneous big-bang nucleosynthesis (IBBN) one assumes that $`\eta `$ is inhomogeneous. To get a significant effect on BBN this inhomogeneity has to be large, $`\delta \eta /\eta >1`$. Since the baryons make an insignificant contribution to the energy density at nucleosynthesis time, the total energy density may still be essentially homogeneous. The inhomogeneity could be caused by, e.g., first-order phase transitions. The distance scale of this inhomogeneity is of crucial importance for IBBN. Without inflation, causal physics can only produce significant inhomogeneity at subhorizon scales (see Table 1).
Mechanisms connected with inflation can produce inhomogeneity at any scale. The isotropy of the cosmic microwave background (CMB) rules out significant inhomogeneity at $`>10`$ Mpc scales, and it is difficult to construct an acceptable IBBN scenario which would explain inhomogeneity in observations. In the usual IBBN models one considers a significantly smaller distance scale, so that while $`\eta `$ is inhomogeneous during BBN, resulting in inhomogeneous abundances at first, everything gets mixed and becomes chemically homogeneous before or during galaxy formation. Thus the observable primordial abundances are homogeneous, while different from the SBBN predictions.
The simplest version of IBBN is one where SBBN occurs with different $`\eta `$ in different parts of the universe, and the yields get mixed afterwards, so that one obtains the IBBN results by averaging SBBN results over the $`\eta `$ distribution, whose average we denote by $`\overline{\eta }`$. This kind of IBBN has a long history. Typically $`Y_p`$ goes up, <sup>7</sup>Li goes up (down for small $`\overline{\eta }`$), and D goes up for large $`\overline{\eta }`$, and down for small $`\overline{\eta }`$, compared to SBBN with $`\eta =\overline{\eta }`$. Leonard & Scherrer (1996) concluded that this way one can reduce the lower bound to $`\eta `$ from observations (in fact remove it, if arbitrary $`\eta `$ distributions are allowed), but the upper bound is essentially unchanged from SBBN, as <sup>7</sup>Li and <sup>4</sup>He are overproduced for larger $`\overline{\eta }`$. The tension between D an <sup>4</sup>He is worsened at the large end of the SBBN acceptable range. Thus this kind of modification to BBN appears undesirable.
### 2.1. Small Scale Inhomogeneity and Neutron Diffusion
The above applies to inhomogeneity with distance scales significantly larger than the neutron diffusion scale ($`0.1`$ pc). If there is inhomogeneity at smaller scales, neutrons will diffuse out of the high density regions resulting in an inhomogeneous $`n/p`$ ratio. Especially if this results in $`n/p>1`$ in some regions, the consequencies for BBN may be dramatic. This scenario (Applegate, Hogan, & Scherrer 1987) looked very exciting about ten years ago when it was noted that the QCD (quark-hadron) transition seemed likely to produce strong inhomogeneity at just the right distance scale, and early IBBN calculations indicated a large reduction in $`Y_p`$ and increase in D/H allowing very large $`\eta `$, even a critical density in baryons only. More detailed calculations showed that the effects were less dramatic, and the upper limit to $`\eta `$ given by D/H and $`Y_p`$ is raised at most by a factor of 2 or 3 as compared to SBBN, and this only if the inhomogeneity was at near the optimal distance scale ($`10^3\mathrm{}10^2`$ pc), and most of the baryon number was in the high density regions. The most severe problem for this kind of IBBN is <sup>7</sup>Li overproduction. Some <sup>7</sup>Li depletion (by a factor of 2 or 3) in Pop II stars is needed to allow for larger $`\eta `$ than in SBBN. Figure 1 is from a recent review of this scenario by Kainulainen, Kurki-Suonio, & Sihvola (1999).
Recent lattice QCD calculations favor a much smaller distance scale, although uncertainties are big enough so that the optimal distance scale cannot be ruled out. The distance scale from the electroweak (EW) phase transition must be so small that the effects on BBN cannot be large; in the best case they could be comparable to other small effects that have recently been included in accurate BBN codes.
### 2.2. Regions of Antimatter
A less-studied variant of IBBN is one where $`\eta `$ is allowed to have negative values, i.e., there are antimatter regions. This is possible in some baryogenesis scenarios (Dolgov 1996). Antimatter in cosmology has been reviewed by Steigman (1976). If the distance scale of antimatter regions is small, antimatter and matter will mix and annihilate in the early universe, and the presence of matter today implies that there was initially more matter than antimatter. If the distance scale is large, so that antimatter regions will survive till present, observational constraints require either the amount of antimatter to be very small, or the distance scale to be very large, comparable to the present horizon or larger (Cohen, De Rújula, & Glashow 1998), so that the case of large regions is not of interest for BBN.
The smaller the antimatter regions are, the earlier they annihilate. Rehm & Jedamzik (1998) considered annihilation immediately before nucleosynthesis. Kurki-Suonio & Sihvola (1999) extended these results to larger distance scales where annihilation occurs during or after nucleosynthesis (see Figure 2). So far the focus has been on obtaining upper limits to the amount of antimatter at various scales in the early universe, but clearly there is also potential for obtaining acceptable abundances with nonstandard values of $`\eta `$, although probably only with fine-tuned model parameters.
## 3. Neutrinos and Big Bang Nucleosynthesis
Neutrinos affect BBN in two ways, through the energy density effect and the $`\nu _e`$ effect. The most significant effect is on $`Y_p`$ in both cases.
The energy density in neutrinos affects the expansion rate of the universe. The simplest way to increase the energy density of the early universe from the standard model is to have additional particle species (sterile neutrinos or other hypothetical particles). The custom is to parametrize this by an “effective number of neutrino species”. The standard case is $`N_\nu =3`$. We now know that there are only three “active” neutrino species, so any additional species must be “sterile” neutrinos or other very weakly interacting particles. A higher energy density means faster expansion. This leads to $`n/p`$ freezeout at a higher temperature, leaving more neutrons, and resulting in a higher <sup>4</sup>He yield. The D yield is also increased, so an increased energy density is disfavored by BBN, and one gets an upper limit, e.g., $`N_\nu <3.2`$ (Burles et al. 1999) or $`N_\nu <4`$ (Lisi, Sarkar, & Villante 1999), depending on what observational constraints one uses.
Electron neutrinos affect the weak $`np`$ reactions directly. More $`\nu _e`$ leads to fewer neutrons and thus to less <sup>4</sup>He (and everything else), whereas more $`\overline{\nu }_e`$ leads to more neutrons and more <sup>4</sup>He.
### 3.1. Neutrino Degeneracy
In SBBN one assumes that the neutrino asymmetry (difference between the number of neutrinos and antineutrinos),
$$L_\nu \frac{n_\nu n_{\overline{\nu }}}{n_\gamma }=0.069\left(\frac{T_\nu }{T}\right)^3(\pi ^2\xi +\xi ^3),$$
(2)
which is related to the neutrino chemical potential $`\mu _\nu `$, or the degeneracy parameter $`\xi \mu _\nu /T`$, is small, $`1`$. This seems natural, since the comparable baryon asymmetry $`\eta `$ is small. However, the neutrino background is unobservable, so we cannot rule out a large neutrino asymmetry. A larger asymmetry always means a larger neutrino energy density, raising $`N_\nu `$. To have a significant effect on BBN, we must have $`|\xi |`$, $`|L_\nu |>0.1`$. There is a separate contribution from each neutrino flavor. Thus there are three indepedent degeneracy parameters, $`\xi _e`$, $`\xi _\mu `$, and $`\xi _\tau `$. The energy density effect is the same for all three flavors, and depends only on $`|\xi |`$. The electron neutrino effect depends only on $`\xi _e`$, but is much stronger, and the direction of the effect depends on the sign.
There are two possible scenarios for affecting BBN. If $`\xi _e`$ is comparable in magnitude to $`\xi _\mu `$ and $`\xi _\tau `$, or larger, one can forget the other two in first approximation. One can then adjust $`\xi _e`$ to dial in the desired value of $`Y_p`$. The other elements are hardly affected. A less natural scenario is one where the asymmetries in the other two neutrino flavors are much larger, and the energy density and $`\nu _e`$ effects are balanced against each other to keep $`Y_p`$ in the acceptable range. This way one can have a significant effect on the other abundances and raise the acceptable range for $`\eta `$. This second scenario is constrained by structure formation, since the large neutrino energy density means that the matter/radiation equality and thus the beginning of structure formation occurs later. Kang & Steigman (1992) used a generous lower limit for matter/radiation equality, $`z_{\mathrm{eq}}>10^3`$ to widen the SBBN acceptable range from $`\eta _{10}`$ = 2.8–4.7 to $`\eta _{10}`$ = 2.8–19.
### 3.2. Inhomogeneous Neutrino Degeneracy
The different results from high-$`z`$ D/H measurements (Tytler, Fan, & Burles 1996; Webb et al. 1997) raised the question whether there might be a large-scale inhomogeneity in primordial abundances. This is very difficult to achieve, since the extreme isotropy of the CMB rules out any significant large-scale inhomogeneity in $`\eta `$ or the energy density. Dolgov & Pagel (1999) have come up with a way of getting around this constraint. In their model the asymmetries of the different neutrino flavors are inhomogeneous but balanced with each other so that they add up to a homogeneous total energy density. The inhomogeneous $`\xi _e`$ is then responsible for the inhomogeneous primordial abundances through the $`\nu _e`$ effect. They suggest that an Afflect-Dine type scenario of generation of leptonic charge asymmetry, respecting the symmetry between different lepton families, could be responsible for creating a domain structure, where the neutrino asymmetries would have the same three values but interchanged with respect to $`e`$, $`\mu `$ and $`\tau `$. To achieve a significant D/H inhomogeneity, a huge $`Y_p`$ inhomogeneity has to be allowed. But since there are no high-$`z`$ $`Y_p`$ determinations, this cannot be used to rule out their model. Table 2 shows an example of what kind of abundances we could have in such a domain structure. The first line would correspond to our local domain; from the other domains we would have only D/H observations.
### 3.3. Decay of a Massive Tau Neutrino
If the rest mass of a neutrino species is much larger than 100 MeV, then it is becoming nonrelativistic before nucleosynthesis and its contribution to the energy density is different from the standard zero-mass case. The laboratory limits for the neutrino masses leave this as a possibility for $`\nu _\tau `$. Above the neutrino decoupling temperature, $`T3`$ MeV, a massive neutrino species contributes less energy density, because of neutrino-antineutrino annihilation, but after neutrino decoupling the annihilation ceases and the rest mass then contributes extra energy density. Neutrinos this heavy must decay to avoid contributing too much to the present energy density. The decay time and mode are of crucial importance to BBN. If the decay time is very short, then the contribution to $`N_\nu `$ will be less than one. The most interesting case is the one where $`\nu _\tau `$ decays into $`\nu _e`$ (and a scalar particle), since then the $`\nu _e`$ effect could cause a significant reduction in $`Y_p`$.
These calculations are difficult since the decisive effects occur near the neutrino decoupling temperature, so thermal equilibrium is not maintained and the neutrino spectra are distorted. The recent results by Hannestad (1998) and Dolgov et al. (1999) are in disagreement with each other. Hannestad gets the maximum reduction of $`Y_p`$, from the SBBN result $`Y_p=0.239`$ to $`Y_p<0.20`$, for $`\nu _\tau `$ mass $`m_\nu `$ = 0.2–0.5 MeV and lifetime $`\tau <100`$ s. According to Dolgov et al., the maximum reduction is less, to $`Y_p0.21`$, and occurs for larger masses, $`m_\nu `$ = 2–3 MeV, and requires a shorter lifetime $`\tau <1`$ s.
The most natural explanation of the SuperKamiokande (1998) result on atmospheric neutrinos is $`\nu _\mu \nu _\tau `$ oscillation. Then $`\nu _\tau `$ cannot be heavy and its mass will not affect BBN significantly. To allow the above scenario, the atmospheric neutrino oscillations would have to be into a sterile neutrino species, $`\nu _\mu \nu _s`$, instead (Kainulainen et al. 1999).
### 3.4. Neutrino Oscillations
Observations of solar neutrinos and atmospheric neutrinos (SuperKamiokande 1998) as well as the LSND (1998) accelerator experiment see different amounts of the different neutrino flavors than predicted by the Standard Model. This can be explained by neutrino oscillations. This is a quantum-mechanical phenomenon where the flavor $`(\nu _e,\nu _\mu ,\nu _\tau )`$ content of the neutrino varies periodically. This requires nonzero neutrino masses and the effect is determined by the difference in mass-squared, $`\mathrm{\Delta }m^2`$, and the “mixing angle”.
All three (solar, atmospheric, and LSND) “neutrino problems” cannot be simultaneously explained by oscillations among three flavors, but require at least a fourth flavor, $`\nu _s`$, which must be “sterile”, i.e., much more weakly interacting than the three known “active” flavors, in order not to violate the limit $`N_\nu 3`$ from $`Z^0`$ decay width (Particle Data Group 1998). A sterile neutrino would also be useful for supernova nucleosynthesis (Peltoniemi 1996; Caldwell, Fuller, & Qian 1999).
The LSND results are controversial, so the other viewpoint is to ignore them until they are confirmed by independent experiments, in which case the solar and atmospheric neutrino problems can be explained just with the three active neutrinos.
Oscillations among (light, non-degenerate, i.e., $`\xi =0`$) active neutrinos do not affect BBN, since they all have equal abundances. If the sterile neutrino exists, it would have thermally decoupled from the other neutrinos very early, much before BBN, so that its contribution to $`N_\nu `$ would be $`1`$. Active-sterile neutrino oscillations before BBN would then lead to production of $`\nu _s`$, increasing $`N_\nu `$ (Enqvist, Kainulainen, & Thomson 1992), which from the BBN point of view is undesirable. The situation is more complicated, however. The oscillation depends on the background temperature, and at a certain temperature there is a resonance. This resonance temperature depends on the neutrino energy, so as the temperature falls, the resonance sweeps through the neutrino spectrum. If there is a small pre-existing asymmetry (this will be the case, since thermal fluctuations suffice), the rates of neutrino and antineutrino oscillation will be different. Resonant active–sterile neutrino oscillations will then lead to a growth of the neutrino asymmetry by a large factor (Barbieri & Dolgov 1991; Foot & Volkas 1995; Shi 1996; Enqvist, Kainulainen, & Sorri 1999; Di Bari & Foot 2000). This may generate a large enough electron neutrino asymmetry to affect BBN (Bell, Foot, & Volkas 1998; Kirilova & Chizhov 1998; Shi, Fuller, & Abazajian 1999).
Depending on the oscillation parameters, the asymmetry may either just grow or oscillate between positive and negative values, so that the final sign of the asymmetry becomes unpredictable. To calculate the effect on BBN is complicated, since the resulting distortion of the $`\nu _e`$ spectrum is also important for BBN, and the process happens near the neutrino decoupling temperature. There are two schemes to generate a large $`\nu _e`$ asymmetry, either directly via $`\nu _e\nu _s`$ oscillations or indirectly via $`\nu _{\mu (\tau )}\nu _s`$ and $`\nu _{\mu (\tau )}\nu _e`$ oscillations.
This scenario is under active study and there is much controversy among the different research groups. In Fig. 3 we show results obtained by Shi et al. (1999). The maximal effect on $`Y_p`$ seems to be at the $`\pm 0.01`$ level.
## 4. Conclusions
At present, no NSBBN scenario appears as convincing as SBBN, which is the simplest of all. Often the real world has turned out to be more complicated in the end than first assumed, but for the early universe a simple picture has been very successful. However, it is healthy to keep in mind the possibility that SBBN might not be the full story, and that any discrepancies between observations and SBBN might actually be telling us something important about the early universe or particle physics.
#### Acknowledgments.
I thank K. Kainulainen, A. Kalliomäki, J. Peltoniemi, and A. Sorri for advice on neutrino physics.
## References
Applegate, J. H., Hogan, C. J., & Scherrer, R. J. 1987, Phys.Rev.D 35, 1151
Barbieri, R. & Dolgov, A. 1991, Nucl.Phys.B 237, 742
Bell, N. F., Foot, R., & Volkas, R. R. 1998, Phys.Rev.D 58, 105010
Burbidge, G., & Hoyle, F. 1998, ApJ 509, L1
Burles, S., Nollett, K. M., Truran, J. W., & Turner, M. S. 1999, Phys.Rev.Lett 82, 4176
Caldwell, D. O., Fuller, G. M., & Qian, Y.-Z. 1999, astro-ph/9910175
Cohen, A. G., De Rújula, A., & Glashow, S. L. 1998, ApJ 495, 539
Di Bari, P. & Foot, R. 2000, Phys.Rev.D, to be published, hep-ph/9912215
Dimopoulos, S., Esmailzadeh, R., Hall, L. J., & Starkman, G. D. 1988, ApJ 330, 545
Dolgov, A. D. 1996, in Proceedings of the International Workshop on Baryon Instability, Oak Ridge, March 1996, hep-ph/9605280
Dolgov, A. D., Hansen, S. H., Pastor, S., & Semikoz, D. V. 1999, Nucl.Phys.B 548, 385
Dolgov, A. D., & Pagel, B. E. J. 1999, New Astron 4, 223
Enqvist, K., Kainulainen, K., & Thomson, M. 1992, Nucl.Phys.B 373, 498
Enqvist, K., Kainulainen, K., & Sorri. A. 1999, Phys.Lett.B 464, 199
Foot, R. & Volkas, R. R. 1995, Phys.Rev.Lett 75, 4350
Hannestad, S. 1998, Phys.Rev.D 57, 2213
Hata, N., Scherrer, R. J., Steigman, G., Thomas, D., Walker, T. P., Bludman, S., & Langacker, P. 1995, Phys.Rev.Lett 75, 3977
Kainulainen, K., Kurki-Suonio, H., & Sihvola, E. 1999, Phys.Rev.D 59, 083505
Kang, H.-S., & Steigman, G. 1992, Nucl.Phys.B 372, 494
Kirilova, D. P. & Chizhov, M. V. 1998, Phys.Rev.D 58, 073004
Kurki-Suonio, H., & Sihvola, E. 1999, astro-ph/9912473
Leonard, R. E., & Scherrer, R. J. 1996, ApJ 463, 420
Lisi, E., Sarkar, S., & Villante, F. L 1999, Phys.Rev.D 59, 123520
LSND Collaboration (Athanassopoulos, C., et al.) 1998, Phys.Rev.Lett 81, 1774; Phys.Rev.C 58, 2489
Malaney, R. A., & Mathews, G. J. 1993, Phys.Rep 229, 145
Particle Data Group (Caso, C., et al.) 1998, Eur.J.Phys.C 3, 1
Peltoniemi, J. 1996, in Proceedings of the 3rd Tallinn Symposium on Neutrino Physics, ed. I. Ots, J. Lõhmus, P. Helde & L. Palgi, hep-ph/9511323
Rehm, J. B., & Jedamzik, K. 1998, Phys.Rev.Lett 81, 3307
Sarkar, S. 1996, Rep.Prog.Phys 59, 1493
Shi, X. 1996, Phys.Rev.D 54, 2753
Shi, X., Fuller, G. M., & Abazajian, K. 1999, Phys.Rev.D 60, 063002
Steigman, G. 1976, ARA&A 14, 339
Super-Kamiokande Collaboration (Fukuda, Y., et al.) 1998, Phys.Rev.Lett 81, 1562
Tytler, D., Fan, X. M., & Burles, S. 1996, Nature 381, 207
Webb, J. K., Carswell, R. F., Lanzetta, K. M., Ferlet, R., Lemoine, M., Vidal-Madjar, A., & Bowen, D. V. 1997, Nature 388, 250
|
warning/0002/math0002019.html
|
ar5iv
|
text
|
# Lattice Substitution Systems and Model Sets
## 1 Introduction
There have been two very successful approaches to building discrete mathematical structures with long-range aperiodic order. These are the substitution methods, notably symbolic substitutions and tiling substitutions, and the cut and project method. In the first case the structure is typically generated by successive substitution from a finite starting configuration. In the second it typically appears in one shot as the (partial) projection of a periodic structure in some “higher” dimensional embedding space.
The principal focus in this paper is the relationship between matrix substitution systems on a lattice and a naturally related cut and project formalism. We start with a partition of a lattice $`L`$ in $`^n`$ into a finite number of point sets $`\stackrel{~}{U}=(U_1,\mathrm{},U_m)`$ and a finite set of substitution rules $`\mathrm{\Phi }`$ which are affine inflations and under which $`\stackrel{~}{U}`$ is invariant. The main theorem (Theorem 3) provides conditions on $`\mathrm{\Phi }`$ which are equivalent to $`U_1,\mathrm{},U_m`$ being regular model sets (i.e. cut and project sets). One of the characterizations (modular coincidence) affords a simple computational approach to testing for model sets. In a later section we go beyond the context of substitution systems and provide an alternative characterization (Theorem 4) of model sets. We use both types of characterization in showing that the sphinx and $`n`$-dimensional chair tilings are based on model sets.
The connection between substitution systems and cut and project sets is nothing new, e.g. the Fibonacci chain is often described in terms of a cut by a strip through $`^2`$, and the klotz construction of Kramer et al. is a sophisticated elaboration of the same idea. Nonetheless, substitution systems and cut and project sets are not different formulations of the same thing, and the relationship between them remains inadequately understood.
In the early study of aperiodic order, the cut and project formalism was always based on projection into $`^n`$ from a lattice in some higher space $`^n\times ^p`$, the projection being controlled by a compact set $`W^p`$. However, it was already implicit in the much earlier work of Y. Meyer that $`^p`$ can be replaced by any locally compact abelian group $`H`$ and $`WH`$ by any compact set with non-empty interior, and the projection method still produces discrete aperiodic sets with diffractive properties (hence long-range order). Meyer’s terminology for such sets was “model sets” and we use it here in deference to its priority and to emphasize the greater generality of the internal space $`H`$. Model sets have been studied in detail in . The relevance of more general internal spaces to tiling theory and symbolic substitutions was made explicit in where $`p`$-adic and mixed $`p`$-adic and real spaces naturally appear.
One of the important features of making the connection to model sets is that once it is established, pure point diffractivity is assured (see Theorem 2 for a precise statement of this). This type of information is generally quite difficult to obtain. For example, our results shown here prove that the $`n`$-dimensional chair tiling and the $`2`$-dimensional sphinx tiling are pure point diffractive. The former has been established for $`n=2`$ previously . The latter is claimed in as being provable by a geometric form of “coincidence” established there (see below for more on the concept of coincidence).
The $`p`$-adic type internal spaces occur when the aperiodic set in question is based on the points of a lattice and its sublattices in $`^n`$. An important class of examples of this type arises from the equal length symbolic substitution systems. Suppose that $`A=\{a_1,\mathrm{},a_m\}`$ is a finite alphabet with associated monoid of words $`A^{}`$, and we are given a primitive substitution $`\sigma :AA^{}`$ for which the length $`l`$ of each of the words $`\sigma (a_i)`$ is the same. This substitution leads to a tiling of $``$ of tiles of equal length, say equal to $`1`$. Matching the coordinate of the left end of each tile with its tile type $`a_i`$, we obtain a partition $`U_1\mathrm{}U_m`$ of $``$, and $`\sigma `$ may be viewed as comprised of a set of affine mappings $`xlx+v`$ where $`v`$.
A lot more is known about equal length substitutions than the arbitrary ones, a particularly important example of this being Dekking’s criterion for diffraction . An equal length aperiodic tiling is pure point diffractive if and only if it admits a coincidence ($`\sigma `$ is said to admit a coincidence if there is a $`k`$, $`1kl^n`$, for which the $`k`$th letter of each word $`\sigma ^n(a_i)`$ for some $`n`$ is the same).
In this paper we prove a related result, but this time the dimension is not restricted. Namely, there is a notion of coincidence (in fact there are two such notions) and either of these is equivalent to the sets $`U_1,\mathrm{},U_m`$ being regular model sets. One of the criteria for coincidence that we give is a straightforward algorithm and thus in principle is computable.
As we have already pointed out, a consequence of our result is that coincidence implies the pure point diffractivity of $`U_1,\mathrm{},U_m`$. We do not know yet to what extent the condition is equivalent to pure point diffractivity.
The setting of the paper is entirely at the level of point sets, so necessarily the strong conditions implicit in the tiling situation are replaced here by a corresponding algebraic condition on the matrix substitution system: the Perron-Frobenius eigenvalue of the substitution system should equal its inflation constant. This is in fact a compatibility condition which is necessary for the model set connection to exist. This condition, not surprisingly, has occurred elsewhere in the literature (see for instance, ). The important result that gets the process off the ground is Theorem 1, which is largely due to Martin Schlottmann.
Matrix substitution systems, treated at the level of point sets, have recently appeared in Lagarias and Wang under the name of self-replicating Delone sets. In that paper, point sets $`X`$ are not restricted to lattices and the principal question revolves around the interesting question of existence of tilings of $`^n`$ by translations of certain prototiles for which the points of $`X`$ are the appropriate translational vectors. Also related to our paper is the study of sets of affine mappings in the context of lattice tilings (see for a nice recent survey on this). In relation to our paper, the situation there corresponds to the $`1\times 1`$ matrix substitution systems and the problems become entirely different. Since the tilings there are lattice tilings, the whole issue of model sets and diffracion is trivial, and the issues lie more around the complex nature of the tiles themselves.
## 2 Definitions and Notation
Let $`X`$ be a nonempty set. For $`m_+`$, an $`m\times m`$ matrix function system (MFS) on $`X`$ is an $`m\times m`$ matrix $`\mathrm{\Phi }=(\mathrm{\Phi }_{ij})`$, where each $`\mathrm{\Phi }_{ij}`$ is a set (possibly empty) of mappings $`X`$ to $`X`$.
The corresponding matrix $`S(\mathrm{\Phi }):=(\text{card}(\mathrm{\Phi }_{ij}))_{ij}`$ is called the substitution matrix of $`\mathrm{\Phi }`$. The MFS is primitive if $`S(\mathrm{\Phi })`$ is primitive, i.e. there is an $`l>0`$ for which $`S(\mathrm{\Phi })^l`$ has no zero entries.
In this paper we deal only with MFSs which are finite in the sense that card$`(\mathrm{\Phi }_{ij})<\mathrm{}`$ for all $`i,j`$. Of particular importance are the Perron-Frobenius (PF) eigenvalue and the corresponding PF eigenvector (unique up to a scalar factor) of $`S(\mathrm{\Phi })`$. We will also have use for the incidence matrix $`I(\mathrm{\Phi })`$ of $`\mathrm{\Phi }`$, which is defined by
$$(I(\mathrm{\Phi }))_{ij}=\{\begin{array}{cc}1\hfill & \text{if}\text{card}(\mathrm{\Phi }_{ij})0,\hfill \\ 0\hfill & \text{else}.\hfill \end{array}$$
Let $`P(X)`$ be the set of subsets of $`X`$. Any MFS induces a mapping on $`P(X)^m`$ by
$`\mathrm{\Phi }\left[\begin{array}{c}U_1\\ \mathrm{}\\ U_m\end{array}\right]`$ $`=`$ $`\left[\begin{array}{c}_j_{f\mathrm{\Phi }_{1j}}f(U_j)\\ \mathrm{}\\ _j_{f\mathrm{\Phi }_{mj}}f(U_j)\end{array}\right]_,`$ (7)
which we call the substitution determined by $`\mathrm{\Phi }`$. We sometimes write $`\mathrm{\Phi }_{ij}(U_j)`$ to mean $`_{f\mathrm{\Phi }_{ij}}f(U_j)`$.
In the sequel, $`X`$ will be a lattice $`L`$ in $`^n`$ and the mappings of $`\mathrm{\Phi }`$ will always be affine linear mappings of the form $`xQx+a`$, where $`Q\text{End}_{}(L)`$ is the same for all the maps. Such maps extend to $`^n`$. For any affine mapping $`f:xQx+b`$ on $`L`$ we denote the translational part, $`b`$, of $`f`$ by $`t(f)`$. We say that $`f,g\mathrm{\Phi }`$ are congruent mod $`QL`$ if $`t(f)t(g)`$ mod $`QL`$. This equivalence relation partitions $`\mathrm{\Phi }`$ into congruence classes. For $`aL`$, $`\mathrm{\Phi }[a]:=\{f_{i,j}\mathrm{\Phi }_{ij}|t(f)a\text{mod}QL\}.`$
We say that $`\mathrm{\Phi }`$ admits a coincidence if there is an $`i,1im`$, for which $`_{j=1}^m\mathrm{\Phi }_{ij}\mathrm{}`$, i.e. the same map appears in every set of the $`i`$-th row for some $`i`$. Furthermore, if $`\mathrm{\Phi }^M[a]`$ is contained entirely in one row of the MFS $`(\mathrm{\Phi }^M)`$ for some $`M>0`$, $`aL`$, then we say that $`(\stackrel{~}{U},\mathrm{\Phi })`$ admits a modular coincidence.
Let $`\mathrm{\Phi },\mathrm{\Psi }`$ be $`m\times m`$ MFSs on $`X`$. Then we can compose them :
$$\mathrm{\Psi }\mathrm{\Phi }=((\mathrm{\Psi }\mathrm{\Phi })_{ij}),$$
(8)
where $`(\mathrm{\Psi }\mathrm{\Phi })_{ij}=_{k=1}^m\mathrm{\Psi }_{ik}\mathrm{\Phi }_{kj}`$and$`\mathrm{\Psi }_{ik}\mathrm{\Phi }_{kj}:=\{\begin{array}{c}\{gf|g\mathrm{\Psi }_{ik},f\mathrm{\Phi }_{kj}\}\hfill \\ \mathrm{}\text{if}\mathrm{\Psi }_{ik}=\mathrm{}\text{or}\mathrm{\Phi }_{kj}=\mathrm{}.\hfill \end{array}`$ Evidently, $`S(\mathrm{\Psi }\mathrm{\Phi })S(\mathrm{\Psi })S(\mathrm{\Phi })`$ (see (16) for the definition of the partial order).
For an $`m\times m`$ MFS $`\mathrm{\Phi }`$, we say that $`\stackrel{~}{U}:=[U_1,\mathrm{},U_m]^TP(X)^m`$ is a fixed point of $`\mathrm{\Phi }`$ if $`\mathrm{\Phi }\stackrel{~}{U}=\stackrel{~}{U}`$.
## 3 Substitution Systems on Lattices
Let $`L`$ be a lattice in $`^n`$. A mapping $`Q\text{End}_{}(L)`$ is an inflation for $`L`$ if $`detQ0`$ and
$$\underset{k=0}{\overset{\mathrm{}}{}}Q^kL=\{0\}.$$
(9)
Let $`Q`$ be an inflation. Then $`q:=|detQ|=[L:QL]>1`$. We define the $`Q`$-adic completion
$$\overline{L}=\overline{L_Q}=\underset{k}{lim}L/Q^kL$$
(10)
of $`L`$. $`\overline{L}`$ will be supplied with the usual topology of a profinite group. In particular, the cosets $`a+Q^k\overline{L},aL,k=0,1,2,\mathrm{}`$ , form a basis of open sets of $`\overline{L}`$ and each of these cosets is both open and closed. When we use the word coset in this paper, we mean either a coset of the form $`a+Q^k\overline{L}`$ in $`\overline{L}`$ or $`a+Q^kL`$ in $`L`$, according to the context. An important observation is that any two cosets in $`\overline{L}`$ are either disjoint or one is contained in the other. The same applies to cosets of $`L`$.
We let $`\mu `$ denote Haar measure on $`\overline{L}`$, normalized so that $`\mu (\overline{L})=1`$. Thus for cosets,
$$\mu (a+Q^k\overline{L})=\frac{1}{|detQ|^k}=\frac{1}{q^k}.$$
(11)
We also have need of the metric $`d`$ on $`\overline{L}`$ defined via the standard norm:
$$x:=\frac{1}{q^k}\text{if}xQ^k\overline{L}\backslash Q^{k+1}\overline{L},0=0.$$
(12)
From $`_{k=0}^{\mathrm{}}Q^kL=\{0\}`$, we conclude that the mapping $`x\{x\text{mod}Q^kL\}_k`$ embeds $`L`$ in $`\overline{L}`$. We identify $`L`$ with its image in $`\overline{L}`$. Note that $`\overline{L}`$ is the closure of $`L`$, whence the notation.
An affine lattice substitution system on $`L`$ with inflation $`Q`$ is a pair $`(\stackrel{~}{U},\mathrm{\Phi })`$ consisting of disjoint subsets $`\{U_i\}_{i=1}^m`$ of $`L`$ and an $`m\times m`$ MFS $`\mathrm{\Phi }`$ on $`L`$ for which $`\stackrel{~}{U}=[U_1,\mathrm{},U_m]^T`$ is a fixed point of $`\mathrm{\Phi }`$, i.e.
$$U_i=\underset{j=1}{\overset{m}{}}\underset{f\mathrm{\Phi }_{ij}}{}f(U_j),i=1,\mathrm{},m,$$
(13)
where the maps of $`\mathrm{\Phi }`$ are affine mappings of the form $`xQx+a,aL`$, and in which the unions in (13) are disjoint.<sup>1</sup><sup>1</sup>1 In the case that one has unions (13) which are not disjoint there arises the natural question of the mulitplicities of points, or more generally densities of points. For more on this see . In this paper all our matrix substitution systems are composed of affine mappings on a lattice and we often drop the words ‘affine lattice’, speaking simply of substitution systems.
We say that the substitution system $`(\stackrel{~}{U},\mathrm{\Phi })`$ is primitive if $`\mathrm{\Phi }`$ is primitive. A second substitution system $`(\stackrel{~}{U^{^{}}},\mathrm{\Psi })`$ is called equivalent to $`(\stackrel{~}{U},\mathrm{\Phi })`$ if $`\stackrel{~}{U^{^{}}}=\stackrel{~}{U},`$ $`\mathrm{\Psi }`$ and $`\mathrm{\Phi }`$ have the same inflation, and $`S(\mathrm{\Psi }),S(\mathrm{\Phi })`$ have the same PF-eigenvalue and right PF-eigenvector (up to scalar factor).
Let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a substitution system on $`L`$. Identifying $`L`$ as a dense subgroup of $`\overline{L}`$, we have a unique extension of $`\mathrm{\Phi }`$ to a MFS on $`\overline{L}`$ in the obvious way. Thus if $`f\mathrm{\Phi }_{ij}`$ and $`f:xQx+a`$, then this formula defines a mapping on $`\overline{L}`$, to which we give the same name. Note that $`f`$ is a contraction on $`\overline{L}`$, since $`Qx=\frac{1}{q}x`$ for all $`x\overline{L}`$. Thus $`\mathrm{\Phi }`$ determines a multi-component iterated function system on $`\overline{L}`$. Furthermore defining the compact subsets
$$W_i:=\overline{U_i},i=1,\mathrm{},m,$$
(14)
and using (13) and the continuity of the mapping, we have
$$W_i=\underset{j=1}{\overset{m}{}}\underset{f\mathrm{\Phi }_{ij}}{}f(W_j),i=1,\mathrm{},m,$$
(15)
which shows that $`\stackrel{~}{W}=[W_1,\mathrm{},W_m]^T`$ is the unique attractor of $`\mathrm{\Phi }`$ (see ).
We call $`(\stackrel{~}{W},\mathrm{\Phi })`$ the associated Q-adic system. We cannot expect in general that the decomposition in (15) will be disjoint, so we do not call $`(\stackrel{~}{W},\mathrm{\Phi })`$ a substitution system.
For $`X,Y^n`$, we write
$$\begin{array}{c}XY\text{if}X_iY_i\text{for all}1in\\ X<Y\text{if}X_i<Y_i\text{for all}1in.\end{array}$$
Similarly, for $`A,BM_n()`$
$$\begin{array}{c}AB\text{if}A_{ij}B_{ij}\text{for all}1i,jn\\ A<B\text{if}A_{ij}<B_{ij}\text{for all}1i,jn.\end{array}$$
(16)
We begin by recalling a couple of results from the Perron-Frobenius theory.
###### Lemma 1
Let $`A`$ be a non-negative primitive matrix with PF-eigenvalue $`\lambda `$. If $`0\lambda XAX`$, then $`AX=\lambda X`$.
proof : We can assume $`X0`$. Since $`0\lambda X`$ and $`\lambda >0`$, $`X0`$. Let $`X^{}>0`$ be a PF right-eigenvector of $`A`$. Let $`\alpha =\text{max}\{\frac{X_i}{X_i^{}}|1im\}`$. Then $`X\alpha X^{}`$ and $`X`$ is not strictly less than $`\alpha X^{}`$. Claim $`X=\alpha X^{}`$. If $`X\alpha X^{}`$, then $`0<A^N(\alpha X^{}X)=\alpha \lambda ^NX^{}A^NX`$ for some $`N`$, since A is primitive. So $`\lambda ^NXA^NX<\alpha \lambda ^NX^{},`$ i.e. $`X<\alpha X^{}`$. This is a contradiction. Therefore $`AX=\lambda X`$. $`\mathrm{}`$
###### Lemma 2
Let $`\lambda `$ be the PF-eigenvalue of the non-negative primitive matrix $`A`$ and $`\mu `$ be an eigenvalue of a matrix $`B`$ where $`0BA`$. If $`AB`$, then $`|\mu |<\lambda `$.
proof : Let $`Y`$ be a right eigenvector for eigenvalue $`\mu `$ of $`B`$, with $`Y=[Y_1,\mathrm{},Y_m]^T.`$ Let $`\overline{Y}=[|Y_1|,\mathrm{},|Y_m|]^T0`$. Then $`|\mu |\overline{Y}B\overline{Y}A\overline{Y}`$. Let $`\overline{X}^T`$ be a positive left eigenvector for $`A`$ with PF-eigenvalue $`\lambda `$. So $`|\mu |\overline{X}^T\overline{Y}\overline{X}^TB\overline{Y}\overline{X}^TA\overline{Y}=\lambda \overline{X}^T\overline{Y}`$. This shows that $`|\mu |\lambda `$. If $`|\mu |=\lambda `$, then $`\lambda \overline{Y}A\overline{Y}`$. By Lemma 1, $`\lambda \overline{Y}=A\overline{Y}`$. Since $`A`$ is a primitive matrix, $`\lambda ^m\overline{Y}=A^m\overline{Y}>0`$ for some $`m`$. So $`\overline{Y}>0`$. From $`\lambda \overline{Y}B\overline{Y}A\overline{Y}=\lambda \overline{Y}`$, we have $`A\overline{Y}=B\overline{Y}`$. Therefore $`A=B`$. $`\mathrm{}`$
###### Lemma 3
Let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a primitive substitution system. Then for all $`l=1,2,\mathrm{},(\stackrel{~}{U},\mathrm{\Phi }^l)`$ is a primitive substitution system.
proof : Let $`i,j,k\{1,2,\mathrm{},m\}`$. All the maps $`g\mathrm{\Phi }_{ik}`$ have domain $`U_k`$ and disjoint images in $`U_i`$. Moreover all the mappings $`g`$ are injective. Likewise all the maps $`f`$ of $`\mathrm{\Phi }_{kj}`$ have domain $`U_j`$ and disjoint images in $`U_k`$. Thus all the maps $`gf\mathrm{\Phi }_{ik}\mathrm{\Phi }_{kj}`$ have domain $`U_j`$ and disjoint images in $`U_i`$. Furthermore $`\mathrm{\Phi }^2\stackrel{~}{U}=\mathrm{\Phi }(\mathrm{\Phi }\stackrel{~}{U})=\mathrm{\Phi }(\stackrel{~}{U})=\stackrel{~}{U}`$. So $`(\stackrel{~}{U},\mathrm{\Phi }^2)`$ is a substitution system. The argument extends in the same way to $`(\stackrel{~}{U},\mathrm{\Phi }^l)`$. The statement on primitivity is clear. $`\mathrm{}`$
###### Theorem 1
Let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a primitive substitution system with inflation $`Q`$ on $`L`$. Let $`(\stackrel{~}{W},\mathrm{\Phi })`$ be the corresponding associated $`Q`$-adic system. Suppose that the PF-eigenvalue of $`S(\mathrm{\Phi })`$ is $`|detQ|`$ and $`\overline{L}=_{i=1}^mW_i`$. Then
$$\begin{array}{c}\text{(i)}S(\mathrm{\Phi }^r)=(S(\mathrm{\Phi }))^r,r1;\hfill \\ \text{(ii)}\mu (W_i)=\frac{1}{q^r}_{j=1}^m(S(\mathrm{\Phi }^r))_{ij}\mu (W_j),\text{for all}i=1,\mathrm{},m,r1;\hfill \\ \text{(iii)}\text{For all}i=1,\mathrm{},m,\stackrel{}{W_i}\mathrm{}\text{and}\mu (W_i)=0.\hfill \end{array}$$
proof : For every measurable set $`EL`$ and all $`f\mathrm{\Phi }_{ij},\mu (f(E))=\mu (Q(E)+a)=\frac{1}{|detQ|}\mu (E),\text{where}f:xQx+a`$. In particular, $`\mu (f(W_j))=\frac{1}{q}w_j,\text{where}w_j:=\mu (W_j)\text{and}q=|detQ|`$. We obtain
$$w_i\underset{j=1}{\overset{m}{}}\frac{1}{q^r}\text{card}((\mathrm{\Phi }^r)_{ij})w_j$$
from (15).
Let $`w=[w_1,\mathrm{},w_m]^T`$. Since $`_{i=1}^mW_i=\overline{L}`$, the Baire category theorem assures us that for at least one $`i`$,
$$\stackrel{}{W_i}\mathrm{}$$
(17)
and then the primitivity gives this for all $`i`$. So $`w>0`$ and
$$w\frac{1}{q^r}S(\mathrm{\Phi }^r)w\frac{1}{q^r}S(\mathrm{\Phi })^rw,\text{for any}r1.$$
(18)
Since the PF-eigenvalue of $`S(\mathrm{\Phi })^r`$ is $`q^r=|detQ|^r`$ and $`S(\mathrm{\Phi })^r`$ is primitive, we have from Lemma 1 that
$$w=\frac{1}{q^r}S(\mathrm{\Phi }^r)w=\frac{1}{q^r}S(\mathrm{\Phi })^rw,\text{for any}r1.$$
(19)
The positivity of $`w`$ together with $`S(\mathrm{\Phi }^r)S(\mathrm{\Phi })^r`$ shows that $`S(\mathrm{\Phi }^r)=S(\mathrm{\Phi })^r`$. This proves (i) and (ii).
Fix any $`i\{1,\mathrm{},m\}`$, let $`\stackrel{}{W_i}`$ contain a basis open set $`a+Q^r\overline{L}`$ with some $`r_0`$ by (17). Since $`(\stackrel{~}{U},\mathrm{\Phi }^r)`$ is a substitution system, $`a+Q^r\overline{L}\stackrel{}{W_i}W_i=_{j=1}^m(\mathrm{\Phi }^r)_{ij}W_j`$. In particular, $`(a+Q^r\overline{L})g(W_k)\mathrm{}`$ for some $`k\{1,\mathrm{},m\}`$ and some $`g(\mathrm{\Phi }^r)_{ik}`$. However $`g(\overline{L})=b+Q^r\overline{L}`$ for some $`bL`$, so $`(a+Q^r\overline{L})(b+Q^r\overline{L})\mathrm{}`$. This means $`a+Q^r\overline{L}=b+Q^r\overline{L}`$. Thus
$$g(W_k)g(\overline{L})=a+Q^r\overline{L}\stackrel{}{W_i}.$$
(20)
For all $`f(\mathrm{\Phi }^r)_{ij},j\{1,2,\mathrm{},m\}`$, $`f`$ is clearly an open map, so $`_{j=1}^m(\mathrm{\Phi }^r)_{ij}(\stackrel{}{W_j})\stackrel{}{W_i}.`$ Thus
$`W_i=W_i\backslash \stackrel{}{W_i}`$ $`=`$ $`\left({\displaystyle \underset{j=1}{\overset{m}{}}}(\mathrm{\Phi }^r)_{ij}(W_j)\right)\backslash \stackrel{}{W_i}`$ (21)
$``$ $`{\displaystyle \underset{j=1}{\overset{m}{}}}\left((\mathrm{\Phi }^r)_{ij}(W_j)\backslash (\mathrm{\Phi }^r)_{ij}(\stackrel{}{W_j})\right)`$
$``$ $`{\displaystyle \underset{j=1}{\overset{m}{}}}(\mathrm{\Phi }^r)_{ij}(W_j).`$
Note that due to (20) at least one $`g`$ in $`(\mathrm{\Phi }^r)_{ij}`$ does not contribute to the relation (21).
Let $`v_i:=\mu (W_i),i=1,\mathrm{},m`$ and $`v:=[v_1,\mathrm{},v_m]^T`$. So $`v\frac{1}{q^r}S(\mathrm{\Phi }^r)v`$. Actually, by what we just said,
$$0v\frac{1}{q^r}S^{}v\frac{1}{q^r}S(\mathrm{\Phi }^r)v=\frac{1}{q^r}S(\mathrm{\Phi })^rv,$$
(22)
where $`S^{}S(\mathrm{\Phi })^r,S^{}S(\mathrm{\Phi })^r`$. Now applying the Lemma 1 again we obtain equality throughout (22). But by Lemma 2 the eigenvalues of $`\frac{1}{q^r}S^{}`$ are strictly less in absolute value than the PF-eigenvalue of $`\frac{1}{q^r}S(\mathrm{\Phi })^r`$, which is $`1`$. This forces $`v=0`$, and hence $`\mu (W_i)=0,i=1,\mathrm{},m`$. $`\mathrm{}`$
In the sequel, the central concern is to relate the sets $`U_i`$ and the sets $`\mathrm{\Lambda }_i:=W_iL`$. Clearly $`\mathrm{\Lambda }_iU_i`$. The next lemma groups a circle of ideas that relate this question to the boundaries and interiors of the $`W_i`$.
###### Lemma 4
Let $`U_i,i=1,\mathrm{},m,`$ be point sets of the lattice $`L`$ in $`^n`$. Let $`Q`$ be an inflation of $`L`$ and identify $`L`$ with its image in its $`Q`$-adic completion $`\overline{L}`$. Define $`W_i:=\overline{U_i}`$ in $`\overline{L}`$ and $`\mathrm{\Lambda }_i:=W_iL`$.
* If $`U_1,\mathrm{},U_m`$ are disjoint and $`\mu (\overline{\mathrm{\Lambda }_i\backslash U_i})=0\text{for all}i=1,\mathrm{},m`$, then $`\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}`$ for all $`ij`$.
* If $`L=_{i=1}^mU_i`$ and $`\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}`$ for all $`ij`$, where $`i,j\{1,\mathrm{},m\}`$, then $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j\text{for all}i=1,\mathrm{},m`$.
* If $`\mu (W_j)=0`$ for all $`j=1,\mathrm{},m`$ and $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j`$, then $`\mu (\overline{\mathrm{\Lambda }_i\backslash U_i})=0`$.
proof: (i) Suppose there are $`i,j\{1,\mathrm{},m\}`$ with $`\stackrel{}{W_i}\stackrel{}{W_j}\mathrm{}`$. We can choose $`a(\stackrel{}{W_i}\stackrel{}{W_j})L`$, since $`L`$ is dense in $`\overline{L}`$ and $`\stackrel{}{W_i}\stackrel{}{W_j}`$ is open. Choose $`k_+`$ so that $`a+q^k\overline{L}\stackrel{}{W_i}\stackrel{}{W_j}`$. Note that $`a+q^kL\mathrm{\Lambda }_i\mathrm{\Lambda }_j`$. Then
$`{\displaystyle \underset{i=1}{\overset{m}{}}}(\mathrm{\Lambda }_i\backslash U_i)`$ $``$ $`\left((a+q^kL)\backslash U_i\right)\left((a+q^kL)\backslash U_j\right)`$
$`=`$ $`(a+q^kL)\backslash (U_iU_j)`$
$`=`$ $`a+q^kL,\text{since the }U_i,i=1,\mathrm{},m,\text{are disjoint}.`$
So
$`{\displaystyle \underset{i=1}{\overset{m}{}}}\mu (\overline{\mathrm{\Lambda }_i\backslash U_i})`$ $``$ $`\mu ({\displaystyle \underset{i=1}{\overset{m}{}}}(\overline{\mathrm{\Lambda }_i\backslash U_i}))`$
$``$ $`\mu (a+q^k\overline{L})`$
$`>`$ $`0,`$
contrary to assumption.
(ii) Assume $`\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}`$ for all $`ij`$. For any $`i\{1,\mathrm{},m\}`$,
$`(\mathrm{\Lambda }_i\backslash U_i)`$ $``$ $`({\displaystyle \underset{ji}{}}U_j)W_i,\text{since}L={\displaystyle \underset{i=1}{\overset{m}{}}}U_i`$
$``$ $`{\displaystyle \underset{ji}{}}(W_jW_i){\displaystyle \underset{j=1}{\overset{m}{}}}W_j,\text{since}\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}\text{for all}ij.`$
(iii) Obvious. $`\mathrm{}`$
## 4 Model Sets
Let us recall the notion of a model set (or cut and project set). A cut and project scheme (CPS) consists of a collection of spaces and mappings as follows;
$$\begin{array}{ccccc}^n& \stackrel{\pi _1}{}& ^n\times G& \stackrel{\pi _2}{}& G\\ & & & & \\ & & \stackrel{~}{L}& & \end{array}$$
(23)
where $`^n`$ is a real Euclidean space, $`G`$ is some locally compact Abelian group, and $`\stackrel{~}{L}^n\times G`$ is a lattice, i.e. a discrete subgroup for which the quotient group $`(^n\times G)/\stackrel{~}{L}`$ is compact. Furthermore, we assume that $`\pi _1|_{\stackrel{~}{L}}`$ is injective and $`\pi _2(\stackrel{~}{L})`$ is dense in $`G`$.
A model set in $`^n`$ is a subset of $`^n`$ which, up to translation, is of the form $`\mathrm{\Lambda }(V)=\{\pi _1(x)|x\stackrel{~}{L},\pi _2(x)V\}`$ for some cut and project scheme as above, where $`VG`$ has non-empty interior and compact closure (relatively compact). When we need to be more precise we explicitly mention the cut and project scheme from which a model set arises. This is quite important in some of the theorems below. Model sets are always Delone subsets of $`^n`$, that is to say, they are relatively dense and uniformly discrete.
We call the model set $`\mathrm{\Lambda }(V)`$ regular if the boundary $`V=\overline{V}\backslash \stackrel{}{V}`$ of $`V`$ is of (Haar) measure $`0`$. We will also find it convenient to consider certain degenerate types of model sets. A weak model set is a set in $`^n`$ of the form $`\mathrm{\Lambda }(V)`$ where we assume only that $`V`$ is relatively compact, but not that it has a non-empty interior. When $`V`$ has no interior, $`\mathrm{\Lambda }(V)`$ is not necessarily relatively dense in $`^n`$ but regularity still means that the boundary of $`V`$ is of measure $`0`$.
###### Theorem 2
(Schlottmann ) If $`\mathrm{\Lambda }=\mathrm{\Lambda }(V)`$ is a regular model set, then $`\mathrm{\Lambda }`$ is a pure point diffractive set, i.e. the Fourier transform of its volume averaged autocorrelation measure is a pure point measure.
$`\mathrm{}`$
It is this theorem that is a prime motivation for finding criteria for sets to be model sets.
Now let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a substitution system with inflation $`Q`$ on a lattice $`L`$ of $`^n`$ and let $`\overline{L}`$ be the $`Q`$-adic completion of $`L`$. This gives rise to the cut and project scheme.
$$\begin{array}{ccccc}^n& \stackrel{\pi _1}{}& ^n\times \overline{L}& \stackrel{\pi _2}{}& \overline{L}\\ & & & & \\ L& & \stackrel{~}{L}& & L\\ t& & (t,t)& & t\end{array}$$
(24)
where $`\stackrel{~}{L}:=\{(t,t)|tL\}^n\times \overline{L}`$.
We claim that $`(^n\times \overline{L})/\stackrel{~}{L}`$ is compact. $`\stackrel{~}{L}`$ is clearly discrete and closed in $`^n\times \overline{L}`$. Since $`(^n\times \overline{L})/\stackrel{~}{L}`$ is Hausdorff and satisfies the first axiom of countability, it is enough to show that it is sequentially compact . If $`\{(x_i,z_i)+\stackrel{~}{L}\}`$ is a countable sequence in $`(^n\times \overline{L})/\stackrel{~}{L}`$, then there is a subsequence $`\{(x_i,z_i)+\stackrel{~}{L}\}_S`$ with $`\{x_i+L\}_S`$ convergent sequence, since $`^n/L`$ is compact. We can rewrite $`\{(x_i,z_i)+\stackrel{~}{L}\}_S`$ as $`\{(x_i^{},z_i^{})+\stackrel{~}{L}\}_S`$, where $`\{x_i^{}\}_{iS}`$ converges to $`x`$ in $`^n`$. Since $`\overline{L}`$ is compact, there is a convergent subsequence $`\{z_i^{}\}_S^{}`$ to some $`z`$ in $`\overline{L}`$. Thus $`\{(x_i^{},z_i^{})\}_S^{}`$ converges to $`(x,z)`$ in $`^n\times \overline{L}`$. Therefore $`(^n\times \overline{L})/\stackrel{~}{L}`$ is sequentially compact.
Note also that $`\pi _1|_{\stackrel{~}{L}}`$ is injective and $`\pi _2(\stackrel{~}{L})`$ is dense in $`\overline{L}`$.
###### Lemma 5
Let $`U_i,i=1,\mathrm{},m,`$ be disjoint point sets of the lattice $`L`$ in $`^n`$. Identify $`L`$ and its image in $`\overline{L}`$. Let $`W_i:=\overline{U_i}`$ in $`\overline{L}`$ and $`\mathrm{\Lambda }_i:=W_iL`$. Suppose that $`\mu (W_i)=0`$ for all $`i=1,\mathrm{},m`$.
(i) If $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j`$ then, relative to the CPS(24), $`U_i`$ is a regular weak model set when $`\stackrel{}{W_i}`$ is empty, and $`U_i`$ is a regular model set when $`\stackrel{}{W_i}`$ is non-empty.
(ii) If $`L=_{j=1}^mU_j`$ and each $`U_i`$ is a regular model set, then $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j`$ for all $`i=1,\mathrm{},m`$.
proof: (i) Assume that $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j`$ for all $`i=1,\mathrm{},m`$. Since $`\mu (W_i)=0`$ for all $`i=1,\mathrm{},m`$,
$$\mu (W_i)=\mu (\stackrel{}{W_i})=\mu (\stackrel{}{W_i}\backslash \underset{j=1}{\overset{m}{}}W_j)$$
(25)
Since $`\mathrm{\Lambda }_i=W_iL`$, $`U_i=V_iL`$ where $`V_i:=W_i\backslash (\mathrm{\Lambda }_i\backslash U_i)`$. Now $`V_i\stackrel{}{W_i}\backslash _{j=1}^mW_j`$. From $`\stackrel{}{W_i}\backslash _{j=1}^mW_j\stackrel{}{V_i}V_i\overline{V_i}=W_i`$ and (25), $`\mu (\overline{V_i}\backslash \stackrel{}{V_i})=0`$. So $`U_i`$ is regular. If $`\stackrel{}{W_i}=\mathrm{}`$, then $`\stackrel{}{V_i}=\mathrm{}`$ also. Thus $`U_i`$ is a regular weak model set. On the other hand, for any $`i`$ with $`\stackrel{}{W_i}\mathrm{}`$, $`\stackrel{}{V_i}\mathrm{}`$ and $`\overline{V_i}`$ is compact. It follows that $`U_i=\mathrm{\Lambda }(V_i)`$ is a regular model set for the CPS (24).
(ii) Suppose that $`\stackrel{}{V_i}\mathrm{}`$, $`\mu (\overline{V_i}\backslash \stackrel{}{V_i})=0`$, where $`U_i=V_iL`$, and $`L=_{j=1}^mU_j`$. Then from $`\overline{\mathrm{\Lambda }_i\backslash U_i}=\overline{\mathrm{\Lambda }(W_i)\backslash \mathrm{\Lambda }(V_i)}\overline{W_i\backslash V_i}W_i\backslash \stackrel{}{V_i}=\overline{V_i}\backslash \stackrel{}{V_i}`$, we have $`\mu (\overline{\mathrm{\Lambda }_i\backslash U_i})=0`$ for all $`i=1,\mathrm{},m`$. By Lemma 4 (i) and (ii) , $`\stackrel{}{W_i}\stackrel{}{W_j}=0`$ for all $`ij`$ and $`\mathrm{\Lambda }_i\backslash U_i_{j=1}^mW_j`$. $`\mathrm{}`$
###### Theorem 3
Let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a primitive substitution system with inflation $`Q`$ on the lattice $`L`$ in $`^n`$. Suppose that PF-eigenvalue of the substitution matrix $`S(\mathrm{\Phi })`$ is equal to $`|detQ|`$ and $`L=_{i=1}^mU_i`$. Then the following are equivalent.
(i) There is a primitive substitution matrix $`\mathrm{\Psi }`$ admitting a coincidence, where $`(\stackrel{~}{U},\mathrm{\Psi })`$ is equivalent to $`(\stackrel{~}{U},\mathrm{\Phi }^M)`$ for some $`M1`$.
(ii) The sets $`U_i,i=1,\mathrm{},m,`$ of $`\stackrel{~}{U}`$ are model sets for the CPS (24).
(iii) For at least one $`i`$, $`U_i`$ contains a coset $`a+Q^ML`$.
(iv) $`(\stackrel{~}{U},\mathrm{\Phi })`$ admits a modular coincidence.
proof :
(i) $``$ (ii): Suppose that $`(\stackrel{~}{U},\mathrm{\Psi })`$ admits a coincidence and is equivalent to $`(\stackrel{~}{U},\mathrm{\Phi }^M)`$. Fix $`i\{1,\mathrm{},m\}`$ with $`_{j=1}^m\mathrm{\Psi }_{ij}\mathrm{}`$ and let $`g`$ be in this intersection. Recalling equation (15), and in view of the choice of $`g`$, we have
$$\mu (W_i)\left(\underset{j=1}{\overset{m}{}}\underset{f\mathrm{\Psi }_{ij}}{}\mu (f(W_j))\right)\mu (g(W_k)g(W_l)),$$
for any $`k,l\{1,\mathrm{},m\}`$ with $`kl`$. On the other hand, from Theorem 1 (ii)
$$\mu (W_i)=\frac{1}{q^M}\underset{j=1}{\overset{m}{}}(S(\mathrm{\Psi }))_{ij}\mu (W_j)=\underset{j=1}{\overset{m}{}}\underset{f\mathrm{\Psi }_{ij}}{}\mu (f(W_j)).$$
(26)
Thus, in fact, $`\mu (g(W_k)g(W_l))=0`$ whenever $`kl`$. It follows at once that $`\stackrel{}{W_k}\stackrel{}{W_l}=\mathrm{}`$ for all $`kl`$, since the measure of any open set is larger than $`0`$.
Recall that $`\stackrel{}{W_i}\mathrm{}`$ and $`\mu (W_i)=0`$ for all $`i=1,\mathrm{},m`$ . Then by Lemma 4(ii) and Lemma 5, $`U_i,i=1,\mathrm{},m,`$ are model sets in CPS(24).
(ii) $``$ (iii): Assume that $`U_i,i=1,\mathrm{},m,`$ are model sets in CPS(24), i.e. $`U_i=\mathrm{\Lambda }(V_i)=V_iL`$ for some $`V_i`$ with $`\stackrel{}{V_i}\mathrm{}`$. Thus there is a coset $`a+Q^M\overline{L}\stackrel{}{V_i}`$ and, since we can always choose the coset representative from the dense lattice $`L`$, we can arrange that $`a+Q^MLU_i`$.
(iii) $``$ (iv): Assume that for at least one $`i`$, $`U_i`$ contains a coset $`a+Q^ML`$. Fix $`i`$. Iterate $`\mathrm{\Phi }`$ M-times. Then each function $`f`$ in the substitution system $`\mathrm{\Phi }^M`$ has the form $`f:xQ^Mx+b`$. For each $`j`$, let $`G_j:=\{f(\mathrm{\Phi }^M)_{ij}|t(f)a\text{mod}Q^ML\}`$. (Recall that $`t(f)`$ is the translational part of $`f`$). From $`U_i=_{j=1}^m_{f(\mathrm{\Phi }^M)_{ij}}f(U_j)`$, we obtain $`a+Q^ML_{j=1}^m_{fG_j}f(U_j)`$. In fact
$$a+Q^ML=\underset{j=1}{\overset{m}{}}\underset{fG_j}{}f(U_j),$$
(27)
since the right hand side is clearly inside $`a+Q^ML`$. From the fact $`a+Q^MLU_i`$, we get $`\mathrm{\Phi }^M[a]=_{j=1}^mG_j_{j=1}^m(\mathrm{\Phi }^M)_{ij}`$. Therefore $`\mathrm{\Phi }^M`$ has a row containing an entire congruence class $`\mathrm{\Phi }^M[a]`$.
(iv) $``$ (i): Assume $`\mathrm{\Phi }^M`$ has a row, say $`i`$-th row, containing an entire congruence class $`\mathrm{\Phi }^M[a]`$. Let $`G_j:=\mathrm{\Phi }^M[a](\mathrm{\Phi }^M)_{ij}`$. Then $`_{j=1}^m_{fG_j}f(U_j)a+Q^ML`$. Recall that $`_{j=1}^mU_j=L`$ and $`\stackrel{~}{U}=\mathrm{\Phi }^M(\stackrel{~}{U})`$. It follows that the elements of $`a+Q^ML`$ can be obtained from the substitution system $`\mathrm{\Phi }^M`$ only from the mappings of $`\mathrm{\Phi }^M[a]`$, and indeed they must all appear as images of the mappings of $`\mathrm{\Phi }^M[a]`$. Thus
$$a+Q^ML=\underset{j=1}{\overset{m}{}}\underset{fG_j}{}f(U_j)U_i.$$
(28)
On the other hand,
$$a+Q^ML=\underset{j=1}{\overset{m}{}}Q^M(U_j)+a,$$
(29)
which is a disjoint union.
We now alter our substitution system $`\mathrm{\Phi }^M`$ as follows: Define $`g:LL`$ by $`g(x)=Q^Mx+a`$. We may, by restriction of domain, consider $`g`$ as a function on $`U_j,j=1,\mathrm{},m`$. We define $`\mathrm{\Psi }`$ by
$$\{\begin{array}{c}\mathrm{\Psi }_{ij}=((\mathrm{\Phi }^M)_{ij}\backslash G_j)\{g\}\hfill \\ \mathrm{\Psi }_{kj}=(\mathrm{\Phi }^M)_{kj}\text{if}ki,\hfill \end{array}$$
for all $`j`$. From (28) and (29), the $`\mathrm{\Psi }_{ij},j=1,\mathrm{},m`$, consist of maps from $`U_j`$ to $`U_i`$ and have the same total effect on $`U_i`$ as the $`(\mathrm{\Phi }^M)_{ij},j=1,\mathrm{},m`$. Thus $`(\stackrel{~}{U},\mathrm{\Psi })`$ is a substitution system admitting a coincidence.
Since $`S(\mathrm{\Phi }^M)`$ is primitive, the incidence matrix $`I(\mathrm{\Phi }^M)`$ is primitive. Then $`I(\mathrm{\Psi })`$ is also primitive, since $`I(\mathrm{\Phi }^M)I(\mathrm{\Psi })`$. So $`\mathrm{\Psi }`$ is primitive. In addition, $`\mathrm{\Psi }`$ has the inflation $`Q^M`$ for $`L`$ which is an inflation in $`\mathrm{\Phi }^M`$.
We claim that $`S(\mathrm{\Psi }),S(\mathrm{\Phi }^M)`$ have the same PF-eigenvalue and right PF-eigenvector. Then $`(\stackrel{~}{U},\mathrm{\Psi })`$ is equivalent to $`(\stackrel{~}{U},\mathrm{\Phi }^M)`$.
We verify first that $`\stackrel{}{W_k}\stackrel{}{W_j}=\mathrm{}\text{for all}kj`$. We can assume that $`m>1`$, since there is nothing to prove when $`m=1`$. Let $`g_1G_l=(\mathrm{\Phi }^M)_{il}[a]\mathrm{}`$ for some $`l`$. Take any $`k\{1,\mathrm{},m\}`$. There is $`M_0_+`$ for which $`(\mathrm{\Phi }^{M_0})_{lk}\mathrm{}`$. Choose $`f(\mathrm{\Phi }^{M_0})_{lk}`$. Let $`g_1:xQ^Mx+a_1`$, where $`a_1a\text{mod}Q^ML`$, and $`f:xQ^{M_0}x+b`$ with $`bL`$. Then $`g_1f:xQ^{M+M_0}x+Q^Mb+a_1`$. So $`g_1f(\mathrm{\Phi }^{M+M_0})_{ik}[a_1+Q^Mb]`$. Furthermore $`(a_1+Q^Mb)+Q^{M+M_0}(L)a_1+Q^MLU_i`$.
Let $`N:=M+M_0,c:=a_1+Q^Mb,\text{and}p:=g_1f`$. Note that
$$c+Q^NL=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}h(U_j),$$
(30)
where $`H_j=(\mathrm{\Phi }^N)_{ij}[c].`$
There are at least two functions in $`_{j=1}^mH_j`$, since for all $`j`$ $`U_j,L`$. We can write $`c+Q^NL`$ in the form
$$c+Q^NL=\{Q^NU_j+Q^N\alpha _h+c|j\{1,\mathrm{},m\},hH_j,\alpha _hL\},$$
(31)
where we have used the explicit form of each of the mappings $`hH_j`$. This union is disjoint, and as a consequence the elements $`\alpha _hL`$ for $`h`$ in any single $`H_j`$ are all distinct. In particular we have $`\alpha _p`$ coming from $`H_k`$. From (31) we have
$$L=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}(U_j+\alpha _h)$$
(32)
and separating off $`U_k`$,
$$L=U_k\underset{j=1}{\overset{m}{}}\underset{hH_j^{}}{}(U_j+\alpha _h\alpha _p),$$
(33)
where $`H_j^{}:=H_j`$ if $`jk`$ and $`H_k^{}:=H_k\backslash \{p\}`$. Again these decompositions are disjoint. But we also know that $`U_k`$ and $`_{\stackrel{j=1}{jk}}^mU_j`$ are disjoint, and it follows that
$$\underset{\stackrel{j=1}{jk}}{\overset{m}{}}U_j\underset{j=1}{\overset{m}{}}\underset{hH_j^{}}{}(U_j+\alpha _h\alpha _p).$$
Taking closures,
$$\underset{\stackrel{j=1}{jk}}{\overset{m}{}}W_j\underset{j=1}{\overset{m}{}}\underset{hH_j^{}}{}(W_j+\alpha _h\alpha _p).$$
(34)
On the other hand, if we apply Theorem 1(ii) to $`\mathrm{\Phi }^N`$ and look at (30) we see that
$$\mu (c+Q^N\overline{L})=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}\mu (h(W_j))=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}\mu (Q^N(W_j+\alpha _h)+c),$$
and hence
$$\mu (\overline{L})=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}\mu (W_j+\alpha _h)=\underset{j=1}{\overset{m}{}}\underset{hH_j}{}\mu (W_j+\alpha _h\alpha _p).$$
Thus
$$\mu (\overline{L})=\mu (W_k)+\left(\underset{j=1}{\overset{m}{}}\underset{hH_j^{}}{}\mu (W_j+\alpha _h\alpha _p)\right)$$
which, after taking closures in (33), gives
$$\mu \left(W_k\left(\underset{j=1}{\overset{m}{}}\underset{hH_j^{}}{}(W_j+\alpha _h\alpha _p)\right)\right)=0.$$
(35)
Finally from (34) and (35) we obtain
$$\mu (W_k(\underset{\stackrel{j=1}{jk}}{\overset{m}{}}W_j))=0,$$
from which $`\stackrel{}{W_k}\stackrel{}{W_j}=\mathrm{}`$ for all $`kj`$. This establishes the claim.
Now
$`\mu \left({\displaystyle \underset{j=1}{\overset{m}{}}}g(W_j)\right)`$ $`=`$ $`{\displaystyle \frac{1}{|detQ^M|}}\mu \left({\displaystyle \underset{j=1}{\overset{m}{}}}W_j\right)`$ (36)
$`=`$ $`{\displaystyle \frac{1}{|detQ^M|}}{\displaystyle \underset{j=1}{\overset{m}{}}}\mu (W_j),`$
$`\text{from}\mu (W_j)=0,\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}\text{for all}ij`$
$`=`$ $`{\displaystyle \underset{j=1}{\overset{m}{}}}\mu (g(W_j)).`$
Again using Theorem 1 (ii), this time for $`\mathrm{\Phi }^M`$, we obtain
$$w=\frac{1}{|\text{det}Q^M|}S(\mathrm{\Phi }^M)w,$$
where $`w=[\mu (W_1),\mathrm{},\mu (W_m)]^T`$. The part of this relation in $`W_i`$ which pertains to the coset $`a+Q^M\overline{L}`$ is
$$\mu (a+Q^M\overline{L})=\underset{j=1}{\overset{m}{}}\underset{fG_j}{}\mu (f(W_j)).$$
(37)
But from (29)
$$\mu (a+Q^M\overline{L})=\mu \left(\underset{j=1}{\overset{m}{}}g(W_j)\right).$$
(38)
Together, (36), (37), and (38) show
$$w=\frac{1}{|\text{det}Q^M|}S(\mathrm{\Psi })w.$$
Since $`w>0`$ and $`S(\mathrm{\Psi })`$ is primitive, $`S(\mathrm{\Psi })`$ has PF-eigenvalue $`|\text{det}Q^M|`$ and PF-eigenvector $`w`$ as required. $`\mathrm{}`$
Remark: Let $`A=\{a_1,\mathrm{},a_m\}`$ be an alphabet of $`m`$ symbols and let $`\sigma `$ be a primitive equal-length alphabetic substitution system on $`A`$, that is,
* $`\sigma :AA^q`$ for some $`q_+`$;
* the $`m\times m`$ matrix $`S=(S_{ij})`$, whose $`i,j`$ entry is the number of appearances of $`a_i`$ in $`\sigma (a_j)`$, is primitive.
According to Gottschalk , for some iteration $`\sigma ^k`$ of $`\sigma `$, there is a word $`wA^{}`$ which is fixed by $`\sigma `$ in the sense that
$$\begin{array}{ccc}\sigma ^k(w_0w_1\mathrm{})& =& w_0w_1\mathrm{}\\ \sigma ^k(\mathrm{}w_2w_1)& =& \mathrm{}w_2w_1.\end{array}$$
(39)
Replacing $`\sigma ^k`$ by $`\sigma `$ and $`q^k`$ by $`q`$ if necessary we can suppose that $`k=1`$, and assume then that $`\sigma (w)=w`$.
We can view $`w`$ as a tiling of $``$ by tiles of types $`a_1,\mathrm{}a_m`$, all of the same length $`1`$. If we coordinatize each tile by its lefthand end point so that $`w_l`$ gets coordinate $`l`$, then we obtain a partition $`U_1\mathrm{}U_m`$ of $``$ and an $`m\times m`$ matrix substitution system $`\mathrm{\Phi }`$ of $`q`$-affine mappings derived directly from $`\sigma `$: namely, $`\sigma a_j=a_{i_1}\mathrm{}a_{i_q}`$ gives rise to the mappings $`(xqx+l1)\mathrm{\Phi }_{i_lj}`$, $`l=1,\mathrm{},q`$.
We take as our cut and project scheme
$$\begin{array}{ccccc}& & \times _q& & _q\\ & & & & \\ & & \stackrel{~}{}& & \\ z& & (z,z)& & z\end{array}$$
(40)
(see 24), where $`_q`$ is the $`q`$-adic completion of $``$.
According to Theorem 3, the $`U_i`$ are model sets for (40) if and only if for some iteration $`\sigma ^M`$ of $`\sigma `$, there is a $`k`$ for which all the mappings $`f_l:xq^Mx+l`$ with $`lk(\mathrm{mod}q^M)`$ lie in one row of $`\mathrm{\Phi }^M`$.
Since $`\sigma ^Ma_j`$ has $`q^M`$ letters in it, there are $`q^M`$ mappings in the $`j`$th column of $`\mathrm{\Phi }^M`$. Furthermore, since the letters $`\sigma ^Ma_j`$ are represented by contiguous tiles, their coordinates fall in a range of consecutive integers, and so the mappings of the $`j`$th column of $`\mathrm{\Phi }^M`$ are the maps $`f_l`$, where $`0l<q^M`$, in some order. In particular, all of the mappings in $`\mathrm{\Phi }^M`$ are of this restricted form. It follows that modular coincidence is equivalent to the existence of a row of $`\mathrm{\Phi }^M`$, say the $`i`$th row, and a $`k`$, $`0k<q^M`$, so that $`f_k`$ belongs to each of $`\mathrm{\Phi }_{i1}^M,\mathrm{},\mathrm{\Phi }_{im}^M`$.
This condition precisely says that there is a $`k`$ so that the $`k`$th position of $`\sigma ^M(a_j)`$ contains the same letter $`a_i`$ for all $`j`$. This is the well-known coincidence condition of Dekking , and he has proved that for non-periodic primitive equal-length substitutions, this condition is equivalent to pure point diffractivity. It is straighforward to show that $`S(\mathrm{\Phi })`$ has its PF-eigenvalue equal to $`|detQ|`$. Thus we have
###### Corollary 1
Let $`\sigma `$ be a primitive equal-length ($`=q`$) alphabetic substitution with a fixed bi-infinite word $`w`$, and assume that $`w`$ is not periodic. Let $`\mathrm{\Phi }`$ be the corresponding matrix substitution system and let $`=U_1\mathrm{}U_m`$ be the corresponding partition of $``$. Then the following are equivalent:
* there is an $`M`$ so that $`\sigma ^M`$ has a coincidence in the sense of Dekking;
* $`\mathrm{\Phi }`$ has a modular coincidence;
* the $`U_i`$’s are model sets for (40);
* the $`U_i`$’s are pure point diffractive.
We note that this interesting equivalence of model sets and pure point diffractivity is more than we can yet prove in the higher dimensional substitution systems.
## 5 Sphinx tiling
Long we sought the wayward lynx
And bowed before the subtle sphinx
But solved we not the cryptic sphinx
Before we found the wayward links.
-Anon
In this section we take up the sphinx tiling. This is a substitution tiling whose subdivision rule is shown in Figure 1 and Figure 2.
It has $`12`$ sphinx-like tiles (up to translation). If we choose a single point in the same way in each sphinx then we arrive at $`12`$ sets of points. We wish to show that each of these sets is a regular model set. Actually we make a slight alteration to this, choosing several points from each tile, but this is equivalent to our original problem.
Each sphinx can be viewed as consisting of $`6`$ equilateral triangles of two orientations. In this way, any sphinx tiling determines a tessellation of the plane by equilateral triangles. We consider the centre points of the triangles of one orientation. These clearly form a lattice $`L`$, once we have chosen one of them as the origin. Note that some sphinxes have two points and others have four points in $`L`$. We give names to each tile and the points in it as shown in Figure 3. Then the $`12`$ types of sphinx partition $`L`$ into 36 subsets forming a matrix substitution system. We show that these are model sets for a $`2`$-adic-like cut and project scheme of the form of (24).
With the origin as shown, the coordinates are chosen so that in the standard rectangular system $`(1,0)`$ is the lattice point directly to the right of $`(0,0)`$. It is more convenient to replace this by an oblique coordinate system: $`L=\{ae+bw|a,b\}`$, where $`e=(1,0),w=(\frac{1}{2},\frac{\sqrt{3}}{2})`$ in the standard rectangular system and relative to this basis we can identify $`L`$ and $`^2`$ and denote $`ae+bw`$ by $`(a,b)`$. The basic inflation shown in Figure 1 gives rise to the map
$$T:x2Rx+(1,0),$$
where $`R`$ is a reflection in $`^2`$ through $`x`$-axis, i.e. in the new coordinates, $`R(1,0)=(1,0),R(0,1)=(1,1)`$.
The various types of points are designated by letter pairs $`i\alpha `$, where $`i\{1,\mathrm{},12\}`$ and $`\alpha \{a,\mathrm{},d\}`$ (of which only 36 actually occur). Let $`U_{i\alpha }`$ be the set of points of type $`i\alpha `$. On the basis of this we can make mappings of each point set to other point set.
Define
$$\begin{array}{cc}h_1:xTx+(0,0),\hfill & h_2:xTx+(1,0)\hfill \\ h_3:xTx+(0,1),\hfill & h_4:xTx+(1,1)\hfill \\ h_5:xTx+(1,0),\hfill & h_6:xTx+(0,1)\hfill \\ h_7:xTx+(1,1),\hfill & h_8:xTx+(2,1)\hfill \\ h_9:xTx+(1,2),\hfill & h_{10}:xTx+(1,1).\hfill \end{array}$$
Let $`f_{i\alpha j\beta }`$ be the function which maps $`j\beta `$-point set into $`i\alpha `$-point set.
\[Type 1\]
$$\begin{array}{cc}f_{9a\mathrm{\hspace{0.17em}1}a}=h_4:xTx+(1,1),\hfill & f_{1a\mathrm{\hspace{0.17em}1}b}=h_2:xTx+(1,0)\hfill \\ f_{9b\mathrm{\hspace{0.17em}1}a}=h_9:xTx+(1,2),\hfill & f_{1b\mathrm{\hspace{0.17em}1}b}=h_1:xTx+(0,0)\hfill \\ f_{9c\mathrm{\hspace{0.17em}1}a}=h_3:xTx+(0,1),\hfill & f_{4a\mathrm{\hspace{0.17em}1}b}=h_5:xTx+(1,0)\hfill \\ f_{9d\mathrm{\hspace{0.17em}1}a}=h_2:xTx+(1,0),\hfill & f_{4b\mathrm{\hspace{0.17em}1}b}=h_{10}:xTx+(1,1)\hfill \\ f_{4a\mathrm{\hspace{0.17em}1}a}=h_1:xTx+(0,0),\hfill & f_{4c\mathrm{\hspace{0.17em}1}b}=h_6:xTx+(0,1)\hfill \\ f_{4b\mathrm{\hspace{0.17em}1}a}=h_6:xTx+(0,1),\hfill & f_{4d\mathrm{\hspace{0.17em}1}b}=h_7:xTx+(1,1)\hfill \\ f_{4c\mathrm{\hspace{0.17em}1}a}=h_7:xTx+(1,1),\hfill & \\ f_{4d\mathrm{\hspace{0.17em}1}a}=h_8:xTx+(2,1).\hfill & \end{array}$$
\[Type 2\]
$$\begin{array}{cc}f_{12a\mathrm{\hspace{0.17em}4}a}=h_1:xTx+(0,0),\hfill & f_{1a\mathrm{\hspace{0.17em}4}b}=h_2:xTx+(1,0)\hfill \\ f_{12b\mathrm{\hspace{0.17em}4}a}=h_4:xTx+(1,1),\hfill & f_{1b\mathrm{\hspace{0.17em}4}b}=h_1:xTx+(0,0)\hfill \\ & \\ f_{4a\mathrm{\hspace{0.17em}4}c}=h_1:xTx+(0,0),\hfill & f_{1a\mathrm{\hspace{0.17em}4}d}=h_1:xTx+(0,0)\hfill \\ f_{4b\mathrm{\hspace{0.17em}4}c}=h_6:xTx+(0,1),\hfill & f_{1b\mathrm{\hspace{0.17em}4}d}=h_5:xTx+(1,0)\hfill \\ f_{4c\mathrm{\hspace{0.17em}4}c}=h_7:xTx+(1,1),\hfill & \\ f_{4d\mathrm{\hspace{0.17em}4}c}=h_8:xTx+(2,1).\hfill & \end{array}$$
All points in a sphinx having 2-points in it are mapped as in \[Type 1\] changing the translation part according to the orientation of the sphinx relative to sphinx 1. Likewise, all points in a sphinx having 4-points in it are mapped as in \[Type 2\] relative to sphinx 4.
Now we can list the $`36\times 36`$ matrix($`\mathrm{\Phi }`$) of affine mappings that make up our substitution system (Figure 4).
We can check that $`S(\mathrm{\Phi })`$ has PF-eigenvalue $`4`$ and is a primitive matrix and the union of point sets is L. We used Mathematica to check that property (iv) in Theorem 3 is satisfied in $`\mathrm{\Phi }^8`$ (it may actually be satisfied at some lower power). Certainly in $`\mathrm{\Phi }^8`$ there are a large number of modular coincidences. Theorem 1 and 3 say that all 36 point sets are regular model sets in CPS (24).
## 6 The total index and model sets
In this section we derive another criterion for determining when a partition of a lattice is a partition into $`Q`$-adic model sets, the difference this time being that there is no substitution system involved.
We assume that we are given a lattice $`L`$ in $`^n`$ and an inflation $`Q`$ on $`L`$ as in (9). The notation remains the same as before. The main ingredient is a non-negative sub-additive function called the total index which is defined on the subsets of $`L`$ and its $`Q`$-adic completion $`\overline{L}`$.
For any subset $`V`$ of $`L`$ the coset part of $`V`$ is defined as
$$𝒞(V):=\{C|C\text{is a coset in}V\}.$$
(41)
The key point to remember in what follows is that two cosets in $`L(\overline{L})`$ are either disjoint or one of them is contained in the other. If $`C=a+Q^kL`$ is a coset then we write $`[L:C]`$ for the index of the subgroup $`Q^kL`$ in $`L`$.
###### Lemma 6
The coset part of $`V`$ can be written as a disjoint union of cosets in $`V`$.
proof: If $`V`$ contains no cosets, then the result is clear. Suppose $`V`$ contains cosets. Let $`C_1=a_1+Q^{k_1}L`$ be a coset in $`V`$ with $`k_1`$ minimal. Consider $`V\backslash C_1`$. No coset can be partly in $`C_1`$ and partly in $`V\backslash C_1`$. Thus, if $`V\backslash C_1`$ contains no cosets, then $`𝒞(V)=C_1`$. Otherwise let $`C_2`$ be a coset $`a_2+Q^{k_2}L`$ with $`k_2`$ minimal in $`V\backslash C_1`$. Then $`𝒞(V)C_1\stackrel{}{}C_2`$. We continue this process. Since there are only finitely many cosets for $`Q^kL`$ in $`L`$, either we obtain $`𝒞(V)=C_1\stackrel{}{}\mathrm{}\stackrel{}{}C_r`$ for some $`r`$ or $`𝒞(V)C_1\stackrel{}{}C_2\stackrel{}{}\mathrm{},`$ where $`k_1k_2\mathrm{}`$ is infinite and unbounded. In the latter case, $`𝒞(V)=_{i=1}^{\mathrm{}}C_i`$ is our required decompostion. If not, there is a coset $`C=a+Q^kL`$ in $`V`$ such that $`C_{i=1}^{\mathrm{}}C_i`$. Then there is $`C_i`$ with $`k_{i1}k<k_i.`$ This contradicts the choice of $`C_i.`$ $`\mathrm{}`$
For $`VL`$, we call a decomposition $`𝒞(V)=_iC_i`$ of $`𝒞(V)`$ into mutually disjoint cosets using the algorithm of Lemma 6, an efficient decomposition of $`V`$ into cosets. In this case we call $`c(V):=_i[L:C_i]^1`$ the total index of $`V`$. Since any coset is an efficient decomposition of itself, we have $`c(V)=_ic(C_i)`$. We will see shortly that the total index is finite.
It is useful to note that an efficient decomposition of $`𝒞(V)=_iC_i`$ of $`𝒞(V)`$ into cosets has the following special property: if $`D`$ is any coset of $`V`$ then necessarily $`DC_i`$ for some $`i`$.
###### Lemma 7
Any two efficient decompositions of $`𝒞(V)`$ are the same up to rearrangement of the order of the cosets. In particular the total index is well-defined.
proof: Let $`𝒞(V)=C^{}`$ be a second decomposition of $`𝒞(V)`$ determined by the same algorithm as in Lemma 6. Then with $`k_1`$ as in the Lemma, let $`D_1,\mathrm{},D_r`$ be all the cosets of $`V`$ of the form $`a+Q^{k_1}L`$. These are all disjoint and by the algorithm all of them must be chosen in the decomposition of $`𝒞(V)`$, and they all occur before all the others. Thus $`C_1,\mathrm{},C_r`$ and $`C_1^{},\mathrm{},C_r^{}`$ are $`D_1,\mathrm{},D_r`$ in some order. Removing these and continuing in the same way the result is clear. $`\mathrm{}`$
We have similar concepts in $`\overline{L}`$. For $`W\overline{L}`$ we have the coset part $`𝒞^{}(W)`$ of $`W`$ and $`𝒞^{}(W)`$ can be written as a disjoint union of cosets in $`W`$. Let $`𝒞^{}(W)=_iD_i`$ where $`D_i,i=1,2,\mathrm{},`$ are mutually disjoint cosets in $`W`$. We call $`c^{}(W):=_i[\overline{L}:D_i]^1`$ the total index of $`W`$. This time we do not need to be careful about the way in which the decomposition is obtained since the total index is nothing else than the measure $`\mu (𝒞^{}(W))`$ of $`𝒞^{}(W)`$.
Given an efficient decomposition $`𝒞(V)=_{i=1}C_i`$ into disjont cosets in $`L`$, we define $`\overline{𝒞}(V):=_{i=1}\overline{C_i}\overline{L}`$. This is actually an open set in $`\overline{L}`$. Since $`[L:C]=[\overline{L}:\overline{C}]`$ we see that $`c(V)=c^{}(\overline{𝒞}(V))`$. In particular it follows that the total index of any subset $`V`$ of $`L`$ is finite and bounded by $`\mu (\overline{𝒞}(V))`$.
###### Lemma 8
For $`X,YL`$ and $`XY`$, and any decomposition $`𝒞(X)=_iC_i`$ into disjoint cosets, $`_ic(C_i)c(Y)`$. In particular, $`c(X)c(Y)`$.
proof: Assume first that $`Y`$ is a single coset $`C`$. Then
$$\underset{i}{}c(C_i)=\underset{i}{}c^{}(\overline{C}_i)=\underset{i}{}\mu (\overline{C}_i)\mu (\overline{C})=c^{}(\overline{C})=c(C),$$
(42)
since the cosets remain distinct after closing them in $`\overline{L}`$.
In the general case, let $`𝒞(Y)=\stackrel{}{}_{j=1}C_j^{}`$ be an efficient decomposition of $`Y`$. Since $`XY`$, each $`C_iY`$. In view of the remark above about efficient decompositions, there is for each $`i`$ a unique $`j`$ for which $`C_iC_j^{}`$. Thus we can arrange the $`C_i`$’s so that
$$𝒞(X)=\underset{j=1}{\overset{\mathrm{}}{}}\underset{iA_j}{}C_i$$
(43)
where $`A_j:=\{i|C_iC_j^{}\}`$. Now $`_{iA_j}C_iC_j^{}`$, so by the first part of the proof, $`_{iA_j}c(C_i)c(C_j^{})`$. Finally
$$c(X)=\underset{j}{}\underset{iA_j}{}c(C_i)c(C_j^{})=c(Y).$$
(44)
$`\mathrm{}`$
###### Lemma 9
Let $`U_i,i=1,\mathrm{},m,`$ be disjoint point sets of the lattice $`L`$ in $`^n`$. Let $`\mathrm{\Lambda }_i=\overline{U_i}L`$ and $`𝒞(U_i)`$ be the coset part in $`U_i`$. Then $`_{i=1}^m(\mathrm{\Lambda }_i\backslash U_i)L\backslash _{i=1}^m𝒞(U_i)`$, with equality if $`L=_{i=1}^mU_i`$.
proof For $`x_{i=1}^m𝒞(U_i)`$ there is a coset $`C𝒞(U_i)`$ for which $`xCU_i`$. Let $`C=a+Q^kL,aL`$. Suppose $`x`$ is a limit point of $`U_j`$ in $`\overline{L}`$ for some $`ji`$. Then, since $`a+Q^k\overline{L}`$ is an open neighborhood of $`x`$, $`(a+Q^k\overline{L})U_j\mathrm{}`$ i.e. $`(a+Q^kL)U_j\mathrm{}`$. But then $`U_iU_j\mathrm{}`$, contrary to the assumption. This means $`x_{i=1}^m(\mathrm{\Lambda }_i\backslash U_i)`$, proving the first part.
Suppose that $`L=_{i=1}^mU_i`$ and $`xL`$ but $`x_{i=1}^m𝒞(U_i)`$. Then $`xU_i`$ for some $`U_i`$ but there is no coset in $`U_i`$ which contains $`x`$. For any $`k_+,B_k(x):=x+Q^k\overline{L}`$ is an open neighborhood of $`x`$ in $`\overline{L}`$ and $`LB_k(x)U_i`$, by assumption. Since $`L=_{i=1}^mU_i,(LB_k(x))U_j\mathrm{}`$ for some $`ji`$. So we can choose $`x_k^j(LB_k(x))U_j`$. Then we get a sequence $`\{x_k^j\}`$ convergent to $`x`$ as $`k\mathrm{}`$. Choosing a subsequence lying entirely in one $`\mathrm{\Lambda }_j`$ shows that $`x\mathrm{\Lambda }_j`$ for some $`ji`$. Since $`xU_i`$, and $`U_i,U_j`$ are disjoint, $`x\mathrm{\Lambda }_j\backslash U_j`$. $`\mathrm{}`$
###### Theorem 4
Let $`U_i,i=1,\mathrm{},m,`$ be disjoint nonempty point sets of the lattice $`L`$ in $`^n`$. Let $`𝒞(U_i)`$ be the coset part in $`U_i`$, $`c(U_i)`$ the total index of $`U_i`$, and $`W_i`$ the closure of $`U_i`$ in $`\overline{L}`$. Then $`_{i=1}^mc(U_i)=1`$ if and only if the sets $`U_i,i=1,\mathrm{},m,`$ are regular weak model sets in the CPS(24) and $`\overline{L}=_{i=1}^mW_i`$.
proof
$`()`$ Assume that $`_{i=1}^mc(U_i)=1`$. Let $`U_{m+1}:=L\backslash _{i=1}^mU_m`$. Using Lemma 8 and the fact that $`c(L)=1`$, we see that $`c(U_{m+1})=0`$ and $`_{i=1}^{m+1}c(U_i)=1`$. For this reason we can assume, in proving that the $`U_i`$ are weak model sets, that $`_{i=1}^mU_i=L`$ in the first place.
For $`jk`$ the cosets of $`𝒞(U_j)`$ (of which there may be none!) and those of $`𝒞(U_k)`$ are disjoint from one another, and the same applies to $`\overline{𝒞}(U_j)`$ and $`\overline{𝒞}(U_k)`$. Thus
$$\mu \left(\underset{i=1}{\overset{m}{}}\overline{𝒞}(U_i)\right)=\underset{i=1}{\overset{m}{}}\mu (\overline{𝒞}(U_i))=\underset{i=1}{\overset{m}{}}c(U_i)=1,$$
and
$$\mu \left(\overline{L}\backslash (\underset{i=1}{\overset{m}{}}\overline{𝒞}(U_i))\right)=0.$$
(45)
Now note that $`W_j_{i=1}^m\overline{𝒞}(U_i)=\mathrm{}`$ for any $`j`$. If not let $`aW_j\overline{𝒞}(U_k)`$ for some $`k`$. Since $`\overline{𝒞}(U_k)\stackrel{}{W_k}`$, we see that $`jk`$. But $`aW_j`$, so $`a`$ is a limit point of $`U_j`$ ,and $`\overline{𝒞}(U_k)`$ is an open neighborhood of $`a`$, so $`U_j𝒞(U_k)\mathrm{}`$. This violates the disjointness of the $`U_i`$’s. We conclude that $`W_j\overline{L}\backslash (_{i=1}^m\overline{𝒞}(U_i))`$ and hence that
$$\mu (W_j)=0,$$
(46)
for all $`j=1,\mathrm{},m.`$ Note also
$`\mathrm{\Lambda }_i\backslash U_i`$ $``$ $`{\displaystyle \underset{j=1}{\overset{m}{}}}(\mathrm{\Lambda }_j\backslash U_j)`$
$``$ $`L\backslash {\displaystyle \underset{j=1}{\overset{m}{}}}𝒞(U_j)\text{by Lemma 8}`$
$`=`$ $`L\backslash {\displaystyle \underset{j=1}{\overset{m}{}}}(\overline{𝒞}(U_j)L)=L\backslash {\displaystyle \underset{j=1}{\overset{m}{}}}\overline{𝒞}(U_j).`$
This shows that
$$\mu (\overline{\mathrm{\Lambda }_i\backslash U_i})\mu \left(\overline{L\backslash (\underset{i=1}{\overset{m}{}}\overline{𝒞}(U_i))}\right)\mu \left(\overline{L}\backslash \underset{i=1}{\overset{m}{}}\overline{𝒞}(U_i)\right)=0.$$
(47)
By Lemma 4 (i) and (ii), $`\stackrel{}{W_i}\stackrel{}{W_j}=\mathrm{}`$ for all $`ij`$ and $`(\mathrm{\Lambda }_i\backslash U_i)_{j=1}^mW_j`$ for all $`i=1,\mathrm{},m`$.
Using Lemma 5(i) we obtain that the sets $`U_i,i=1,\mathrm{},m,`$ are regular weak model sets in the CPS(24).
Remark : Whenever $`\stackrel{}{W_i}\mathrm{}`$, $`U_i`$ is actually a regular model set.
Since $`_{i=1}^m\overline{𝒞}(U_i)_{i=1}^mW_i,\mu (_{i=1}^mW_i)=1`$. Thus $`\overline{L}\backslash _{i=1}^mW_i`$ is open of measure $`0`$ and $`\overline{L}=_{i=1}^mW_i`$. This last argument does not require that $`_{i=1}^mU_i=L`$.
$`()`$ Assume that $`U_i=\mathrm{\Lambda }(V_i)=V_iL`$ where $`\overline{V_i}\backslash \stackrel{}{V_i}`$ has measure $`0`$ and $`\overline{L}=_{i=1}^mW_i`$. Thus $`U_iV_i`$ and $`W_i:=\overline{U_i}\overline{V_i}`$. Since $`L`$ is dense in $`\overline{L}`$ and for $`x\stackrel{}{V_i}`$ each ball around $`x`$ of radius $`ϵ>0`$ contains points of $`\stackrel{}{V_i}LU_i`$, it follows that $`\overline{U_i}\stackrel{}{V_i}`$. This proves that $`\stackrel{}{V_i}W_i\overline{V_i}`$. So $`\mu (\stackrel{}{V_i})=\mu (W_i)`$ and $`\mu (W_i\backslash \stackrel{}{V_i})=0`$. Now
$$\underset{i=1}{\overset{m}{}}W_i=(\underset{i=1}{\overset{m}{}}\stackrel{}{V_i})(\underset{i=1}{\overset{m}{}}(W_i\backslash \stackrel{}{V_i})).$$
So $`\mu (_{i=1}^mW_i)=\mu (_{i=1}^m\stackrel{}{V_i})`$. Also the disjointness of the $`U_i`$ gives $`\stackrel{}{V_i}\stackrel{}{V_j}=\mathrm{}`$ for $`ij`$ (since $`L`$ is dense in $`\overline{L}`$). Finally
$$1=\mu (\underset{i=1}{\overset{m}{}}W_i)=\mu (\underset{i=1}{\overset{m}{}}\stackrel{}{V_i})=\underset{i=1}{\overset{m}{}}\mu (\stackrel{}{V_i})=\underset{i=1}{\overset{m}{}}c^{}(\stackrel{}{V_i})\underset{i=1}{\overset{m}{}}c(\stackrel{}{V_i}L)c(U_i)1.$$
$`\mathrm{}`$
###### Corollary 2
Let $`(\stackrel{~}{U},\mathrm{\Phi })`$ be a primitive substitution system with inflation $`Q`$ on the lattice $`L`$ in $`^n`$. Suppose that PF-eigenvalue of the substitution matrix $`S(\mathrm{\Phi })`$ is equal to $`|detQ|`$ and $`L=_{i=1}^mU_i`$. Then $`_{i=1}^mc(U_i)=1`$, where $`c(U_i)`$ is the total index of $`U_i`$ if and only if the sets $`U_i,i=1,\mathrm{},m,`$ are model sets in CPS(24).
proof Use Theorem 1 to determine that for all $`i,\stackrel{}{W_i}\mathrm{}`$. Now use Theorem and Remark in the proof. $`\mathrm{}`$
## 7 Chair tiling
The two dimensional chair tiling is generated by the inflation rule shown in Figure 5. There are $`4`$ orientations of the chairs in any chair tiling. In it was shown that the chair tiling has an interpretation in terms of model sets based on the lattice $`^2`$ and its 2-adic completion as internal space.
In this section we generalize this result to the $`n`$-dimensional chair tiling using the results of the last section (see Figure 7 for an example of the $`3`$-dimensional chair). To make things clearer we begin with the case $`n=2`$.
### I. Chair tiling in $`^2`$
The starting point is to replace each tile by $`3`$ oriented squares. Figure 6 shows the inflation rule, for one chair, in terms of oriented squares. The resulting tiling is a square tiling of the plane in which each of the squares has one of $`4`$ orientations. The centre points of each square form a square lattice which we identify with $`^2`$ by assigning coordinates as shown.
Let $`U_i`$ be the set of centre points corresponding to squares of orientation $`(i)`$ as given in Figure 6. We start out from a basic generating set $`A_2:=\{(x_1,x_2)|x_i\{0,1\}\}`$ and determine the precise maps for the substitution rules of Figure 6.
Letting $`e_1:=(0,0),e_2:=(1,0),e_3:=(1,1),e_4:=(0,1),`$ these maps are defined as:
$$\begin{array}{cc}f_{j,i}:U_iU_j\text{by}(x,i)(2x+e_j,j)\hfill & \text{if}ji\pm 2\hfill \\ f_{i,i}^{(2)}:U_iU_i\text{by}(x,i)(2x+e_j,i)\hfill & \text{if}j=i\pm 2,\hfill \end{array}$$
where $`i,j\{1,2,3,4\},x^2,i\pm 2:=\{\begin{array}{cc}i+2\hfill & \text{if}i2\hfill \\ i2\hfill & \text{if}i>2.\hfill \end{array}`$
These are the maps of an affine substitution system $`\mathrm{\Phi }`$. In fact, if we define
$$h_1:x2x+e_1,h_2:x2x+e_2,h_3:x2x+e_3,h_4:x2x+e_4,$$
then
$$\begin{array}{cccc}f_{1,1}=h_1,\hfill & f_{1,2}=h_1,\hfill & f_{1,1}^{(2)}=h_3,\hfill & f_{1,4}=h_1\hfill \\ f_{2,1}=h_2,\hfill & f_{2,2}=h_2,\hfill & f_{2,3}=h_2,\hfill & f_{2,2}^{(2)}=h_4\hfill \\ f_{3,3}^{(2)}=h_1,\hfill & f_{3,2}=h_3,\hfill & f_{3,3}=h_3,\hfill & f_{3,4}=h_3\hfill \\ f_{4,1}=h_4,\hfill & f_{4,4}^{(2)}=h_2,\hfill & f_{4,3}=h_4,\hfill & f_{4,4}=h_4,\hfill \end{array}$$
and
$$\mathrm{\Phi }=\left(\begin{array}{cccc}\{h_1,h_3\}\hfill & \{h_1\}\hfill & \{\}\hfill & \{h_1\}\hfill \\ \{h_2\}\hfill & \{h_2,h_4\}\hfill & \{h_2\}\hfill & \{\}\hfill \\ \{\}\hfill & \{h_3\}\hfill & \{h_3,h_1\}\hfill & \{h_3\}\hfill \\ \{h_4\}\hfill & \{\}\hfill & \{h_4\}\hfill & \{h_4,h_2\}\hfill \end{array}\right)_.$$
Inflating $`A_2`$ by the substitutions above we generate the $`4`$ point sets $`U_i,i=1,2,3,4`$. The precise description of $`U_i`$ is the following :
$`U_1`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((0,0)+2^k(2,0)+t(1,1)+2^k4^2)`$
$`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((0,0)+2^k(0,2)+t(1,1)+2^k4^2){\displaystyle \underset{t=\mathrm{}}{\overset{\mathrm{}}{}}}\{t(1,1)\}`$
$`U_2`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((1,0)+2^k(2,0)+t(1,1)+2^k4^2)`$
$`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((1,0)+2^k(0,2)+t(1,1)+2^k4^2){\displaystyle \underset{t=0}{\overset{\mathrm{}}{}}}\{(0,1)+t(1,1)\}`$
$`U_3`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((1,1)+2^k(2,0)+t(1,1)+2^k4^2)`$
$`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((1,1)+2^k(0,2)+t(1,1)+2^k4^2)`$
$`U_4`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((0,1)+2^k(2,0)+t(1,1)+2^k4^2)`$
$`{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((0,1)+2^k(0,2)+t(1,1)+2^k4^2){\displaystyle \underset{t=0}{\overset{\mathrm{}}{}}}\{(1,0)+t(1,1)\}.`$
Each of these decompositions is basically into cosets, with the exception of three trailing sets in types $`1,2,4`$ which we will designate by $`V_1,V_2,V_4`$ respectively.
We can prove the correctness of this as follows:
Let $`U_1^{},U_2^{},U_3^{},U_4^{}`$ be the sets on the right hand sides respectively. Note that
1. The generating set $`A_2`$ is contained in $`U_i^{}`$ adequately, i.e.
$`(0,0)U_1^{},(0,1)U_2^{},(1,1)U_1^{},(1,0)U_4^{}`$.
2. Claim that $`U_i^{}_{j=1}^4\mathrm{\Phi }_{ij}U_j^{},i=1,2,3,4`$.
Check that for any $`i`$,
$`h_i(U_i^{})`$ $``$ $`{\displaystyle \underset{\stackrel{j=1}{ji,i\pm 2}}{\overset{4}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((e_i)+2^{k+1}(2(e_ie_j))+2t(e_{i\pm 2}e_i)`$
$`+2^{k+1}4^2)V_i`$
$``$ $`U_i^{}`$
$`h_{i\pm 2}(U_i^{})`$ $``$ $`{\displaystyle \underset{\stackrel{j=1}{ji,i\pm 2}}{\overset{4}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((e_i)+2^{k+1}(2(e_ie_j))+(2t+1)(e_{i\pm 2}e_i)`$
$`+2^{k+1}4^2)V_i`$
$``$ $`U_i^{}`$
$`h_i(U_l^{})`$ $``$ $`(2e_l+e_i+4^2)(2e_{l\pm 2}+e_i+4^2)`$
$``$ $`U_i^{},\text{where}li,i\pm 2,l\{1,2,3,4\}`$
3. $`U_i^{},i=1,2,3,4,`$ are all disjoint.
Indeed within each $`U_i`$, all the cosets and the non-coset part are clearly disjoint. And two cosets or non-coset sets chosen from $`U_i^{}`$ and $`U_j^{}`$, where $`ji,i\pm 2`$, cannot intersect, since they are different by mod 2. Futhermore two of cosets or non-coset sets chosen from $`U_i^{}`$ and $`U_{i\pm 2}^{}`$ cannot intersect either, since for $`a+2^k4^2U_i^{},b+2^l4^2U_{i\pm 2}^{}`$ with $`kl`$, $`ab0`$ mod $`2^k4^2`$.
Now since $`U_1^{},U_2^{},U_3^{},U_4^{}`$ are generated from $`A_2`$ by $`\mathrm{\Phi }`$, $`U_iU_i^{}`$ for all $`i=1,2,3,4`$. Also from $`U_i=^2`$, we get $`U_i^{}=^2`$ . Since all $`U_i^{},i=1,2,3,4,`$ are disjoint, $`U_i=U_i^{}`$ for all $`i=1,2,3,4`$.
Finally, for any $`i=1,2,3,4`$,
$$c(U_i)2\underset{k=0}{\overset{\mathrm{}}{}}\underset{t=0}{\overset{2^k1}{}}\frac{1}{(2^k4)^2}=2\underset{k=0}{\overset{\mathrm{}}{}}\frac{2^k}{16(2^k)^2}=\frac{1}{4}.$$
Thus $`_{i=1}^4c(U_i)=1`$. Theorem 4 shows that $`U_i,i=1,2,3,4,`$ are regular model sets.
### II. Chair tiling in $`^n`$
In this section we are going to generalize the foregoing to the $`n`$-dimensional chair tilings for all $`n2`$. The $`n`$-chair is an $`n`$-cube with a corner taken out of it. The inflation rule, which we spell out algebraically below, is geometrically the obvious generalization of the $`2`$-dimensional case.
We transform the geometry by replacing each chair by a $`2^n1`$ oriented cubes, as before, and coordinatize the lattice formed by the centres of the cubes, starting from the basic generating set $`A_n:=\{(x_1,\mathrm{},x_n)|x_i\{0,1\}\}`$. There are $`2^n`$ orientations of cubes and hence $`2^n`$ types of points (but only $`2^n1`$ of these types appear in the starting set $`A_n`$).
For each $`k0`$ let $`\beta (k)`$ be the binary expansion $`ϵ_o+ϵ_12+ϵ_22^2+\mathrm{}`$ of $`k`$$`ϵ_l\{0,1\}`$. We define the basic orientation vectors $`e_1,\mathrm{}e_{2^n}`$ by
$$e_i:=\{\begin{array}{cc}(ϵ_0,\mathrm{},ϵ_{n1})\text{ the binary digits of}\beta (i1)\hfill & \text{if}i2^{n1},\hfill \\ (1,\mathrm{}1)e_{i2^{n1}}\hfill & \text{if}i>2^{n1}.\hfill \end{array}$$
We determine the sets $`U_i`$, $`i=1,\mathrm{},2^n`$, of all $`i`$type points in $`^n`$ from the points of the basic generating set $`A_n`$, using the inflation rules below.
The types of the points of $`A_n`$ are as follows: for $`x=(x_1,\mathrm{},x_n)A_n`$,
* when $`x_n=1`$ ;
+ $`xU_i`$, for which $`\beta (i1)=(1,\mathrm{},1)+x`$,
* when $`x_n=0`$ ;
+ if $`x=(0,\mathrm{},0)`$, $`xU_1`$
+ otherwise, $`xU_{i+2^{n1}}`$, for which $`\beta (i1)=(1,\mathrm{},1)((1,\mathrm{},1)+x)`$.
The idea of considering our vectors in the form $`(1,\mathrm{},1)+x`$ is to make it easy to compare them with the basic orientation vectors.
This conforms with what happens when $`n=2`$: there are $`2^n1`$ types in the basic starting set that are in $`2^{n1}1`$ complementary pairs and $`1`$ pair of vectors $`(0,\mathrm{},0)`$ and $`(1,\mathrm{},1)`$ of the same type, namely of type $`1`$.
Define
$$\begin{array}{cc}f_{j,i}:U_iU_j\text{by}(x,i)(2x+e_j,j)\hfill & \text{if}ji\pm 2^{n1}\hfill \\ f_{i,i}^{(2)}:U_iU_i\text{by}(x,i)(2x+e_j,i)\hfill & \text{if}j=i\pm 2^{n1},\hfill \end{array}$$
where $`i,j\{1,\mathrm{},2^n\},x^n,i\pm 2^{n1}:=\{\begin{array}{cc}i+2^{n1}\hfill & \text{if}i2^{n1}\hfill \\ i2^{n1}\hfill & \text{if}i>2^{n1}.\hfill \end{array}`$
Let $`\mathrm{\Phi }`$ be the matrix function system. Define $`h_i:x2x+e_i,i\{1,\mathrm{},2^n\}`$.
$$\mathrm{\Phi }=\left(\begin{array}{c}\{h_1,h_{1+2^{(n1)}}\}\{h_1\}\mathrm{}\stackrel{\stackrel{1+2^{n1}}{}}{\{\}}\mathrm{}\{h_1\}\\ \mathrm{}\\ \{h_{2^n}\}\{h_{2^n}\}\mathrm{}\stackrel{\stackrel{2^n2^{n1}}{}}{\{\}}\mathrm{}\{h_{2^n2^{n1}},h_{2^n}\}\end{array}\right)_.$$
Inflating $`A_n`$ by the maps, we get the precise description of $`U_i`$ :
$$U_i=\underset{\stackrel{j=1}{ji,i\pm 2^{n1}}}{\overset{2^n}{}}\underset{k=0}{\overset{\mathrm{}}{}}\underset{t=0}{\overset{2^k1}{}}\left((e_i)+2^k(2(e_ie_j))+t(e_{i\pm 2^{n1}}e_i)+2^k4^n\right)V_i,$$
where
$$V_i=\{\begin{array}{cc}_{t=\mathrm{}}^{\mathrm{}}\{t(e_{1\pm 2^{n1}}e_1)\}\hfill & \text{if}i=1\hfill \\ _{t=0}^{\mathrm{}}\{t(e_ie_{i\pm 2^{n1}})+(e_{i\pm 2^{n1}})\}\hfill & \text{if}i1,1\pm 2^{n1}\hfill \\ \mathrm{}\hfill & \text{if}i=1+2^{n1}.\hfill \end{array}$$
(48)
The equalities can be proved in the same way as in the 2-dimensional case.
Let $`U_{}^{}{}_{i}{}^{}`$ be the set of the right hand side in (48). Note that
1. The generating set $`A_n`$ is contained in $`U_i^{}`$ adequately, i.e.
* $`e_1U_1^{}`$ if $`i=1`$
* $`e_{i\pm 2^{n1}}U_i^{}`$ if $`i1,1\pm 2^{n1}`$
* $`e_{1\pm 2^{n1}}U_1^{}`$ if $`i=1\pm 2^{n1}.`$
2. Claim that $`U_i^{}_{j=1}^{2^n}\mathrm{\Phi }_{ij}U_j^{},i=1,\mathrm{},2^n`$.
Indeed for $`i\{1,\mathrm{},2^n\}`$
$`h_i(U_i^{})`$ $``$ $`{\displaystyle \underset{\stackrel{j=1}{ji,i\pm 2^{n1}}}{\overset{2^n}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((e_i)+2^{k+1}(2(e_ie_j))+2t(e_{i\pm 2^{n1}}e_i)`$
$`+2^{k+1}4^n)V_i`$
$``$ $`U_i^{}`$
$`h_{i\pm 2^{n1}}(U_i^{})`$ $``$ $`{\displaystyle \underset{\stackrel{j=1}{ji,i\pm 2^{n1}}}{\overset{2^n}{}}}{\displaystyle \underset{k=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{t=0}{\overset{2^k1}{}}}((e_i)+2^{k+1}(2(e_ie_j))+(2t+1)(e_{i\pm 2^{n1}}e_i)`$
$`+2^{k+1}4^n)V_i`$
$``$ $`U_i^{}`$
$`h_i(U_l^{})`$ $``$ $`(2e_l+e_i+4^n)(2e_{l\pm 2^{n1}}+e_i+4^n)`$
$``$ $`U_i^{},\text{where}li,i\pm 2^{n1},l\{1,\mathrm{},2^n\}.`$
3. $`U_i^{},i=1,\mathrm{},2^n,`$ are disjoint.
Indeed all cosets and a non-coset set in each $`U_i^{}`$ are all disjoint. And two cosets or non-coset sets chosen from $`U_i^{}`$ and $`U_j^{}`$, where $`ji,i\pm 2^{n1}`$, cannot intersect, since they are different by mod 2. Futhermore two of cosets or non-coset sets chosen from $`U_i^{}`$ and $`U_{i\pm 2^{n1}}^{}`$ cannot intersect either, since for $`a+2^k4^nU_i^{},b+2^l4^nU_{i\pm 2^{n1}}^{}`$ with $`kl`$, $`ab0`$ mod $`2^k4^n`$.
Now since $`U_i^{},i=1,\mathrm{},2^n,`$ are generated from $`A_n`$ by $`\mathrm{\Phi }`$, $`U_iU_i^{}`$ for all $`i=1,\mathrm{},2^n`$. Also from $`_{i=1}^{2^n}U_i=^n`$, $`_{i=1}^{2^n}U_i^{}=^n`$. Since all $`U_i^{},i=1,\mathrm{},2^n,`$ are disjoint, $`U_i=U_i^{}`$ for all $`i=1,\mathrm{},2^n`$.
For any $`i=1,\mathrm{},2^n`$,
$$c(U_i)(2^n2)\underset{k=0}{\overset{\mathrm{}}{}}\underset{t=0}{\overset{2^k1}{}}\frac{1}{(2^k4)^n}=(2^n2)\underset{k=0}{\overset{\mathrm{}}{}}\frac{2^k}{2^{2n}(2^k)^n}=\frac{1}{2^n}.$$
Thus $`_{i=1}^{2^n}c(U_i)=1`$. Theorem 4 shows that $`U_i,i=1,\mathrm{},2^n,`$ are regular model sets.
To get a model set interpretation of the chair tiling itself we proceed as follows. We observe that every arrow points to the inner corner of exactly one chair. Let us label each chair by its inner corner point which is at the tip of exactly $`2^n1`$ arrows. These corner points give us $`2^n`$ sets $`X_1,\mathrm{},X_{2^n}`$ according to the type, and all lie in the shift $`L^{}=(\frac{1}{2},\mathrm{},\frac{1}{2})+^n`$ of our lattice $`^n`$. Let $`f_i,i=1,\mathrm{},2^n`$, be $`(\frac{1}{2},\mathrm{},\frac{1}{2})e_i`$ respectively. Then $`U_i+f_i`$ is the set of tips of all arrows of type $`i`$ and $`U_i+f_i=L^{}(V_i+f_i),`$ for some $`V_i_2^n`$ for which $`\overline{V_i}`$ compact, $`\stackrel{}{V_i}\mathrm{}`$ and $`\mu (V_i)=0`$. Now
$$X_i=L^{}(\underset{ji\pm 2^{n1}}{}(V_j+f_j))$$
which is the required regular model set description of $`X_i`$, since
$$(\underset{ji\pm 2^{n1}}{}(V_j+f_j))\underset{ji\pm 2^{n1}}{}(V_j+f_j)$$
and $`\mu ((V_j+f_j))=0`$ for all $`j=1,\mathrm{},2^n`$.
From this result we can show that if we mark each chair with a single point in a consistent way, then the set of points obtained from all the chairs of any one type also forms a regular model set, and hence a pure point diffractive set.
## Acknowledgment
It is a pleasure to thank Martin Schlottmann for his encouragement and in particular for his help in Theorem 1. We are also indebted to Michael Baake for his interest and insights into this work.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.